test(cli): a neutral fixture rig name in the never-started set-runtime cell

A seat from rig import --materialize-only that has never started has no conversation, occupant or native history for
a switch to carry, and its first start already builds the seat from its saved node row. So set-runtime now changes
what it starts as and switches nothing: the node's runtime, with the source runtime's settings cleared as a switch
clears them (or set from the given flags), the account home, the native permission choice, the member in its launch
recipe that a restore's first start reads, and a snapshot. There's no successor or switch id, and the CLI says so.

A seat with history keeps today's refusal. A never-started seat that wasn't materialized takes its runtime from its
rig spec at first start, so its refusal now says to declare the other runtime there and create the seat again.

CHANGELOG.md

@@ -8,6 +8,957 @@ deprecations, and behavioral changes. Breaking changes are called out explicitly
 
 ---
 
+## [0.7.8]
+
+**Not yet published.**
+
+OpenRig 0.7.8 lets a person answer a runner seat's permission requests, with a
+conformance check and an example runner to build against. It adds a maintained
+way to run a Claude worker in an NVIDIA OpenShell sandbox, and includes 0.7.7's
+fixes plus repairs across seats, tmux, Slack and the TUI.
+
+### Before you upgrade
+
+- **Restart the daemon after installing:**
+  ```bash
+  npm install -g @openrig/cli@0.7.8
+  rig daemon stop && rig daemon start
+  ```
+  Four migrations (142 to 145) store a Claude seat's MCP opt-out, seat activity
+  grants, OpenShell environment bindings and the person chosen to answer a
+  seat's permission requests.
+  ([#2177](https://github.com/mvschwarz/openrig/pull/2177), [#2295](https://github.com/mvschwarz/openrig/pull/2295), [#2296](https://github.com/mvschwarz/openrig/pull/2296), [#2301](https://github.com/mvschwarz/openrig/pull/2301))
+- **The bundled openrig-core plugin advances from 0.1.23 to 0.1.24.** One line
+  of its seat-continuity skill changed. Managed refresh preserves replaced files
+  under `$OPENRIG_HOME/backups/managed-refresh/<id>/`; reapply local edits you
+  still need from that copy. ([#2287](https://github.com/mvschwarz/openrig/pull/2287))
+
+### Permissions for runner seats
+
+- **A runner seat's permission requests now reach a person, and the default is
+  permit.** The runner contract's first version refused every request. Section 7
+  of `docs/reference/harness-runner-contract.md` now covers who is asked:
+  - **With a default person,** "They get an ordinary ask in the Feed and on
+    Slack. No setup is needed."
+  - **To choose someone else,** run
+    `rig seat set-permissions-to <seat> --to <person> --reason "…"`. `--clear`
+    returns to the default.
+  - **When no one can be asked,** the request is "Permitted at once, with the
+    agent's own allow option."
+  - **Refusal is opt-in,** with `--deny`.
+
+  "A person's no is never a grant." An unanswered request "stays pending and
+  visible" until the agent's turn ends: "It's never timed out into a grant."
+  ([#2301](https://github.com/mvschwarz/openrig/pull/2301))
+- **A runner conformance check:**
+  `node docs/reference/runner-conformance/check-runner.mjs --runner '<runner command with {agent}>' --name <harness>`.
+  It covers the answered path, and needs Node and tmux but no daemon, model,
+  credentials or network.
+  ([#2303](https://github.com/mvschwarz/openrig/pull/2303), [#2306](https://github.com/mvschwarz/openrig/pull/2306))
+- **An example runner for DeepSeek Harness (`dsh`),** with its preset, is under
+  `docs/reference/community-harness-examples/dsh/`. OpenRig doesn't install it.
+  ([#2305](https://github.com/mvschwarz/openrig/pull/2305))
+- **A seat activity grant** lets a runner outside the host's hook setup report
+  activity for one seat and launch, without the install-wide hook token. ([#2295](https://github.com/mvschwarz/openrig/pull/2295))
+
+### Claude workers in OpenShell sandboxes
+
+- **`rig environment setup` installs a maintained reference** that binds a seat
+  to one NVIDIA OpenShell sandbox (`rig environment create` or `attach`, then
+  `rig up`).
+  - **Context and results:** the seat's guidance, skills, plugins and startup
+    files go into the sandbox, and verified JSON results come back through
+    queue work.
+  - **Stop and delete:** stopping the agent, stopping compute and deleting the
+    sandbox are separate actions.
+  - **Restore:** a recreated sandbox resumes the same conversation.
+
+  The steps are in `daemon/assets/openshell/README.md`.
+  ([#2296](https://github.com/mvschwarz/openrig/pull/2296), [#2300](https://github.com/mvschwarz/openrig/pull/2300))
+
+### Seats, delivery and the TUI
+
+- **Seats:**
+  - an isolated tmux server per OpenRig home
+    (`rig config set terminal.tmux_server isolated`);
+  - `claude_mcp: none` for a Claude member;
+  - resuming with a new token file;
+  - first starts on `rig launch --seats`;
+  - an accurate account record and kept effort after `rig provider switch`;
+  - skill copies, activity hooks and startup declarations kept across
+    `rig down`, resumes and exports;
+  - clearer stops for handover's `fork:<seat>`, a missing conversation,
+    `claude` not found, Codex restart refusals and failed unpins.
+  ([#1260](https://github.com/mvschwarz/openrig/pull/1260), [#2177](https://github.com/mvschwarz/openrig/pull/2177), [#2269](https://github.com/mvschwarz/openrig/pull/2269), [#2342](https://github.com/mvschwarz/openrig/pull/2342), [#2344](https://github.com/mvschwarz/openrig/pull/2344), [#2337](https://github.com/mvschwarz/openrig/pull/2337), [#2338](https://github.com/mvschwarz/openrig/pull/2338), [#2190](https://github.com/mvschwarz/openrig/pull/2190), [#2297](https://github.com/mvschwarz/openrig/pull/2297), [#2287](https://github.com/mvschwarz/openrig/pull/2287), [#2255](https://github.com/mvschwarz/openrig/pull/2255), [#2282](https://github.com/mvschwarz/openrig/pull/2282), [#2330](https://github.com/mvschwarz/openrig/pull/2330), [#2332](https://github.com/mvschwarz/openrig/pull/2332))
+- **tmux:**
+  - with `OPENRIG_TMUX_CONTROL=1`, no crash on tmux 3.3a to 3.7c when the
+    control client ends while another leaves;
+  - a submit key that a read-only client blocks is pasted instead.
+  ([#2317](https://github.com/mvschwarz/openrig/pull/2317), [#2325](https://github.com/mvschwarz/openrig/pull/2325), [#2345](https://github.com/mvschwarz/openrig/pull/2345))
+- **Wakes:**
+  - `OPENRIG_WAKE_SUSPEND` matches destinations with a colon;
+  - prompt escalations stay deliverable without a reachable person;
+  - reminders stop at a full quota;
+  - `human@kernel` raises no false finding;
+  - OpenRig's own Codex hooks are trusted when no choice is recorded;
+  - refusals name their fix.
+  ([#2353](https://github.com/mvschwarz/openrig/pull/2353), [#2343](https://github.com/mvschwarz/openrig/pull/2343), [#2281](https://github.com/mvschwarz/openrig/pull/2281), [#2286](https://github.com/mvschwarz/openrig/pull/2286), [#2294](https://github.com/mvschwarz/openrig/pull/2294), [#2290](https://github.com/mvschwarz/openrig/pull/2290), [#2284](https://github.com/mvschwarz/openrig/pull/2284), [#2280](https://github.com/mvschwarz/openrig/pull/2280), [#2273](https://github.com/mvschwarz/openrig/pull/2273), [#2202](https://github.com/mvschwarz/openrig/pull/2202))
+- **Slack and usage:**
+  - typed replies to parked decisions are recorded as answers;
+  - row bodies render Markdown;
+  - resolved decisions stop showing as live asks;
+  - OpenRouter charge retries aren't delayed.
+  ([#2293](https://github.com/mvschwarz/openrig/pull/2293), [#2299](https://github.com/mvschwarz/openrig/pull/2299), [#2316](https://github.com/mvschwarz/openrig/pull/2316), [#2349](https://github.com/mvschwarz/openrig/pull/2349), [#2314](https://github.com/mvschwarz/openrig/pull/2314))
+- **Hosts and setup:**
+  - `rig host edit` and `rig host remove`;
+  - a warning when another OS user's daemon holds the port;
+  - `rig daemon status` shows a pending start.
+  ([#2310](https://github.com/mvschwarz/openrig/pull/2310), [#2346](https://github.com/mvschwarz/openrig/pull/2346), [#2302](https://github.com/mvschwarz/openrig/pull/2302), [#2348](https://github.com/mvschwarz/openrig/pull/2348))
+- **The TUI and views:**
+  - agent pages show health and recent work;
+  - pod, spec and instance-table views read correctly;
+  - `rig terminal open mission:<id> --follow --provider herdr` follows a
+    mission.
+  ([#2263](https://github.com/mvschwarz/openrig/pull/2263), [#2264](https://github.com/mvschwarz/openrig/pull/2264), [#2265](https://github.com/mvschwarz/openrig/pull/2265), [#2266](https://github.com/mvschwarz/openrig/pull/2266), [#2327](https://github.com/mvschwarz/openrig/pull/2327), [#2260](https://github.com/mvschwarz/openrig/pull/2260), [#2289](https://github.com/mvschwarz/openrig/pull/2289))
+- **Also:**
+  - MCP text is cut between whole characters;
+  - an ACP cancel survives a failed note;
+  - long transcript lines read faster;
+  - there are Pi and kernel dashboard recovery guides.
+  ([#2268](https://github.com/mvschwarz/openrig/pull/2268), [#2262](https://github.com/mvschwarz/openrig/pull/2262), [#2274](https://github.com/mvschwarz/openrig/pull/2274), [#2186](https://github.com/mvschwarz/openrig/pull/2186), [#2275](https://github.com/mvschwarz/openrig/pull/2275))
+
+### Behaviour that changed
+
+- **A runner following section 7 gets answers or permits, not refusals.**
+  `--deny` keeps a seat refusing.
+- **On tmux older than 3.7c,** `OPENRIG_TMUX_CONTROL=1` uses spawned clients
+  instead of a control client.
+
+### Known limits
+
+- **No harness has run end to end under the contract yet.** The dsh example and
+  the conformance check were checked against stand-ins.
+- **The answer owner is a registered person.** "A colleague's answer is
+  followed on this host. If the ask is handed to another host, the request
+  waits, shown as needs input, until the agent's turn ends."
+- **`allow_always` from a permit:** with no one to ask, "If the agent offers
+  only `allow_always`, the harness then remembers a permission nobody chose."
+- **"Permits and answers aren't recorded beyond the runner's pane and the ask
+  itself."**
+- **Coming as follow-ups:** recording a permit made when no one can be asked,
+  and passing a person's typed words that aren't a plain yes or no to the agent.
+- **The dsh example keeps one hold per seat:** "if dsh asks twice in one turn,
+  settling one ask releases the hold while the other may still wait on a
+  person."
+- **The OpenShell reference requires:**
+  - NVIDIA OpenShell 0.1.6;
+  - a Linux image;
+  - an API-backed provider using `ANTHROPIC_API_KEY`. Subscription or OAuth
+    credentials aren't supported.
+
+  "Provider registration and image construction can incur costs." The worker's
+  activity reads unknown, from pane sampling, never idle.
+- **"Never grant the guest access to the daemon's main API port (normally
+  7433)."** The guest-only listener is what confines a guest's reach. Its grant
+  only authorizes requests for its own seat and launch.
+- **The 0.7.7 and 0.7.6 limits below still apply.**
+
+### Credits
+
+Thanks to @Aschenbath, @dajiaohuang, @Vaishnavi220506, @Nikhila-Dhevi-R,
+@Nikhi00718, @oodadoudou, @DeryFerd, @ch-davidriott, @vanducng, @Ethereal49,
+@AdityaxDeore and @Mydudeclaps for merged contributions. Thanks to @salbang,
+@ziadeh, @davey64, @bpamiri and @oneandonlydean for reports carried into this
+release, and to @bpamiri for dsh-seat, whose fake agent the conformance check's
+stand-in follows.
+
+## [0.7.7]
+
+**Not yet published.** 0.7.7 is a separate reliability patch for 0.7.6. Its three
+fixes are also in 0.7.8.
+
+OpenRig 0.7.7 is 0.7.6 with three reliability fixes for managed Claude seats: an
+idle prompt under Claude's "Update installed" line is recognized again, managed
+compaction no longer types into a message you haven't sent, and a Claude seat
+launched through a wrapper script can be rebuilt again.
+
+### Before you upgrade
+
+- **Restart the daemon after installing** so the fixes take effect:
+  ```bash
+  npm install -g @openrig/cli@0.7.7
+  rig daemon stop && rig daemon start
+  ```
+  There are no database migrations. The bundled openrig-core plugin stays at
+  0.1.23 with unchanged files.
+
+### Fixed
+
+- **A Claude seat under the "Update installed" line reads idle again.** After
+  Claude installs an update, it prints "✔ Update installed · /restart to apply"
+  (or "Restart to apply") below its mode bar. OpenRig didn't recognize that line,
+  so a seat at an empty prompt read unknown instead of idle, and managed
+  compaction waited. Both spellings are now recognized; lines that only look
+  similar still read unknown. ([#2311](https://github.com/mvschwarz/openrig/pull/2311))
+- **Managed compaction no longer types into a draft.** With the optional typing
+  guard off (the default), compaction's preparation prompt could be pasted onto
+  a message someone had typed but not sent. Now the preparation prompt,
+  `/compact` and the restore prompts check the input box first: typed or
+  unreadable input holds the step with a reason. A held restore prompt goes ahead
+  on its own once the box is empty; a manual `rig compact` says why it stopped,
+  to be retried once the box is empty. An empty box or Claude's own suggestion
+  goes ahead.
+  ([#2347](https://github.com/mvschwarz/openrig/pull/2347))
+- **A Claude seat launched through a wrapper script can be rebuilt again.** On
+  0.7.6, `rig seat handover --source rebuild` on a seat whose `claude` is a `#!`
+  script that starts the real Claude without `exec` stopped after 30 seconds with
+  "Waiting to confirm the new Claude process in this pane", though the new Claude
+  was running. The readiness check now also accepts the script's own direct child
+  when it names this launch's conversation; every other check is unchanged.
+  ([#2341](https://github.com/mvschwarz/openrig/issues/2341), [#2354](https://github.com/mvschwarz/openrig/pull/2354))
+- **A failed handover no longer moves the seat's resume target** to the new,
+  unprimed conversation while the seat keeps its previous occupant.
+  ([#2354](https://github.com/mvschwarz/openrig/pull/2354))
+
+### Known limits
+
+- **Other reasons managed compaction can wait are unchanged.**
+- **The draft check reads the input box just before pasting.** Typing that starts
+  in the same moment can still be appended to.
+- **A Claude seat whose current occupant generation was never registered,** such as
+  an adopted Claude carrying another generation, or a managed launch whose tenure
+  record failed to write, keeps its first conversation token, but later hooks don't
+  move it.
+- **The 0.7.6 known limits below still apply.**
+
+### Thanks
+
+- @oneandonlydean for the report and the evidence in #2341.
+
+## [0.7.6]
+
+OpenRig 0.7.6 lets one terminal workspace follow the current owners of a slice's
+open work through handoffs and completions. It also keeps more seat settings
+through expansion and restore, reports retained dead panes honestly, and repairs
+queue and Slack delivery paths.
+
+### Before you upgrade
+
+- **Restart the daemon after installing** so the new code and database updates
+  take effect:
+  ```bash
+  npm install -g @openrig/cli@0.7.6
+  rig daemon stop && rig daemon start
+  ```
+  Two migrations add the selected account-home path to nodes and the runner
+  declaration to sessions. Both have prefix 141; their distinct full filenames
+  identify them. ([#2143](https://github.com/mvschwarz/openrig/pull/2143), [#2239](https://github.com/mvschwarz/openrig/pull/2239))
+- **The bundled openrig-core plugin advances from 0.1.22 to 0.1.23.** Claude's
+  activity hook clears permission marks after a tool batch. Managed refresh
+  preserves replaced files under `$OPENRIG_HOME/backups/managed-refresh/<id>/`;
+  reapply local edits you still need from that copy. ([#2209](https://github.com/mvschwarz/openrig/pull/2209))
+
+### Follow the work in a slice
+
+- **Open `rig terminal open slice:<name-or-id> --follow --provider herdr`.**
+  Membership comes from pending, in-progress and blocked queue rows tagged with
+  that exact slice. Owners appear once with all their obligations listed;
+  unavailable owners and tile overflow stay named.
+  Handoffs retain the association unless their tags are replaced. New slice
+  views default to four automatic tiles; `--max-tiles` changes that. Known empty
+  slices stay following; incomplete reads retain the previous viewers. Pins,
+  drafts, unknown input and focus retain departed viewers, and closing a viewer
+  never stops its agent. Slice following targets the Linux Herdr 0.9.5 /
+  protocol 22 path. Mission following is not included.
+  ([#2259](https://github.com/mvschwarz/openrig/pull/2259))
+- **Named Herdr sessions use their configured endpoint throughout viewer and
+  follow operations.** The TUI's terminal action opens the agent, pod or rig
+  currently in view. ([#2205](https://github.com/mvschwarz/openrig/pull/2205), [#2232](https://github.com/mvschwarz/openrig/pull/2232))
+- **Client-visible HTTP change:** `/api/terminal/viewer/ensure` now returns 400
+  for `view_required` and `unknown_provider`, where it returned 200 with an
+  error body. Clients should continue reading the body code. Missing slice
+  identities return 404; ambiguous or unavailable identities return 409.
+  ([#2259](https://github.com/mvschwarz/openrig/pull/2259))
+
+- **When a slice's folder name repeats across missions, tag its work with the
+  slice's unique ID** (`rig queue create --slice <id>`) or add
+  `--mission <mission>`. A row tagged only with the shared name cannot be placed,
+  so the view reads incomplete until it is.
+  ([#2261](https://github.com/mvschwarz/openrig/pull/2261))
+
+- **A join refused before a pane split can join on a later refresh.** Hold
+  notices keep the seat and phase. The first notice keeps the original reason;
+  later notices can show a generic reason, while the original failure remains
+  in the journal. A created viewer whose attach has not yet run can continue
+  without replacing its pane.
+  ([#2271](https://github.com/mvschwarz/openrig/pull/2271))
+
+### Seats, accounts and harnesses
+
+- **Expansion keeps the member settings accepted by `rig up`,** including env,
+  effort and startup settings. Restores retain the model saved with
+  `rig seat set-model`. ([#2250](https://github.com/mvschwarz/openrig/pull/2250), [#2227](https://github.com/mvschwarz/openrig/pull/2227), [#2228](https://github.com/mvschwarz/openrig/pull/2228))
+- **Move a running Codex or Claude Code seat to an existing account home** with
+  `rig provider switch --seat <seat> --to-account <name> --account-home <dir>`;
+  preview with `--dry-run`. This starts a fresh conversation with a handover,
+  retaining the seat's address and work. `--clear-account-home` returns to the
+  configured defaults. ([#2143](https://github.com/mvschwarz/openrig/pull/2143))
+- **Claude subscription/API-key switching has a same-conversation path:**
+  `rig provider switch --seat <seat> --to-account api --api-key-file <abs>`;
+  return with `--to-account plan --token-file <abs>`. In-flight work is
+  interrupted. The API-key path remains unqualified; verify the resulting
+  account and conversation. A new Claude member can instead name a prepared
+  subscription token in its spec with `claude_token: {label: name, file: /abs/token}`.
+  ([#2191](https://github.com/mvschwarz/openrig/pull/2191), [#2246](https://github.com/mvschwarz/openrig/pull/2246))
+- **Relaunch retains the selected seat and its observed executable.** Codex
+  identity reads preserve OS argument boundaries, and in-place restart can
+  recognize a verified Node launcher named `MainThread`. This does not establish
+  that the reported #2200 restart refusal is fixed. ([#2196](https://github.com/mvschwarz/openrig/pull/2196), [#2212](https://github.com/mvschwarz/openrig/pull/2212), [#2247](https://github.com/mvschwarz/openrig/pull/2247))
+- **Managed seats inherit the daemon's SSH agent socket** for git over SSH;
+  check the daemon's agent with `ssh-add -l`. ([#2222](https://github.com/mvschwarz/openrig/pull/2222))
+- **Community presets can report turns and accept a declared model.**
+  `runner: true` with `launch_result: true` records runner activity per launch.
+  `passes_model: true` passes the requested model as `OPENRIG_MODEL`; it does not
+  observe the effective model. The Cursor and Hermes examples leave this off
+  pending native qualification. These are opt-in preset fields, not a completed
+  integration with every harness. ([#2239](https://github.com/mvschwarz/openrig/pull/2239), [#2243](https://github.com/mvschwarz/openrig/pull/2243))
+
+- **Draft integration author contracts** describe harness, runner and platform
+  duties, distinguishing implemented interfaces from proposals. Start at
+  [the author-contract entry point](docs/reference/author-contracts.md).
+  ([#2272](https://github.com/mvschwarz/openrig/pull/2272))
+
+### Delivery and Slack
+
+- **Original queue wakes retry after a selection prompt clears.** Retries identify
+  OpenRig as the sender, workflow entry keeps its caller, and prompt-blocked
+  alerts remain routable when the original sender is on another host.
+  ([#2235](https://github.com/mvschwarz/openrig/pull/2235), [#2229](https://github.com/mvschwarz/openrig/pull/2229), [#2237](https://github.com/mvschwarz/openrig/pull/2237))
+- **Work sent to a terminal-only seat gets a recovery finding for a person.**
+  Cross-host `rig queue create --verify` reads the destination's receipt;
+  unreadable receipts remain unconfirmed without replaying the create.
+  ([#2192](https://github.com/mvschwarz/openrig/pull/2192), [#2193](https://github.com/mvschwarz/openrig/pull/2193))
+- **Slack button answers resume decisions parked on the person asked.** Replies
+  Slack places outside a deleted thread are recorded at their actual location,
+  so the next reply reaches the right conversation. ([#2245](https://github.com/mvschwarz/openrig/pull/2245), [#2240](https://github.com/mvschwarz/openrig/pull/2240))
+- **Use `rig slack setup --reactions off` to stop acknowledgement emoji waking
+  agents.** The default remains `route`, and regenerating the app manifest
+  preserves the choice. Alert selection filters eligible notifications without
+  dropping older eligible asks behind an arbitrary row cap. ([#2241](https://github.com/mvschwarz/openrig/pull/2241), [#2183](https://github.com/mvschwarz/openrig/pull/2183))
+
+### Status, inventory and help
+
+- **A readiness deadline no longer ends a still-starting daemon.** The CLI
+  returns a nonzero pending result naming the child PID, launch endpoint and
+  log. It leaves the child running and retains the startup reservation, without
+  declaring readiness or allowing a duplicate start. Follow that pending
+  guidance before recovery. ([#2267](https://github.com/mvschwarz/openrig/pull/2267))
+- **A verified exited process in a retained pane reads `pane_exited`,** instead
+  of running and ready. Inventory avoids discarded graph loads, scopes startup
+  history to the requested rig's nodes, and reuses node identity for activity
+  reads. These changes make no universal latency claim. ([#2248](https://github.com/mvschwarz/openrig/pull/2248), [#2225](https://github.com/mvschwarz/openrig/pull/2225),
+  [#2226](https://github.com/mvschwarz/openrig/pull/2226), [#2236](https://github.com/mvschwarz/openrig/pull/2236))
+- **Completed mission labels reflect accepted proof**, without making proof
+  readiness complete a mission. Snapshot ages use UTC correctly; emoji stay
+  whole when TUI health, config and rig-table cells are shortened. ([#2249](https://github.com/mvschwarz/openrig/pull/2249),
+  [#2251](https://github.com/mvschwarz/openrig/pull/2251), [#2252](https://github.com/mvschwarz/openrig/pull/2252), [#2253](https://github.com/mvschwarz/openrig/pull/2253))
+- **Smaller command fixes:** chatroom tmux rig names stay literal; a trailing
+  slash in `OPENRIG_URL` still identifies the same daemon; a trailing slash in
+  the missions root no longer hides a slice's mission. `rig host doctor` names
+  the entry replies actually reach. ([#2218](https://github.com/mvschwarz/openrig/pull/2218), [#2220](https://github.com/mvschwarz/openrig/pull/2220), [#2221](https://github.com/mvschwarz/openrig/pull/2221), [#2214](https://github.com/mvschwarz/openrig/pull/2214))
+- **Help and contributor checks:** edge help explains deferred escalation
+  targets; section addressing ignores HTML comments; the guide covers shared
+  kernel-dashboard recovery. Install pins, help links and earlier release text
+  are corrected. Scenario fixtures honor send verification, execute tmux
+  directly, and reject unknown step verbs at schema load. ([#2204](https://github.com/mvschwarz/openrig/pull/2204), [#2211](https://github.com/mvschwarz/openrig/pull/2211),
+  [#2213](https://github.com/mvschwarz/openrig/pull/2213), [#2238](https://github.com/mvschwarz/openrig/pull/2238), [#2244](https://github.com/mvschwarz/openrig/pull/2244), [#2254](https://github.com/mvschwarz/openrig/pull/2254), [#2256](https://github.com/mvschwarz/openrig/pull/2256), [#2257](https://github.com/mvschwarz/openrig/pull/2257), [#2215](https://github.com/mvschwarz/openrig/pull/2215), [#2216](https://github.com/mvschwarz/openrig/pull/2216),
+  [#2217](https://github.com/mvschwarz/openrig/pull/2217))
+
+### Known limits
+
+- **The Codex hook-refresh refusal in [#2200](https://github.com/mvschwarz/openrig/issues/2200)
+  remains unresolved.** If `rig seat restart <seat> --keep-conversation` refuses,
+  nothing is stopped: leave the seat running and report the refusal with CLI
+  and daemon versions. A refused restart has not refreshed the hooks.
+- **A viewer whose attach already ran and then exited back to a shell is not
+  restarted automatically.** The whole slice view can hold until that tile is
+  resolved. No tested workaround is offered for that case.
+- **Following protects uncertain input by retaining tiles.** A larger workspace
+  or explicit `--max-tiles` can help with overflow. Protected Codex delivery can
+  still retain messages it cannot confirm before paste; retained is not
+  delivered. Disabling its typing guard removes protection and is not recovery.
+- **Internet-facing hosted members remain experimental, unsupported, off by
+  default and at your own risk.** Slice following does not require them.
+
+### Credits
+
+Thanks to @seanmcn, @DeryFerd, @Aschenbath, @Vaishnavi220506, @bradvin,
+@jay23606, @rudycelekli and @Nikhi00718 for merged contributions, and to
+@akushonkamen, @jcmurrayii, @KM199 and @pdurlej for contributed changes, reports
+and reproductions carried into this release.
+
+## [0.7.5]
+
+OpenRig 0.7.5 adds `rig terminal open qitem:<id> --follow`: keep one terminal
+workspace as a task's current owners join and leave. Joining viewers do not
+request focus; departures remove viewers, not the agents or conversations.
+Follow mode is scoped to Linux Herdr 0.9.5 / protocol 22. The exercised journey
+used shell stand-ins; real-agent follow qualification and active render cost
+are not established by that result. Changes since 0.7.4:
+
+### Before you upgrade
+
+- **Restart the daemon after installing** to run the new code and database
+  update. Migration 126 adds deferred `escalates_to` target addresses to the
+  edges table, preserving existing edges. Previously inert escalation edges
+  now affect recovery for seats without a delegation parent.
+  ```bash
+  npm install -g @openrig/cli@0.7.5
+  rig daemon stop && rig daemon start
+  ```
+- **The bundled openrig-core plugin goes from 0.1.21 to 0.1.22.** Its
+  `queue-handoff` skill teaches agents to park waits on CI or a long job with
+  a continuation and wake. As before, managed refresh preserves replaced
+  files under `$OPENRIG_HOME/backups/managed-refresh/<id>/`; reapply any local
+  edits you still need from that copy. No setting is added or removed.
+- **A Codex hook-refresh refusal remains unresolved.** The earlier remedy
+  `rig seat restart <seat> --keep-conversation --reason "refresh hooks"`
+  (preview with `--plan`) was reported to refuse one Codex seat upgraded from
+  0.6.x. Its cause is unknown: the inspected 0.7.4 check already accepts the
+  reported process tree. [#2200](https://github.com/mvschwarz/openrig/issues/2200)
+  is not fixed in 0.7.5; this is not an established refusal of all staged-script
+  launches. A refused restart does not complete the hook refresh. If it
+  refuses, nothing is stopped: leave the seat running and report the refusal
+  with CLI and daemon versions. If Codex asks for hook review, review the
+  hooks you intend to enable.
+
+### Follow one task in one workspace
+
+- **Follow the current owners of a queue item:**
+  `rig terminal open qitem:<id> --follow --provider herdr`. The view derives
+  membership from the task's open rows and explicitly linked successors.
+  Joining seats get verified viewer tiles in the same workspace. A complete
+  empty task drains removable tiles and keeps the workspace; after retained
+  tiles are gone, following stops. Run `--follow` again to resume it.
+  ([#2199](https://github.com/mvschwarz/openrig/pull/2199))
+- **Keep control of the view:** `--pin`, `--unpin`, `--suppress`,
+  `--unsuppress`, `--suppression-ms` and `--max-tiles` adjust presentation.
+  `--no-follow` leaves the layout in place. The default is eight tiles,
+  four seconds of departure grace and twelve more seconds for a focused
+  leaver. Pins can exceed the cap; input-retained leavers do not consume a
+  joining seat's slot. See [terminal sessions](docs/reference/terminal-sessions.md).
+  ([#2199](https://github.com/mvschwarz/openrig/pull/2199))
+- **Uncertain input keeps its tile.** Typed drafts and unreadable composers
+  retain departed viewers, including Pi, terminal and community harness
+  seats whose input cannot be confirmed. Unpinning does not override that.
+  There is one delayed input recheck, then queue/session events or an explicit
+  follow command can reconsider it. Failed or incomplete reads retain the
+  previous view with an explanation. Activity hooks do not trigger renders.
+  ([#2199](https://github.com/mvschwarz/openrig/pull/2199))
+- **Protected Codex delivery no longer strands a message it can't submit.** On a
+  Codex seat whose typing guard is draft-aware or openrig-urgent, a message
+  whose complete text can't be confirmed visible before paste is retained, not
+  sent, instead of being pasted where it can't be submitted. The default (guard
+  off) is unchanged. Use a larger pane or a shorter message; turning the typing
+  guard off turns draft protection off and is not a recovery. Pastes over 1,000
+  code points are retained. Wrapped lines use a conservative bound, so most
+  wrapped wakes in common four-tile views can be retained even when they fit
+  natively; approximate standard-envelope examples are one-paragraph bodies over
+  about 115 characters at 80×22, 192 at 80×24, or 526 at 120×30 (examples, not
+  limits). A later resize is not atomic, and layouts with Codex's image-enabled
+  ambient pet are unqualified.
+  ([#2242](https://github.com/mvschwarz/openrig/pull/2242))
+- **Codex 0.164's two-row composer footer is recognized.** A departed Codex
+  agent's empty composer no longer reads as unconfirmed input that keeps its
+  tile. Delivery's draft-aware and openrig-urgent typing guards use the same
+  reader, so an empty Codex 0.164 composer there can now receive typed
+  delivery. ([#2206](https://github.com/mvschwarz/openrig/pull/2206))
+- **Clients can ensure a missing viewer without relaying out existing panes**
+  through `POST /api/terminal/viewer/ensure`. It rechecks task membership and
+  existing viewer identity before creating or reusing a tile. Preview names
+  `not-provisioned`, `verified` or `unconfirmed`; membership gives no input
+  authority. This path has the same Linux Herdr 0.9.5 / protocol 22 scope.
+  ([#2133](https://github.com/mvschwarz/openrig/pull/2133))
+- **An unconfirmed pane no longer counts as opened.** CLI and TUI output name
+  it as degraded with `pane_unconfirmed`, distinguish it from a skipped seat,
+  and omit its restart-command report. An all-unconfirmed open exits nonzero.
+  Inspect before retrying; this is not proof that nothing opened.
+  ([#2173](https://github.com/mvschwarz/openrig/pull/2173))
+
+### Fixes by what you see
+
+- **Slow successful sends have time to answer.** Local and HTTP `rig send`
+  wait 60 seconds, plus an explicit idle wait, instead of five seconds locally
+  or thirty over HTTP. A timeout still means unconfirmed. The advice now says
+  that a busy recipient may hold input off-screen: missing pane text does not
+  prove non-delivery, so reconcile before another send or duplicate queue item.
+  No automatic resend is added. This does not diagnose the daemon's latency.
+  ([#2198](https://github.com/mvschwarz/openrig/pull/2198))
+- **A wake that was unroutable before its seat had a pane is retried once the
+  binding exists,** through the row's existing wake intent. The outbox does
+  not send that same intent again after it is settled as delivered.
+  ([#2170](https://github.com/mvschwarz/openrig/pull/2170))
+- **A parentless seat's recovery can follow its declared `escalates_to`
+  address,** including another rig on the same daemon. A delegation parent
+  still takes precedence. Unavailable or ambiguous targets use the explained
+  fallback; an already-open recovery keeps its recorded owner. This is one
+  escalation hop, not cross-host routing.
+  ([#1772](https://github.com/mvschwarz/openrig/pull/1772))
+- **A stuck finding needs evidence to close.** Delivery, pickup or explicit
+  custody can resolve it; a changed reporting threshold, an unreadable
+  registry or a failed wake becoming unconfirmed cannot silently clear it.
+  ([#2153](https://github.com/mvschwarz/openrig/pull/2153))
+- **Fresh Claude handovers wait for the current launch.** Retained scrollback
+  or an echoed shell command cannot make a new process look ready. The context
+  packet waits for the matching process and Claude's complete drawn input frame.
+  ([#2159](https://github.com/mvschwarz/openrig/pull/2159))
+- **Runtime send refusals return HTTP 409 rather than 500.** A stale recorded
+  Claude pane names `rig reconcile-session <seat>` as the repair; nothing is
+  typed on that refusal. ([#2157](https://github.com/mvschwarz/openrig/pull/2157))
+- **`rig down` keeps a rig down after a rolled-back start.** It records the
+  explicit stop without rewriting the superseded session's history. The
+  `rig start --last` help now explains which rigs return and how an explicit
+  stop keeps them out; the kernel is not selected by `--last`.
+  ([#2166](https://github.com/mvschwarz/openrig/pull/2166))
+- **Status and configuration describe the selected rig.** `rig ps` reports
+  `periodicSnapshotActive` only for rigs the scheduler actually snapshots
+  ([#2168](https://github.com/mvschwarz/openrig/pull/2168)); the web Overview
+  reads the rig's current exported configuration instead of a same-named
+  library template ([#2188](https://github.com/mvschwarz/openrig/pull/2188)).
+- **Host labels and next steps are clearer.** The TUI shows the configured
+  local host name while retaining its stable identity
+  ([#2162](https://github.com/mvschwarz/openrig/pull/2162)). `rig host add` and
+  `pair` explain how to opt an HTTP host into remote Feed attention; registration
+  leaves that setting unchanged ([#2179](https://github.com/mvschwarz/openrig/pull/2179)).
+- **Small CLI corrections:** `attach --self` reads tmux's rewritten separators
+  ([#2178](https://github.com/mvschwarz/openrig/pull/2178)); transcript and
+  capture line counts use their normalized values on local and remote paths
+  ([#2164](https://github.com/mvschwarz/openrig/pull/2164)); stdio MCP warns on
+  out-of-range daemon ports while continuing to serve, without substituting a
+  port ([#2160](https://github.com/mvschwarz/openrig/pull/2160)).
+- **Notification adapters release unread HTTP response bodies,** without
+  changing their send result or waiting for cleanup. No performance improvement
+  is claimed here. ([#2172](https://github.com/mvschwarz/openrig/pull/2172))
+- **Scope audits agree on slice folders numbered 100 and above.** The guide
+  also explains that moving a slice reopens its proof judgments; a clean
+  structural audit is not a proof judgment.
+  ([#2194](https://github.com/mvschwarz/openrig/pull/2194))
+
+### Presets, hosted members and guides
+
+- **The `pi-tui` example runs Pi's own terminal UI as a community harness.**
+  It gives each seat its own Pi agent folder and passes startup input to Pi.
+  One Pi 1.4.0 / OpenRig 0.7.4 run showed a parked wake returning to the same
+  session and an unattended handoff to a second seat. It did not prove Pi
+  chooses that coordination by itself. Resume is not declared. Use it
+  unattended: a wake can submit a person's half-typed draft, and unrecognized
+  dialogs can receive the send. Busy-turn wakes and other Pi versions remain
+  unqualified. ([#2182](https://github.com/mvschwarz/openrig/pull/2182),
+  [#2184](https://github.com/mvschwarz/openrig/pull/2184))
+- **Interactive Pi setup is explicit:** copy the coordination skills and use
+  `delivery_hint: send_text` for a startup file that must reach a preset's
+  first input. A plain terminal seat does not receive queue wakes. Setting its
+  typing guard to `off` can type notices into a shell; keep the default for a
+  seat that can return to one. These are documented limits, not new guards.
+  ([#2174](https://github.com/mvschwarz/openrig/pull/2174),
+  [#2180](https://github.com/mvschwarz/openrig/pull/2180),
+  [#2181](https://github.com/mvschwarz/openrig/pull/2181),
+  [#2185](https://github.com/mvschwarz/openrig/pull/2185))
+- **Hosted-member guidance names its real reach.** A member without a wake
+  trigger checks work when its owner next starts a session; a listed tool is
+  not a grant, and completed work closes with `no-follow-on`. Hosted members
+  exposed over the internet remain **experimental, unsupported, off by default
+  and at your own risk**. The guide explains narrow grants and token revocation;
+  this terminal-follow release requires no internet exposure.
+  ([#2187](https://github.com/mvschwarz/openrig/pull/2187),
+  [#2189](https://github.com/mvschwarz/openrig/pull/2189))
+- **Publishing a team bundle:** add and declare a README after `rig export`,
+  then run `rig bundle check` at the revision being submitted
+  ([#2176](https://github.com/mvschwarz/openrig/pull/2176)). Queue guidance
+  teaches parking long waits with a continuation and wake
+  ([#2167](https://github.com/mvschwarz/openrig/pull/2167)).
+- **Contributor maintenance:** the test guide corrects installed-scenario
+  counts and cases ([#2165](https://github.com/mvschwarz/openrig/pull/2165));
+  the compaction receipt fixture shares one clock, with no product change
+  ([#2169](https://github.com/mvschwarz/openrig/pull/2169)). Published release
+  notes, the capability map and roadmap records were carried forward
+  ([#2161](https://github.com/mvschwarz/openrig/pull/2161),
+  [#2175](https://github.com/mvschwarz/openrig/pull/2175),
+  [#2197](https://github.com/mvschwarz/openrig/pull/2197)).
+
+### Limits carried forward
+
+- **Follow mode is Linux Herdr 0.9.5 / protocol 22 only.** Other platforms,
+  providers and selector kinds retain one-shot open. Its exercised journey
+  used shell stand-ins, not native agents. Active render cost is unmeasured.
+  The first workspace on an empty provider may gain focus, and a viewer can
+  resize the agent's terminal; `ignore-size` is not resize isolation.
+- **Unreadable composers retain departed tiles.** This includes input the
+  inspector cannot read on Pi, terminal or community harness seats; unpinning
+  alone does not establish that closing is safe.
+- **Escalation across rigs is same-daemon only.** Upgrade followed by snapshot
+  restore of deferred escalation targets has not been qualified.
+- **A webhook 2xx is acceptance of the ring, not pickup.** A hosted member's
+  activity remains unknown between calls. Principals record attribution, not
+  authentication. See [hosted members](docs/reference/hosted-members.md).
+- **Reviewed-history resume remains a client operation through startup review,**
+  not a CLI command. Kernel restore evidence lasts one daemon process.
+- **OpenClaw interrupted local turns are not recovered; Hermes has no declared
+  resume or measured idle gateway exit.** `rig down` stops a community harness's pane; survival of independently spawned
+  children remains untested. Inspect actual processes before relying on their exit.
+- **The reported Codex restart refusal in #2200 remains unresolved,** as do
+  the other 0.7.4 limits unless a change above names them fixed: Codex permission qualification
+  only on 0.164.0, earlier removed hook approvals, SSH tiles after a Herdr
+  restart, unqualified native cmux, Slack-only shared channels and the existing
+  browser-sizing and old Slack-thread limits. Isolated account-home switching
+  is not included.
+
+### Credits
+
+Thanks to @icey-zzz (#1772), @Aschenbath (#2153, #2159, #2162, #2169, #2179),
+@DeryFerd (#2164, #2165), @Nikhi00718 (#2172) and @dajiaohuang (#2188).
+Thanks also to @rudycelekli for the contribution from #1416 completed in #2160;
+to @AdamWhitehurst (#1544, #2049), @kevinhsykes (#2151), @aanestesia (#605),
+@todie (#1419) and @tgrundtvig (#21), whose reports led to fixes here; and to
+the other people whose reports and reviews led to these fixes.
+
+## [0.7.4]
+
+OpenRig 0.7.4 lets an assistant you already use work as a member of your team.
+An assistant connected over MCP can be listed on its rig's team as a hosted
+member, learns from the connection how to take, finish and hand on work, can
+ask a person while keeping the item, and can be rung by a webhook when work
+lands for it; when nothing can wake it, its person is told instead.
+`rig mcp serve --http` can check each remote caller's key with an external
+grant service, offer OAuth discovery, record whose grant acted, and serve MCP
+clients on protocol revision 2026-07-28 beside 2025's. You can list the
+conversations a seat could resume, and a client can resume the one a person
+reviewed. Daemon start now brings back only a kernel that this start found
+down. It is built from the commit tagged `v0.7.4`. The changes since 0.7.3:
+
+### Before you upgrade
+
+- **Restart the daemon after installing.** One database update applies on its
+  first start (migration 140). It adds a column, empty on earlier rows, that
+  records which grant's caller acted, to queue transitions and their archive,
+  inbox and outbox entries, and stream items.
+  ```bash
+  npm install -g @openrig/cli@0.7.4
+  rig daemon stop && rig daemon start
+  ```
+- **The bundled openrig-core plugin goes from 0.1.20 to 0.1.21.** The
+  `queue-handoff` skill gains writing guidance, and the Claude hooks add
+  `StopFailure`, which reports a stop caused by a rate limit. As before,
+  startup first copies each file it replaces under
+  `$OPENRIG_HOME/backups/managed-refresh/<id>/`; if you edited one, reapply
+  your change from the copy. A Claude seat gets the new hook at its next
+  launch, resume, restore or handover. Codex sees its openrig-core plugin as
+  changed: if a Codex seat stops at "Hooks need review", review the hooks in
+  Codex.
+- **Going back to 0.7.3** leaves migration 140's column in place, unread.
+
+### What's new and fixed
+
+**Assistants as team members**
+
+- **A rig's team shows the members that live outside a terminal.** A roster
+  entry can mark a member as hosted (`"hosted": true`), such as an assistant
+  connected over MCP. `rig_team` lists a rig's hosted members with their
+  declared capabilities and caveat, no pane, and activity `unknown`: OpenRig
+  doesn't observe a hosted member between its calls, so it never reads idle.
+  A roster that can't be read gives a note, never "no members". `rig roster
+  show` and `find` say the member is hosted.
+  ([#2095](https://github.com/mvschwarz/openrig/pull/2095))
+- **The connection teaches the assistant how to work.** The MCP server's
+  instructions say that its inbox is the items addressed to its seat, that a
+  wake only points at the inbox, and to claim an item, do it, then close it
+  with the answer or hand it on. When an item needs a person's approval first,
+  it keeps the item and asks: it blocks the item on that person with the
+  question as the summary, and the answer returns the item in progress. A
+  hosted member can call `rig_whoami` to learn the seat it acts as.
+  ([#2095](https://github.com/mvschwarz/openrig/pull/2095),
+  [#2127](https://github.com/mvschwarz/openrig/pull/2127))
+- **When work waits for a member nothing can wake, the person is told.** The
+  stuck sweep used to address its finding to that same member. Now it goes to
+  the person, on Steering and the TUI's Feed, and to Slack under the default
+  subscription, saying to ask the member to check or to fix the address.
+  ([#2110](https://github.com/mvschwarz/openrig/pull/2110))
+- **A hosted member can be rung when work lands for it.** Its roster entry
+  declares a webhook trigger: `wake.webhook` with `url_ref`,
+  `authorization_ref` and optionally `secrets_env_file`. These are secret
+  names, read from the daemon's environment or a 0600 env file; values never
+  go in rosters or rows. OpenRig POSTs a pointer to the row (the event
+  `work-waiting`, the row id, destination, summary and sender, never the row's
+  body), waits up to 10 seconds and doesn't follow redirects. A 2xx is
+  recorded as the ring being accepted, not as the member waking. If the member
+  doesn't claim the work by the stuck sweep's unclaimed age, its person is
+  told. ([#2115](https://github.com/mvschwarz/openrig/pull/2115))
+- **The team sees when a hosted member last showed up.** `rig_team` shows when
+  each hosted member last called the team's MCP server, and its latest ring
+  and how that went; `rig roster show` and `find` show the last call. With no
+  evidence it says unknown, never idle, and a claim on a row is still the only
+  pickup. ([#2121](https://github.com/mvschwarz/openrig/pull/2121))
+
+**Remote MCP callers and grants**
+
+- **`rig mcp serve --http --grant-service <url>`** checks each request's key
+  with an external grant service instead of the local token file, on every
+  request, with nothing cached. `--grant-service-header-file <path>` sends one
+  `Name: value` header with each lookup.
+  ([#2096](https://github.com/mvschwarz/openrig/pull/2096),
+  [#2141](https://github.com/mvschwarz/openrig/pull/2141))
+- **OAuth discovery:** `--oauth-resource <url>` with `--oauth-issuer <url>`
+  publishes OAuth protected-resource metadata (RFC 9728), so an MCP client
+  that signs in with OAuth can find out how to authenticate, and a request
+  with no credential or a bad one gets a standard challenge. The key and its
+  grant still decide what the caller may do.
+  ([#2135](https://github.com/mvschwarz/openrig/pull/2135))
+- **Whose grant acted is recorded.** The rows a granted caller's actions write
+  record the caller's principal (`principal:v1`): its source, its grant name
+  or id and, for a grant service, the grant's version. It's attribution, not
+  authentication, and never a credential.
+  ([#2098](https://github.com/mvschwarz/openrig/pull/2098),
+  [#2141](https://github.com/mvschwarz/openrig/pull/2141))
+- **A caller's seat that was created but never launched** (for example with
+  `rig import --materialize-only`) is served, instead of every request being
+  refused with `seat_unavailable`.
+  ([#2125](https://github.com/mvschwarz/openrig/pull/2125))
+- **A caller granted through a grant service** hands its tasks to the team it
+  was granted, never to its own caller seat or another caller's
+  ([#2132](https://github.com/mvschwarz/openrig/pull/2132)), and can't attach
+  tags that claim OpenRig's own Slack provenance or a workflow's internal
+  sign-off, so a person asked about its task sees its whole text
+  ([#2126](https://github.com/mvschwarz/openrig/pull/2126)).
+
+**MCP protocol 2026-07-28**
+
+- `rig mcp serve`, over stdio or `--http`, serves clients on MCP protocol
+  revision 2026-07-28, which has no `initialize` handshake, sends its version
+  on every request and may open with `server/discover`. They reach the same
+  tools and the same work loop. Clients that open with 2025's `initialize`
+  keep working unchanged, on the same endpoint and process.
+  ([#2112](https://github.com/mvschwarz/openrig/pull/2112))
+
+**Resuming the conversation you chose**
+
+- **`rig seat history --rig <id> --node-id <id>`** lists the recorded Claude and Codex
+  conversations this exact node could resume, read-only and without native
+  tokens (`--json`; `--snapshot-id` with `--history-session-id` checks one
+  reference it returned). The rig ID is `rigId` in `rig ps --nodes --json`, and the
+  node ID is `identity.nodeId` in `rig whoami --session <seat> --json`. ([#2100](https://github.com/mvschwarz/openrig/pull/2100))
+- **A client can resume the history a person reviewed** for an exact node,
+  through the daemon's startup review. Applying it rechecks the choice and the
+  settings through the existing restore path, and refuses, rather than
+  substituting another conversation or starting fresh, if the history changed
+  or became unavailable. A running or unverified terminal is left alone. With
+  0.7.3's saved seat settings by node id
+  ([#2051](https://github.com/mvschwarz/openrig/pull/2051)), this completes
+  the three stages agreed in
+  [#1837](https://github.com/mvschwarz/openrig/issues/1837).
+  ([#2109](https://github.com/mvschwarz/openrig/pull/2109))
+
+**Task views**
+
+- `rig terminal open` and the daemon's terminal preview accept `qitem:<id>`: the current local
+  owners of that task, meaning its pending, in-progress and blocked rows
+  through explicitly linked successors, without closed predecessors or past
+  senders. Stopped, missing, external and unresolved owners are named, and
+  incomplete lineage is reported rather than cut short. It's a read-only
+  membership view: it doesn't follow the task as it moves, and an explicit
+  open still has the provider's usual workspace effects. Preview and open
+  responses now carry `viewDescriptor` (`kind`, `name`).
+  ([#2131](https://github.com/mvschwarz/openrig/pull/2131))
+
+**The kernel at daemon start**
+
+- **Startup brings back a kernel only when this start found it down,** which
+  is what a reboot leaves. A kernel that was already down before this start
+  stays down, with its old conversations. `rig start` and `rig daemon start
+  --wait-for-kernel` say why and how to bring it back (`rig up kernel
+  --existing`); `--wait-for-kernel` no longer prints a timeout and an
+  authentication hint for a kernel skipped on purpose, and still exits 1.
+  ([#2146](https://github.com/mvschwarz/openrig/pull/2146),
+  [#2152](https://github.com/mvschwarz/openrig/pull/2152),
+  [#2156](https://github.com/mvschwarz/openrig/pull/2156))
+
+**Seats and runtimes**
+
+- **A Claude seat stopped by a rate limit is recognised** from Claude's
+  `StopFailure` hook, even when its status line shows no usage window, and
+  pending handoffs use the existing wake ladder and retry timer. A delayed
+  older Claude hook no longer replaces the current state in send readiness,
+  attention, review telemetry or the wake scheduler.
+  ([#1908](https://github.com/mvschwarz/openrig/pull/1908),
+  [#2113](https://github.com/mvschwarz/openrig/pull/2113))
+- **`rig handover --source fresh` on Claude** reports the restore packet
+  delivered only when Claude's composer shows it submitted. If it's still
+  staged after one retried Enter, the handover fails with
+  `context_delivery_failed`.
+  ([#1863](https://github.com/mvschwarz/openrig/pull/1863))
+- **A periodic snapshot between separate named-agent launches** keeps a down
+  agent's resume source, including its restore policy and community harness
+  selection, without claiming it was running.
+  ([#2123](https://github.com/mvschwarz/openrig/pull/2123), fixes
+  [#1822](https://github.com/mvschwarz/openrig/issues/1822))
+- **A resumed Pi seat** no longer prints `Entry not found` after `READY` on
+  every restart: a saved history cursor Pi doesn't hold is recovered.
+  ([#2122](https://github.com/mvschwarz/openrig/pull/2122))
+
+**People, parks and reviews**
+
+- **A park on a registered person's own address** (`<id>@external`) resolves
+  only from a Slack reply written after its question was posted
+  ([#2106](https://github.com/mvschwarz/openrig/pull/2106)), keeps its
+  `--summary` and `--evidence-ref`
+  ([#2107](https://github.com/mvschwarz/openrig/pull/2107)), and shows in
+  NEEDS YOU in the slice, mission and rig reviews, with asks to that address,
+  instead of the person being listed as an agent holding work
+  ([#2108](https://github.com/mvschwarz/openrig/pull/2108)).
+- **The reviews' agents band** counts only open rows as held, and a seat with
+  work in progress reads as working even when it also has an older blocked
+  row. ([#2136](https://github.com/mvschwarz/openrig/pull/2136), fixes
+  [#2129](https://github.com/mvschwarz/openrig/issues/2129))
+
+**TUI and CLI**
+
+- `rig` or `rig tui` started in a terminal that reports 0 rows or columns
+  waits and draws on the next resize, instead of crashing on its first frame.
+  ([#2099](https://github.com/mvschwarz/openrig/pull/2099))
+- The TUI lists a host's rigs by name, in `rig ps`'s order.
+  ([#2134](https://github.com/mvschwarz/openrig/pull/2134), fixes
+  [#2130](https://github.com/mvschwarz/openrig/issues/2130))
+- Opening a seat whose runtime doesn't report effort no longer shows a failed
+  read in the TUI header. ([#2102](https://github.com/mvschwarz/openrig/pull/2102))
+- `rig whoami`, `rig down`, `rig ask` and `rig health` in a tmux pane without
+  `OPENRIG_SESSION_NAME` no longer print tmux's `invalid option` errors.
+  ([#2103](https://github.com/mvschwarz/openrig/pull/2103))
+
+**Also:** an artifact whose frontmatter header isn't closed no longer wakes an
+artifact-pool worker as ready
+([#2016](https://github.com/mvschwarz/openrig/pull/2016)); with Pi cost
+reporting on, a seat that now runs Claude or Codex shows its current cost, and
+a charge lookup retried during a re-post finishes
+([#2120](https://github.com/mvschwarz/openrig/pull/2120)); the
+`queue-handoff` skill says to write a handoff that can be acted on after one
+read, as guidance and not a filter
+([#2154](https://github.com/mvschwarz/openrig/pull/2154),
+[#2155](https://github.com/mvschwarz/openrig/pull/2155)); the harness preset
+guide says to check that a harness answers on its own first
+([#2116](https://github.com/mvschwarz/openrig/pull/2116)); seat environments
+are exercised on Linux in CI
+([#2101](https://github.com/mvschwarz/openrig/pull/2101)); and portability and
+test fixes ([#2147](https://github.com/mvschwarz/openrig/pull/2147),
+[#2148](https://github.com/mvschwarz/openrig/pull/2148),
+[#2149](https://github.com/mvschwarz/openrig/pull/2149)).
+
+### Behaviour that changed
+
+- A kernel that was already down before daemon start is no longer restored on
+  that start; see "The kernel at daemon start".
+- `rig daemon start --wait-for-kernel` for a kernel skipped on purpose prints
+  the reason and the way back instead of a timeout header and an
+  authentication fix. It still exits 1.
+- Work waiting for a member nothing can wake is reported to its person, not
+  to that member.
+- A caller granted through a grant service can't address its own or another
+  caller seat, or set OpenRig's Slack provenance or a workflow's internal
+  sign-off tags.
+- A park on a registered person's address resolves only from a reply written
+  after its question.
+
+### Known issues
+
+- **A webhook ring proves only that the endpoint accepted it.** Whether the
+  member woke and took the work is its claim on the row. An ordinary chat app
+  can't receive unsolicited events, so a hosted assistant without a
+  webhook-capable runner sees work only when it checks.
+  ([#2115](https://github.com/mvschwarz/openrig/pull/2115))
+- **A recorded principal is attribution, not authentication.** The daemon
+  records what the MCP adapter tells it about the caller it checked.
+  ([#2098](https://github.com/mvschwarz/openrig/pull/2098))
+- **The task view (`qitem:<id>`) is membership only.** It doesn't follow a
+  task as it moves, keep a reusable viewer, or change who may type into a
+  seat. ([#2131](https://github.com/mvschwarz/openrig/pull/2131))
+- **Resuming a reviewed history has no CLI yet.** `rig seat history` lists
+  and checks; applying a choice is through the daemon's startup review, for
+  clients. Choices are limited to this node's eligible Claude and Codex
+  histories in the current compatible snapshot.
+  ([#2100](https://github.com/mvschwarz/openrig/pull/2100),
+  [#2109](https://github.com/mvschwarz/openrig/pull/2109))
+- **The kernel's restore evidence lasts one daemon process.** If the first
+  daemon after a reboot exits before the kernel's restore starts a seat, the
+  next start finds it already down and leaves it down, saying so. Bring it
+  back with `rig up kernel --existing`. (0.7.3's issue of a long-gone kernel
+  returning on daemon start is fixed by
+  [#2146](https://github.com/mvschwarz/openrig/pull/2146).)
+- **OpenClaw's interrupted local turns are still not recovered.** The seat
+  shows as needing attention, and a send submits a new turn: inspect the pane
+  with `rig capture` first. For an unattended seat, use the Gateway preset.
+  ([#2092](https://github.com/mvschwarz/openrig/pull/2092))
+- **Whether processes a community harness starts outside its pane survive is
+  still not exercised**, and `rig down` stops only the pane.
+  ([#2088](https://github.com/mvschwarz/openrig/pull/2088),
+  [#2077](https://github.com/mvschwarz/openrig/pull/2077))
+- **Hermes:** as in 0.7.3, the example preset doesn't declare `resume`, and
+  whether a seat's Hermes gateway exits once idle is unknown, although
+  `docs/reference/community-harness-presets.md` says it exits. Check with `ps`
+  before relying on it.
+- **A Codex seat's own permission settings** were enforced natively in 0.7.3's
+  test with Codex 0.164.0; other Codex versions weren't tested.
+  ([#2048](https://github.com/mvschwarz/openrig/pull/2048))
+- **Upgrading doesn't restore Codex hook approvals that an earlier OpenRig
+  removed.** If Codex asks for hook review, approve the hooks you mean to
+  enable. ([#2035](https://github.com/mvschwarz/openrig/pull/2035))
+- **After the last managed session ends, an empty tmux server may stay
+  running.** That alone doesn't mean an agent is still running.
+- **Slack threads from before 0.7.3** keep 0.7.3's limits: a reply on an ask
+  parked before the fix can still reach the person asked, and a message
+  received before 0.7.3 can still show OpenRig's routing footer, policy text
+  and attachment paths when an agent asks a person about it.
+  ([#2119](https://github.com/mvschwarz/openrig/pull/2119),
+  [#2124](https://github.com/mvschwarz/openrig/pull/2124))
+- **Shared channels are Slack only**, and the primary user's authority in the
+  TUI is attribution, not authentication.
+  ([#1992](https://github.com/mvschwarz/openrig/pull/1992),
+  [#2006](https://github.com/mvschwarz/openrig/pull/2006))
+- **The cmux path wasn't tried in a real cmux**, and a remote host's rigs are
+  read over HTTP only. ([#2080](https://github.com/mvschwarz/openrig/pull/2080),
+  [#2052](https://github.com/mvschwarz/openrig/pull/2052))
+- **Window sizing:** a browser that closes while the daemon is down leaves the
+  window at the browser's size until the next web-terminal visit, and a
+  `--readable` capture isn't atomic with a person's own edits.
+  ([#2065](https://github.com/mvschwarz/openrig/pull/2065),
+  [#2075](https://github.com/mvschwarz/openrig/pull/2075))
+- **The runtime-switch handover change** hasn't been shown to cure the
+  unsent-handover case that prompted it.
+  ([#2054](https://github.com/mvschwarz/openrig/pull/2054))
+- The community harness field keeps its name (`communityHarness`) and its
+  original database update; renaming them would break a return to 0.7.2.
+- Not in this release: verified task viewers
+  ([#2133](https://github.com/mvschwarz/openrig/pull/2133)), switching a seat
+  to an isolated account home
+  ([#2138](https://github.com/mvschwarz/openrig/issues/2138),
+  [#2143](https://github.com/mvschwarz/openrig/pull/2143)), source evidence for
+  resolving stuck findings
+  ([#2153](https://github.com/mvschwarz/openrig/pull/2153)), the stale-pane
+  repair message ([#2157](https://github.com/mvschwarz/openrig/pull/2157)),
+  current launch evidence for fresh handovers
+  ([#2159](https://github.com/mvschwarz/openrig/pull/2159)), and a warning for
+  out-of-range MCP daemon ports
+  ([#2160](https://github.com/mvschwarz/openrig/pull/2160)). Hermes resume and
+  measured idle exit, and OpenClaw interrupted-turn recovery, aren't in this
+  release either. The limits listed for 0.7.3 still apply unless a change
+  above names them fixed.
+
+### Credits
+
+Thank you to everyone whose work is in this release: @Aschenbath (#1908,
+#2101, #2113), @Nikhi00718 (#1863, #2016, #2106, #2107, #2108), @bradvin
+(#2100, #2109), @dajiaohuang (#2123) and @seanmcn (#2147, #2148, #2149). And
+thank you to @pr0xc3nt4ur1 (#1822), @duncanleung (#2130), @ziadeh (#2129),
+@alperta (#99), @smota (discussion #2117) and @oSquashBlossomo, whose reports
+and reviews led to fixes here.
+
 ## [0.7.3]
 
 OpenRig 0.7.3 lets a team work with several people. You can register more than
@@ -330,10 +1281,11 @@ from the commit tagged `v0.7.3`. The changes since 0.7.2:
   current TUI sends it (a request without a seat identity); an older TUI's
   answer on someone else's ask is refused.
   ([#2006](https://github.com/mvschwarz/openrig/pull/2006))
-- **A Codex seat's permission settings are passed as launch flags; that Codex
-  enforces them was not proven natively.** Going back to an older OpenRig keeps
-  them unread (see Before you upgrade). The Codex trust fixes stop new duplicate
-  hook groups but don't remove existing ones.
+- **A Codex seat's permission settings were enforced natively in our test with
+  Codex 0.164.0,** including after a managed restore and a restart. Other Codex
+  versions weren't tested. Going back to an older OpenRig keeps them unread (see
+  Before you upgrade). The Codex trust fixes stop new duplicate hook groups but
+  don't remove existing ones.
   ([#2048](https://github.com/mvschwarz/openrig/pull/2048),
   [#2035](https://github.com/mvschwarz/openrig/pull/2035))
 - **The cmux path wasn't tried in a real cmux**, and a remote host's rigs are
@@ -398,8 +1350,7 @@ reported [#1653](https://github.com/mvschwarz/openrig/issues/1653) (worked on in
 reported [#980](https://github.com/mvschwarz/openrig/issues/980) (addressed by
 [#2081](https://github.com/mvschwarz/openrig/pull/2081)); @rakoort, whose
 suggestion in [#1679](https://github.com/mvschwarz/openrig/issues/1679) became
-[#2086](https://github.com/mvschwarz/openrig/pull/2086) (the issue's main
-report is still open); @george661, whose
+[#2086](https://github.com/mvschwarz/openrig/pull/2086); @george661, whose
 [#1710](https://github.com/mvschwarz/openrig/issues/1710) led to
 [#2093](https://github.com/mvschwarz/openrig/pull/2093).
 

README.md

@@ -26,9 +26,9 @@ use:
 
 ```sh
 # Claude Code
-npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.0/skills/rigs --skill rigs --agent claude-code
+npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.5/skills/rigs --skill rigs --agent claude-code
 # Codex
-npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.0/skills/rigs --skill rigs --agent codex
+npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.5/skills/rigs --skill rigs --agent codex
 ```
 
 Then type `/rigs` in Claude Code, or `$rigs` in Codex, and say what you want built. The skill has your agent tell
@@ -68,11 +68,11 @@ Requires Node.js 22 or 24 and tmux, on macOS or Linux, and one working Claude Co
 
 ### Install it yourself
 
-This runs OpenRig's install script from the `v0.7.0` release. It checks Node.js and npm, installs the latest published `@openrig/cli` with `npm install -g`, runs the Node.js and SQLite check, then runs `rig setup --dry-run` and `rig setup`. The first line prints that plan and changes nothing:
+This runs OpenRig's install script from the `v0.7.5` release. It checks Node.js and npm, installs the latest published `@openrig/cli` with `npm install -g`, runs the Node.js and SQLite check, then runs `rig setup --dry-run` and `rig setup`. The first line prints that plan and changes nothing:
 
 ```bash
-curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.0/scripts/install.sh | sh -s -- --dry-run
-curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.0/scripts/install.sh | sh
+curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.5/scripts/install.sh | sh -s -- --dry-run
+curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.5/scripts/install.sh | sh
 ```
 
 If a provider isn't signed in yet, the last step reports `FAILED [4/4]`, and `rig setup` lists the sign-in under "Some steps need attention". If the only remaining failures are provider sign-ins, the install steps finished; sign in to each selected provider as below and continue.

ROADMAP.md

@@ -6,23 +6,9 @@ This isn't a schedule or a promise: we release often, and an item ships when a g
 To see what's being worked on, by release and theme, open the [OpenRig roadmap board](https://github.com/users/mvschwarz/projects/1).
 Issues labelled `good first issue` are also listed on the [contribute page](https://github.com/mvschwarz/openrig/contribute).
 
-## Released in 0.7.2
+## Released in 0.7.6
 
-[0.7.2 is published](https://github.com/mvschwarz/openrig/releases/tag/v0.7.2).
-See its release notes for what shipped, upgrade instructions, contributor credits and the remaining testing gaps.
-
-Hermes and Cursor through the shared plug-in point were not included. The Codex counterpart to per-member Claude
-permission rules was not included either; that work has since merged on main in
-[#2048](https://github.com/mvschwarz/openrig/pull/2048). These follow-ups have no release promise here.
-
-## Active: 0.7.3 — people and agent teams through one gateway
-
-- **Participant foundation:** people and vetted shared channels, an explicit default, organization and groups, and readable policy.
-- **Attributed shared-channel conversations:** reach a support seat with who spoke, the configured role and policy, and the thread;
-  send answers back to that thread.
-- **Actions and perspective for the primary user:** see whose work needs attention and preserve who actually spoke or acted.
-
-These are the active themes, not claims that the changes have shipped or a release-date promise.
+[0.7.6 is published](https://github.com/mvschwarz/openrig/releases/tag/v0.7.6): follow a slice's work in one terminal workspace, plus steadier daemon start and delivery. See its release notes for what shipped, known issues, upgrade instructions and contributor credits.
 
 ## How to help
 

docker/testbed/Dockerfile

@@ -49,7 +49,8 @@ COPY ${OPENRIG_TARBALL} /tmp/openrig.tgz
 RUN set -eux; \
     npm install -g /tmp/openrig.tgz; \
     rm -f /tmp/openrig.tgz; \
-    rig --version
+    rig --version; \
+    rig --help
 
 # --- layer 4: stub runtime assets (zero-token by construction; staged by the build verb into the
 #     build context as ./stub-assets/, hashed into the manifest via deriveStubAssetsHash) ---

docs/as-built/architecture/adapters-and-runtimes.md

@@ -281,6 +281,34 @@ resume launches (`codex-runtime-adapter.ts:378`, `codex-resume.ts:103`).
   `no_conversation_found` (`:429–434`) and with `recovery:
   "attention_required"` plus pane evidence on an `attention_required` probe
   (`:444–450`) — the adapter never relaunches on its own.
+- **Fresh in-place handover readiness.** `assessFreshInPlaceProbe` qualifies an
+  otherwise ready history probe only for the matching fresh launch, binding and
+  launch generation. The current pane must show a prompt, and `assessManagedProbe`
+  must confirm the new launch's exact session token in the same foreground process
+  across two observations, with the pane and launch still current. It reuses the
+  existing managed identity proof, including its process fingerprint. A process
+  label alone, old scrollback, an unavailable screen, an unreadable process list
+  or an unidentified Claude process cannot admit the handover. Those cases stay
+  `awaiting_runtime`; a positive identity conflict and visible approval or login
+  gates keep their specific results. The handover waits within its existing
+  readiness budget before sending the context packet.
+
+  The fresh path requires the last `❯` prompt to sit inside Claude's complete
+  input frame: its top rule directly above the prompt and a bottom rule below.
+  A shell's echoed launch command cannot borrow an earlier header or input
+  frame, even when the matching Claude process is already running. A header
+  and prompt alone, or a partially drawn frame, remain pending. This may delay
+  readiness until the full frame is visible; other launch paths keep their
+  existing screen predicates.
+
+  Previously, unavailable observations and ambiguous headerless prompts could
+  retain the old ready result. `claude-fresh-in-place-scope.test.ts` now checks
+  missing observations, unknown or changing process identity, retained UI above
+  direct and script launch commands, partial frames, recovery and normal prompt
+  controls. `claude-handover-readiness.test.ts` exercises the real handover
+  service with migrated SQLite, modeled pane/process observations and one synthetic
+  clock. Readiness and the later packet-submission receipt remain separate checks;
+  these tests do not run a live Claude account.
 
 ### CodexRuntimeAdapter (`codex-runtime-adapter.ts:65`)
 
@@ -370,6 +398,33 @@ adapter never closes a session on it and reports `rex_outcome_unknown`.
 Herdr and Rex build their layout trees from one grid helper
 (`buildTileGrid`, `tile-layout.ts`).
 
+### Tmux server selection
+
+Before reconciliation, `startup.ts` calls `selectTmuxServer` in
+`domain/tmux-server-selection.ts`. `terminal.tmux_server` defaults to `inherit`;
+`isolated` selects a socket derived from the canonical OpenRig home. The effective
+choice is persisted in `tmux-server.json`. A registered live seat, or an
+uninspectable old server, defers a requested change until a later restart.
+The persisted `pinned` flag and active marker make isolation and migration records
+mandatory, while ordinary inherited records remain advisory. Returning to inheritance
+commits an unpinned record before clearing the marker, so offline cleanup cannot reuse
+a stale inherited socket or keep enforcing a completed isolation choice.
+
+`TmuxAdapter` receives the selection's `executionArgs` for both shell and argv
+transports, including its direct probe/cleanup calls. They are empty for ordinary
+`inherit` mode, preserving tmux's environment-based selection; `serverArgs` still
+names the server for attachments. The opt-in control client uses the same execution
+arguments for its connection, probes and fallback commands. The same startup choice feeds node
+inventory, `TerminalService.status()` and previews, composed panes, shared TUI attachment and
+CLI viewer management. Inherited SSH panes retain the remote-tmux replacement
+wrapper and plain recovery command without requiring remote OpenRig. Isolated
+SSH panes resolve the remote instance through its own `rig terminal attach`.
+For isolation, self-attachment checks the caller's socket before binding;
+offline destruction reads the persisted socket before removing state.
+
+See [tmux server selection](../../reference/tmux-servers.md) for configuration
+loading, first-upgrade migration, discovery scope and platform behavior.
+
 ## 3. Resume honesty
 
 The daemon does not assume a harness resumed just because the launch command

docs/as-built/architecture/agent-spec-and-startup.md

@@ -65,6 +65,35 @@ resolve → project → deliver pre-launch files → persist replay context →
 
 ## 2. Execution and projection types
 
+The member's optional `claude_mcp: none` is normalized as `claudeMcp` and
+persisted independently in `nodes.claude_mcp` (migration 141). The
+expansion HTTP route preserves both `claude_mcp` and `claudeMcp`, including
+invalid values for the canonical validator to reject. An authored relaunch
+updates the seat choice after launch eligibility and startup-source resolution;
+an already-bound seat or a failed tmux launch keeps its prior value. Export
+describes the current runtime; a dormant choice on another runtime is omitted
+and must be restored in the Claude declaration when rebuilding from that export.
+The YAML export route names those agents in the existing `# Not carried:` report;
+the JSON export is unchanged. `rigspec-routes.test.ts` covers both pod-aware and
+legacy exports, omission controls and successful JSON export.
+
+The shared `StartupOrchestrator` reads the current seat record, so both an authored start
+and saved-context restore pass the choice to Claude. Legacy resume and handover
+successors read it separately; a runtime-switch successor reads it only when
+the target is Claude. `claude-mcp.ts` builds the strict empty-server arguments
+for the adapters, shutdown resume probes and displayed recovery commands,
+including the conversation picker. Shutdown skips a probe with an unreadable
+saved choice and keeps the resume token without adding a shutdown refusal.
+The native lineage selectors recognize these exact
+arguments without treating JSON contents as session options. Invalid stored
+values refuse launch. With no choice, the adapters add no MCP arguments.
+
+`claude-mcp-lifecycle.test.ts` exercises real schema/storage, export, snapshot
+and restore with simulated tmux/provider processes. `claude-mcp-launch.test.ts`
+executes managed and unmanaged shell commands against a private argv-printing
+fake, including fresh/resume/fork and legacy resume. These tests verify command
+plumbing and identity handling; they do not claim live provider enforcement.
+
 Restore/snapshot types are detailed in `lifecycle-snapshot-restore.md`; the
 spec/projection types live here.
 

docs/as-built/architecture/coordination-primitive.md

@@ -178,15 +178,58 @@ the destination's orchestrator when one is derivable, otherwise to the
 destination itself, and that seat may be dead. Overdue claims go to their
 destination. Repeated detections reuse the existing finding without changing
 its route; when the source condition resolves, the sweep closes its own finding.
+Previously, falling out of a reporting query could itself close a delivery or
+unclaimed finding. An unconfirmed retry, a fresh human-notification post window,
+an unreadable human registry or a larger reporting-age setting could therefore
+look like successful recovery. These findings now require source evidence:
+closure, pickup or parking of the work, a verified/accepted wake, or a posted
+notification for the current human recipient and episode, as appropriate to the
+finding. `delivered-ack-pending`, `indeterminate:…`, missing results and unrelated
+legacy human receipts do not prove resolution. Uncertainty keeps an existing
+finding open without itself sending another wake or creating another finding.
+A verified wake resolves a wake failure, not an unclaimed obligation: delivery
+does not prove pickup.
+
+A legacy human notice with no owner-notification history can still resolve from
+the existing item-scoped posted ledger when its recipient is confirmed human.
+This covers older pending asks and parks without inventing an episode that was
+never recorded. A registered alias needs a readable registry to establish its
+human recipient; reserved human references remain known during an outage.
+Legacy receipts carry no recipient key, so a destination reroute requires a new
+matching notification episode. A legacy park is accepted only when no earlier
+notice or reroute predates it, and later history contains only receipts or
+asking-agent reroutes within that park. A re-park, an exit from the park or an
+ambiguous append cannot establish the same recipient without owner history;
+the finding waits for a new matching episode or other source resolution.
+Rerouting only the asking agent within an otherwise unchanged park preserves
+its human recipient (`blockedOn`).
+Modern notifications still require their current episode, and an unrelated
+legacy human receipt cannot resolve an agent wake.
+
+A known hold still leaves recovery with the existing mechanism: an ask on
+Steering without an alert, a wake held by the seat typing guard, or a live
+quiet-release job. Its closure note names that disposition rather than claiming
+delivery or pickup. An unreadable quiet hold is not positive closure evidence.
+Claimed work in either `pending` or `in-progress` uses the same pickup derivation;
+returning a row to `pending` while retaining its claim no longer hides a later
+stall or closes its existing stalled-claim finding. A retained failed wake cannot
+override that pickup evidence. The agent's claim still does not establish delivery
+of a separate notification to a person.
+
 That automatic closure records `stuck-sweep@system` as the actor in both the
-transition and the `queue.updated` audit event. Previously it credited the
-finding's source agent with the detector's action. The finding's source,
+transition and the `queue.updated` audit event, with the source evidence in the
+delivery and unclaimed-obligation closure notes. Previously it credited the finding's source agent
+with the detector's action. The finding's source,
 destination and claim remain its existing ownership records; the underlying work
 is not reassigned, and existing transition history is not rewritten.
 
-`queue-stuck-sweep.test.ts` checks this with migrated SQLite: pending and claimed
-findings close once, their transition and event name the sweep, and ownership
-stays intact. The posted-human-notice case also checks the actor while leaving
+`queue-stuck-sweep-resolution.test.ts` checks unconfirmed retries, retained
+claims, current human-notification episodes and registry outages with migrated
+SQLite, including positive recovery controls and no repeated send on an
+unresolved retry. `quiet-hold-custody.test.ts` checks existing findings with
+readable and unreadable quiet custody. `queue-stuck-sweep.test.ts` checks that
+pending and claimed findings close once, name the sweep in their transition and
+event, and keep their ownership. The posted-human-notice case also checks the actor while leaving
 the person's original ask open. These checks exercise the detector and queue
 repository, not a live agent or connector.
 

docs/as-built/architecture/lifecycle-snapshot-restore.md

@@ -293,7 +293,7 @@ cited here by number with the code that carries them):
   for one seat. `resume` captures an eligibility-checked `auto-rehydrate`
   snapshot when none exists and goes through `launchSingleNode`.
 - **Seat handover** — `rig seat handover <seat> --source
-  fresh|rebuild|fork:<id>|discovered:<id> --reason <reason> [--dry-run]`
+  fresh|rebuild|fork:<seat>|discovered:<id> --reason <reason> [--dry-run]`
   (the reason is required, dry run included; without it the command exits 2
   before contacting the daemon)
   (`packages/cli/src/commands/seat.ts:364`) calls

docs/as-built/cli-reference.md

@@ -25,7 +25,7 @@ below `rig` (including groups and the hidden `restore apply` command), and
 **1,304 explicitly registered option objects**, including the root version
 option. Aliases do not add command objects; short/long spellings of one
 option do not add option objects. Commander-generated help is additional.
-These are counts for that source snapshot, not a claim about a deployed release.
+These are counts for that source snapshot, not a claim about a deployed release. The provider-switch and set-runtime rows include the later account-home options; those additions are not included in the snapshot counts.
 
 ## Reading this reference
 
@@ -354,6 +354,23 @@ the sections that were not read are `null`.
 | `rig host list` | `ls` | `--json` |
 | `rig host doctor <id>` | — | `--posture <profile>`<br>`--public-addr <ip>`<br>`--json` |
 
+Registration makes a host visible in topology; remote FEED attention remains opt-in.
+The human-readable success output of `host add` and `host pair` now includes the FEED
+next step alongside the existing `host doctor` suggestion. For an HTTP entry named
+`laptop`, run this on the local instance to include its attention items:
+
+```sh
+rig config set feed.subscriptions.laptop.enabled true
+```
+
+The key uses the registry id, including any dots, and defaults to `false`. Registration
+does not change that setting. SSH entries instead explain that FEED needs an HTTP entry;
+ids reserved by feed settings (such as `progress`) get a different-id suggestion rather
+than an invalid command. `--json` keeps its single JSON result without these hints.
+The host CLI tests execute the advertised command against an isolated config and cover
+registration, reserved names and JSON output; pairing responses are simulated, so they
+do not establish connectivity to a live peer.
+
 ### ps
 
 | Invocation | Aliases | Declared options |
@@ -468,7 +485,7 @@ the sections that were not read are `null`.
 | Invocation | Aliases | Declared options |
 |---|---|---|
 | `rig slack` | — | — |
-| `rig slack setup` | — | `--channel <id>`<br>`--inbound-destination <session>`<br>`--minimum-level-that-posts <level>`<br>`--minimum-level-that-interrupts <level>`<br>`--source-label <label>`<br>`--secrets-env-file <path>`<br>`--required-scopes <csv>`<br>`--reason <reason>`<br>`--actor <actor>` |
+| `rig slack setup` | — | `--channel <id>`<br>`--inbound-destination <session>`<br>`--minimum-level-that-posts <level>`<br>`--minimum-level-that-interrupts <level>`<br>`--source-label <label>`<br>`--secrets-env-file <path>`<br>`--required-scopes <csv>`<br>`--reactions <mode>`<br>`--reason <reason>`<br>`--actor <actor>` |
 | `rig slack status` | — | `--json` |
 | `rig slack manifest` | — | `--url`<br>`--json` |
 | `rig slack verify` | — | `--json` |
@@ -777,8 +794,8 @@ Only Claude and Codex tenures can be woken.
 | `rig provider accounts` | — | `--json`<br>`--provider <p>`<br>`--account <a>` |
 | `rig provider bindings` | — | `--json`<br>`--provider <p>`<br>`--account <a>` |
 | `rig provider signals` | — | `--json`<br>`--provider <p>`<br>`--account <a>` |
-| `rig provider precheck` | — | `--seat <s>` **required**<br>`--to-account <a>`<br>`--token-file <path>`<br>`--stored-login`<br>`--json` |
-| `rig provider switch` | — | `--seat <s>` **required**<br>`--to-account <a>`<br>`--token-file <path>`<br>`--stored-login`<br>`--force-unsafe`<br>`--json` |
+| `rig provider precheck` | — | `--seat <s>` **required**<br>`--to-account <a>`<br>`--token-file <path>`<br>`--api-key-file <path>`<br>`--stored-login`<br>`--json` |
+| `rig provider switch` | — | `--seat <s>` **required**<br>`--to-account <a>`<br>`--token-file <path>`<br>`--api-key-file <path>`<br>`--stored-login`<br>`--force-unsafe`<br>`--clear-account-home`<br>`--account-home <path>`<br>`--reason <text>`<br>`--model <id>`<br>`--codex-config-profile <name>`<br>`--dry-run`<br>`--json` |
 | `rig provider moves` | — | `--seat <seat>`<br>`--limit <n>`<br>`--json` |
 | `rig provider moves report` | — | `--file <path>` **required**<br>`--to <seat>` **required**<br>`--follows <qitemId>`<br>`--evidence-ref <ref>`<br>`--json` |
 
@@ -967,9 +984,10 @@ for input and receipt examples, rollback, and snapshot behavior.
 | `rig seat set-terminal-policy <seat>` | — | `--preset <preset>` **required**<br>`--helper <addresses...>`<br>`--rig-default`<br>`--reason <text>` **required**<br>`--operator <address>`<br>`--json`<br>`--peer-compaction <decision>`<br>`--peer-model <decision>`<br>`--self-compaction <decision>`<br>`--self-model <decision>`<br>`--openrig-compaction <decision>`<br>`--openrig-model <decision>` |
 | `rig seat set-permissions <seat>` | — | `--mode <mode>`<br>`--rules <file>`<br>`--codex-permissions <file>`<br>`--codex-profile <name>`<br>`--reason <text>` **required**<br>`--operator <address>`<br>`--json` |
 | `rig seat set-effort <seat>` | — | `--effort <level>`<br>`--clear`<br>`--reason <text>` **required**<br>`--operator <address>`<br>`--json` |
+| `rig seat set-permissions-to <seat>` | — | `--to <person>`<br>`--clear`<br>`--deny`<br>`--reason <text>` **required**<br>`--operator <address>`<br>`--json` |
 | `rig seat effort <seat>` | — | `--json` |
 | `rig seat set-model <seat>` | — | `--model <id>` **required**<br>`--reason <text>` **required**<br>`--operator <address>`<br>`--json` |
-| `rig seat set-runtime <seat>` | — | `--runtime <runtime>`<br>`--reason <text>`<br>`--model <id>`<br>`--effort <level>`<br>`--codex-config-profile <name>`<br>`--permission-mode <mode>`<br>`--confirm-timeout <seconds>`<br>`--dry-run`<br>`--confirm <switch-id>`<br>`--note <text>`<br>`--recover`<br>`--json` |
+| `rig seat set-runtime <seat>` | — | `--runtime <runtime>`<br>`--reason <text>`<br>`--model <id>`<br>`--effort <level>`<br>`--codex-config-profile <name>`<br>`--clear-account-home`<br>`--account-home <path>`<br>`--token-file <path>`<br>`--account-label <label>`<br>`--permission-mode <mode>`<br>`--confirm-timeout <seconds>`<br>`--dry-run`<br>`--confirm <switch-id>`<br>`--note <text>`<br>`--recover`<br>`--json` |
 | `rig seat set-cwd <seat>` | — | `--cwd <directory>` **required**<br>`--reason <text>` **required**<br>`--operator <address>`<br>`--json` |
 | `rig seat launch <seat>` | — | `--fresh` **required**<br>`--reason <text>`<br>`--plan`<br>`--stop`<br>`--operator <address>`<br>`--json` |
 | `rig seat restart <seat>` | — | `--keep-conversation` **required**<br>`--reason <text>`<br>`--plan`<br>`--verify`<br>`--expect-account <text>`<br>`--operator <address>`<br>`--host <id>`<br>`--json` |

docs/as-built/test-layers.md

@@ -88,6 +88,11 @@ are not yet runnable checks.
   credentials, and outbound network blocked except to localhost (see below). A test that reads
   your real home directory, a running daemon, the network or a credential can pass on your
   machine and fail in CI.
+- **Mixed fixture clocks.** `compaction-command-submission.test.ts` uses one clock for the
+  enforcer, the transport's injected `now` and sleeps, and its `Date.now` idle deadline. Previously,
+  real-time transport waits could exhaust the fixture's 20 ms budget during a scheduler pause
+  even though the modeled seat was idle. Restore the `Date.now` spy after each test; keep real
+  deadline behavior covered by the transport tests. This changes the fixture, not production timeouts.
 - **Docs location.** `check-docs-guard` fails if a tracked file under `docs/` sits outside
   `docs/as-built/`, `docs/reference/` or `docs/releases/`. `docs/DESIGN.md` is the one allowed
   exception.
@@ -105,7 +110,7 @@ credentials are supplied.
 | `repo-checks` | ubuntu | `npm run test:repo` |
 | `package-tests` | macOS, matrix `daemon`, `cli`, `tui`, `ui` (fail-fast off) | Installs tmux, runs `npm ci` and `npm run build`. Then runs `npm run test -w packages/<pkg>` under `env -i` with a scratch `HOME`/`TMPDIR`/`TMUX_TMPDIR` and a `sandbox-exec` profile that blocks outbound network except localhost and the job's own socket directory. `scripts/ci-macos-process-probe.mjs` runs first. |
 | `linux-seat-environment` | ubuntu | Builds all packages, then runs `seat-env-parent-session.integration.test.ts` with real CLI, daemon, tmux and stub seats. A network namespace has loopback only; tests run as the unprivileged runner user under `env -i` with scratch `HOME`/`TMPDIR`/`TMUX_TMPDIR`. Runs on pull requests and manual dispatch, like `package-tests`. |
-| `installed-scenario` | ubuntu | `bash scripts/run-pr-scenarios.sh` (eight container runs, described below). Uploads `dist/pr-scenarios/` as `installed-scenario-evidence`, including on failure. |
+| `installed-scenario` | ubuntu | `bash scripts/run-pr-scenarios.sh` (fifteen container runs, described below). Uploads `dist/pr-scenarios/` as `installed-scenario-evidence`, including on failure. |
 
 The Linux seat-environment job covers a gap in the macOS matrix: the integration test
 reads `/proc/<pid>/environ`, so macOS skips it. It checks both a new tmux server and an
@@ -153,7 +158,7 @@ Exit codes: 0 pass, 1 a leg failed, 2 lane busy, 3 the runner itself errored.
 | Piece | Where |
 |---|---|
 | `runtime: stub` adapter | `packages/daemon/src/adapters/stub-runtime-adapter.ts`, with `stub-runner.ts` (the process that runs in the pane), `stub-script.ts` (the launch-script format) and `stub-compaction.ts` / `stub-restore.ts` |
-| Scenario format and runner | `packages/daemon/test/helpers/scenario-*.ts`. Step verbs: `up`, `down`, `send`, `restart`, `daemon`, `seed_regression` are bound. `restore`, `emit`, `mutate`, `policy` parse but are unbound. Assertion surfaces: `ps`, `queue`, `stream`, `scope`, `pane`, `transcript`, `tui_socket`, `policy_provenance`. |
+| Scenario format and runner | `packages/daemon/test/helpers/scenario-*.ts`. Step verbs: `up`, `down`, `send`, `restart`, `daemon`, `seed_regression` are bound. `restore`, `mutate`, `policy` are rejected at load time (`UNBOUND_STEP_VERB`) because the runner has no shipped binding for them; `emit` parses but only `refocus` is bound. Assertion surfaces: `ps`, `queue`, `stream`, `scope`, `pane`, `transcript`, `tui_socket`, `policy_provenance`. |
 | Scenario library | `packages/test-system/scenarios/`: 20 scenarios, 7 stub topologies, `agents/`, `culture.md`, per-seat stub scripts in `scripts/` |
 | Runner fixtures | `packages/daemon/test/fixtures/scenarios/` (`scenario-01-per-seat-scripts`, `scenario-02-baton`, `scenario-10-one-view-state`) |
 | Host runner | `packages/daemon/scripts/run-scenarios.mjs` |
@@ -163,6 +168,28 @@ Exit codes: 0 pass, 1 a leg failed, 2 lane busy, 3 the runner itself errored.
 
 ### Host mode: while you're writing a scenario
 
+The tmux server selection regression also uses these scaffold and lifecycle helpers
+in `packages/daemon/test/tmux-server-isolation-daemon.test.ts`. It runs a real stub
+seat through config changes, daemon restarts, capture, self-attachment and cleanup,
+while native socket observations verify that a personal namesake survives. The
+test also removes the effective-server record and verifies that `destroy --all`
+leaves the daemon, database and both servers' sessions intact before restoring
+the record and checking normal cleanup. The
+YAML grammar has no config-write or socket-inventory surface, so these assertions
+live in the Vitest integration test. It does not exercise a real provider model,
+SSH connection or graphical terminal window.
+
+`packages/daemon/test/tmux-server-two-instances.test.ts` adds two real daemons
+sharing a scratch OS home and inherited personal tmux server, with separate
+OpenRig homes, databases and ports. Both enable isolation and control mode. It
+checks same-named seats through inventory, capture and verified writes, observes
+the marked control session on each selected socket, then stops one instance and
+checks that the other instance's pane identity, control session and delivery still
+work. The personal namesake remains untouched. This test requires a tmux version
+that supports control mode and the built CLI and daemon. Like CI's package suites,
+it should run with scratch homes and outbound network restricted to localhost;
+scripted seats prove terminal routing and rendered delivery, not model receipt.
+
 ```sh
 npm run build    # the runner drives packages/cli/dist/bin-wrapper.js
 node --import tsx packages/daemon/scripts/run-scenarios.mjs     # the fixture scenario-*.yaml files
@@ -185,10 +212,12 @@ node --import tsx packages/daemon/scripts/run-scenarios.mjs packages/daemon/test
 
 In CI the `installed-scenario` job runs `scripts/run-pr-scenarios.sh` with no arguments:
 both baton cases (`fixture` = `scenario-02-baton.yaml`, `library` = `queue-baton-survives-restart.yaml`),
-each run three times as healthy, then lost-baton, then healthy. Then the two stub-script
-cases (`transcript` = `transcript-reads-addressed-seat.yaml`, `capture` =
-`capture-returns-addressed-seat.yaml`) run once each, healthy only. That is eight fresh
-containers.
+each run three times as healthy, then lost-baton, then healthy (six containers); the
+`refocus-parked` effect case runs healthy, seeded, then healthy (three containers); and the
+six passing-only cases (`transcript` = `transcript-reads-addressed-seat.yaml`, `capture` =
+`capture-returns-addressed-seat.yaml`, `daemon-restart`, `refocus-active`,
+`refocus-parked-off`, `prompt-state`) run once each, healthy only (six containers). That is
+fifteen fresh containers.
 
 To run a case before you push, point `DOCKER_HOST` at a Docker engine you control over SSH:
 
@@ -198,10 +227,13 @@ DOCKER_HOST=ssh://<your-host> bash scripts/run-pr-scenarios.sh \
   --remote --case library --mode healthy --out dist/scenario-check
 ```
 
-- Options: `--case fixture|library|transcript|capture` (default: all four), `--mode
-  healthy|lost-baton` (default: healthy, lost-baton, healthy) and `--out <dir>` (default
-  `dist/pr-scenarios`). `--mode` applies to `fixture` and `library`; `transcript` and `capture`
-  always run once, healthy. `--mode lost-baton` leaves them out of a full run, and
+- Options: `--case fixture|library|transcript|capture|daemon-restart|refocus-parked|
+  refocus-active|refocus-parked-off|prompt-state` (default: all nine), `--mode
+  healthy|lost-baton|seeded` (default: healthy, lost-baton, healthy) and `--out <dir>` (default
+  `dist/pr-scenarios`). `--mode` applies to the baton cases (`fixture` and `library`) and to
+  the `refocus-parked` effect case; the six passing-only cases (`transcript`, `capture`,
+  `daemon-restart`, `refocus-active`, `refocus-parked-off`, `prompt-state`) always run once
+  each, healthy. `--mode lost-baton` leaves them out of a full run, and
   `--case transcript|capture --mode lost-baton` is refused with exit 2. Use a fresh `--out`
   directory for each run you want to keep.
 - `--remote` accepts only `ssh://` values for `DOCKER_HOST`, and it refuses if
@@ -268,7 +300,7 @@ means:
 | Queue pickup and wake | No stub worker reacts to a nudge by claiming an item, working on it and handing it back. | The baton scenarios prove the claim *survives a restart*. They don't prove delivery, pickup or wake. |
 | Reboot / tmux reset | `daemon: {op: restart}` restarts only the scenario daemon, and the tmux server keeps running. No step resets tmux. | Recycled pane IDs and stale bindings after a real reboot (the class behind #141) are not covered by a daemon restart. |
 | Multi-rig and same-name identities | The `up` step always uses the scenario's top-level `topology` and ignores a per-step override. No step archives or removes a rig. | Cross-rig scope, same-name generations and archive/remove routing (the class behind #174) need runner work before they can be tested. |
-| Step-time restore, compaction, mutation, policy | `restore`, `emit`, `mutate` and `policy` parse, but the runner throws `UnboundActionError` when it reaches them. Behaviours run only from a per-seat launch script (`env.stub_scripts`). A seat `restart` step maps to `rig launch`, which is not native resume or fork. | Step-time compaction and restore, and native resume and fork, are not covered. |
+| Step-time restore, compaction, mutation, policy | `restore`, `mutate` and `policy` steps are refused at load time (`UNBOUND_STEP_VERB`); `emit` parses but a non-`refocus` behavior throws `UnboundActionError` when the runner reaches it. Behaviours run only from a per-seat launch script (`env.stub_scripts`). A seat `restart` step maps to `rig launch`, which is not native resume or fork. | Step-time compaction and restore, and native resume and fork, are not covered. |
 | Skill projection | The stub projects skills into `<cwd>/.openrig/stub/skills/`. Claude Code projects a team seat's skills into the seat's generated plugin in OpenRig's state, and into `<cwd>/.claude/skills/` only for a seat without one. | A stub green says nothing about provider-specific projection (the class behind #159). |
 | Model and provider | No model, no tokens. Stub scripts refuse `usage_limit` because it only exists on real runtimes. | Model behaviour, provider usage limits and real context usage need live evals or a real runtime. |
 | Platform | CI's scenario job runs in a Linux container. | macOS process behaviour is covered by the macOS `package-tests` jobs, not by scenarios. |
@@ -309,9 +341,9 @@ of them. The other nine scenarios are authored but not yet runnable as meaningfu
 | `send-verify-means-rendered` | The `send` step drops `verify`, and the stub has no input loop | Build the input-consuming stub and assert a response derived from the nonce |
 | `home-divergence-preseed-visible` | The pipeline ignores `env.pre_existing_tmux`, and the default script never prints the expected `trust reached` | Recreate the real HOME split and whatever consumes it |
 | `stream-emit-durable-replay-live` | Step-time `emit` is unbound | Assert the actual stream contract |
-| `policy-posture-survives-structural-edit` | `policy` and `mutate` are unbound | Rewrite around current per-seat permission selection and an observed launch |
-| `locked-restore-never-permissive` | `policy` and `restore` are unbound | Keep the invariant (the selection survives a restore) and use current modes |
-| `one-view-state-after-mutation-storm` | The validator rejects its bare-list `equals` (`EQUALS_NOT_DECLARATIVE`), and `mutate` is unbound | Declare the comparable fields and real mutations |
+| `policy-posture-survives-structural-edit` | The validator refuses its `policy` and `mutate` steps at load time (`UNBOUND_STEP_VERB` — no shipped runner binding) | Rewrite around current per-seat permission selection and an observed launch |
+| `locked-restore-never-permissive` | The validator refuses its `policy` and `restore` steps at load time (`UNBOUND_STEP_VERB` — no shipped runner binding) | Keep the invariant (the selection survives a restore) and use current modes |
+| `one-view-state-after-mutation-storm` | The validator rejects its bare-list `equals` (`EQUALS_NOT_DECLARATIVE`) and refuses its `mutate` steps at load time (`UNBOUND_STEP_VERB`) | Declare the comparable fields and real mutations |
 | `kill-daemon-mid-handoff` | A pane `send` doesn't create the queue item the next assertion expects, and later assertions use a shape the queue reader doesn't return | Rebuild the setup before treating a red result as meaningful |
 
 In most of these files, `seed_regression` steps that come *after* the assertions are

docs/reference/agent-startup-guide.md

@@ -244,6 +244,15 @@ interactive"), and the post-launch files aren't sent. A Codex fork that the rig
 `session_source.mode: fork`, or an agent image's fork) is the exception: its timeout ends as `attention_required`,
 with its startup context kept for `rig seat continue` (see below). A seat handover's fork has its own readiness check.
 
+**Fresh Claude handovers.** `rig handover --source fresh` keeps the previous agent's scrollback. Before sending the
+handover context, OpenRig waits for the latest prompt inside Claude's complete input frame, and confirms that the
+process in the pane belongs to the new launch. A shell's `❯` prompt or echoed launch command is not enough, even
+with an old Claude header or input frame above it and Claude already running. A header alone or a partially drawn
+input frame also keeps the check pending. If the current screen or process cannot be read, or the process identity is unconfirmed, the handover keeps
+waiting within the readiness window and fails if that window expires. Old screen text or the `claude` process label
+alone cannot complete this check. A healthy agent can therefore remain unconfirmed while those observations are
+unavailable; inspect its pane before deciding how to recover.
+
 **Claude submission check.** For each startup message to a Claude seat, OpenRig makes a bounded check that the prompt
 was submitted. When the capture can't tell, it records the submission as unverified and carries on. When it sees the
 same prompt still sitting in the input, it presses Enter once more; if it's still there, it warns "Startup prompt still
@@ -553,6 +562,10 @@ remain shared.
 - Selected skills are projected into the seat's own agent directory under OpenRig's state, not into a shared
   workspace folder
 
+For Pi tool-selection diagnosis and deliberate in-place recovery, see
+[Pi tool-selection recovery](pi-tool-selection-recovery.md). It distinguishes
+configured, currently intended, registered and active tools, including the TUI/RPC boundary.
+
 **Terminal seats** (`builtin:terminal`) get no agent or profile layers and no startup proof.
 
 ### Claude compaction preparation directory

docs/reference/agent-state-taxonomy.md

@@ -102,7 +102,7 @@ time-bounded evidence. Rungs, top down, with each rung's retirement condition:
 - **r3 — Claude `sessions/<pid>.json` self-report** (busy/shell/idle/waiting, since
   v2.1.139): standing, above hooks for working/idle. Undocumented internal — unreadable
   means fall down the ladder, never error.
-- **r2 — lifecycle hooks** (Claude SessionStart/UserPromptSubmit/Stop/Notification;
+- **r2 — lifecycle hooks** (Claude SessionStart/UserPromptSubmit/PostToolBatch/Stop/StopFailure/Notification;
   Codex SessionStart/UserPromptSubmit/Stop/PermissionRequest): standing for Claude; Codex's
   hook rung starts at trial and is promoted after 50 agreeing observations over at least an
   hour. Exactly-once turn boundaries; SubagentStop filtered (it
@@ -139,6 +139,26 @@ time-bounded evidence. Rungs, top down, with each rung's retirement condition:
   idle. Silence never clears a pending attention prompt, and a runner that states no interval (omp, an older pi
   runner) keeps the hold-until-the-next-boundary behaviour.
 
+Claude permission notifications raise needs-input immediately. A visible permission can clear
+after it has been observed and then disappears. A notification whose dialog was never observed
+can also clear when Claude reports `PostToolBatch`: every tool in that actor's batch has resolved,
+so its earlier permission requests are no longer pending. The relay carries the native actor
+identity; a subagent or in-process teammate finishing its batch cannot clear the main session's
+request, and overlapping actors remain separate. An individual `PostToolUse` is insufficient
+because another tool in the batch may still be waiting for permission.
+
+Batch completion must come from the current occupant and have a later source timestamp than
+the matching notification. Missing actor attribution from an older relay, equal or older
+timestamps, and another actor's completion retain attention. An unrecognized or missing screen
+does not expire attention by age. `Stop` and `UserPromptSubmit` remain the turn-boundary fallback.
+The oracle remembers each actor's latest completed batch for the current occupant, so a strictly
+earlier notification arriving late cannot restore the mark, even if the completion arrived first.
+A notification with an equal or later timestamp remains eligible; swapping or retiring the
+occupant discards its remembered completions.
+Completing a batch changes the needs-input signal without inferring that the seat is idle or
+overwriting a later elicitation. The hook contract is documented in
+[Claude's hooks reference](https://code.claude.com/docs/en/hooks#posttoolbatch).
+
 Partial-coverage honesty: a source with partial lifecycle coverage gets identity-only
 trust, unless its inventory declares it positive-only. A positive-only source can report
 working and nothing else; its silence reads unknown, never idle. The ladder is also the migration path — new oracles slot in above existing

docs/reference/author-contracts.md

@@ -0,0 +1,74 @@
+# Integration author contracts — DRAFT, published with 0.7.6
+
+OpenRig composes agents into a team. An integration author supplies the pieces that let a new harness participate, or let an existing harness run in a new environment. This guide divides that work into two contracts:
+
+1. **[Harness and runner](harness-runner-contract.md):** launch the agent, report its conversation identity and activity, deliver input, handle permission requests and cancellation, and reopen the requested conversation.
+2. **[Execution and platform](execution-platform-contract.md):** choose placement, deliver files and credential references, distinguish owned from attached resources, preserve native state, return artifacts and reconcile shutdown.
+
+These are draft author responsibilities. Some use implemented OpenRig interfaces; others describe a proposed integration whose code remains with its author. Including the guide in release documentation does not make every described capability available.
+
+## Release and support status
+
+This draft was published with **OpenRig 0.7.6** (tag [`v0.7.6`](https://github.com/mvschwarz/openrig/tree/v0.7.6), source `c00c6321`). Its implemented-interface baseline is commit [`b6bd4633`](https://github.com/mvschwarz/openrig/tree/b6bd4633fbd94634a2b4bdef2574a6880f97be63). That base includes the runner declaration and `passes_model` changes. The published 0.7.6 adds release repairs and documentation on top of that base. References below use the immutable base instead of a moving branch.
+
+| Area | Status at this baseline | What the status means |
+|---|---|---|
+| Community launcher presets, RigSpec members and AgentSpec role/startup declarations | Implemented interfaces | They compose configuration. A launcher must implement its declared effects; successful parsing does not prove delivery or native behavior. |
+| `runner: true` with `launch_result: true` | Released in 0.7.6 | The declaration is retained at launch, and Core uses the runner identity/activity path. It does not supply a runner executable. |
+| `passes_model: true` | Released in 0.7.6 | Passes the declared model through `OPENRIG_MODEL`. It does not observe or verify the effective native model. Cursor and Hermes examples leave it off pending native qualification. |
+| The dsh/ACP input, cancellation and answered permission recipe (a person answers, and the default is permit) | Proposed author implementation | No end-to-end dsh integration is established. Native reopen capabilities and the tools-off case remain implementer questions. |
+| Exact native resume via an author launcher | Existing interface; integration-specific evidence | A requested conversation must reopen exactly or fail honestly. One provider/harness result does not establish portability to another. |
+| Environment preparation, projection, artifact export and owned-resource cleanup | Measured in a private wrapper; generic seam proposed | The platform chapter makes the responsibilities reusable. Core does not provide a generic platform driver or `DeploySpec` loader at this baseline. |
+| `HarnessSpec`, `DeploySpec`, generic lifecycle fields and wake-on-demand declarations | Proposed vocabulary or unsupported configuration | They are not new accepted YAML keys. Use existing composition and an explicit author implementation. |
+| Cross-host native resume, provider-wide cancellation or permission support | Unresolved unless qualified by an integration | Unsupported and untested are distinct from observed failure. Publish the actual boundary and version. |
+
+## How the pieces compose
+
+```mermaid
+flowchart TD
+  R["RigSpec: members, relationships, work arrangement"] --> A["AgentSpec: role, context and startup declarations"]
+  R --> H["Built-in runtime or community preset"]
+  H --> N["Harness / runner: native conversation and control"]
+  H --> P["Author platform integration: placement and resources"]
+  P --> N
+  P --> F["Workspace, retained state and returned artifacts"]
+  R --> S["Optional application services"]
+```
+
+**RigSpec remains the composer.** AgentSpec describes reusable agent intent; the runtime or preset selects how the native agent starts. An integration can surround that launch with an environment-specific wrapper. A bundled application's database or Compose service is software the team operates; it is distinct from the environment in which an agent executes.
+
+The terminal provider, native harness protocol and network connection also have different jobs. A terminal provider carries input and output; a runner understands native turn boundaries; a granted MCP or other connection exposes coordination operations. An HTTPS connection does not by itself provision compute, awaken a suspended application or establish native resume. Each integration states which of those effects it implements.
+
+| Owner | Responsibility |
+|---|---|
+| OpenRig Core | Team and work identity, queue records, existing launcher contracts, activity interpretation and declared configuration. |
+| Harness/runner author | Native launch and protocol, exact native reference, prompt framing, truthful activity, permission/cancel behavior and native resume. |
+| Platform integration author | Placement binding, destination projection, resource ownership, state transfer, artifact delivery and resource reconciliation. |
+| Operator / credential owner | Selected environments, grants, provider credentials and spending authority. |
+
+The queue records who owns work. It does not prove that an old provider process has stopped, that a new runtime consumed its history, or that guest bytes reached the consumer. Those observations belong in the launch, resume and export receipts alongside the work record. This division reuses the queue rather than creating another task system.
+
+## Start from an existing example
+
+For a terminal harness, begin with the pinned [community preset reference](https://github.com/mvschwarz/openrig/blob/b6bd4633fbd94634a2b4bdef2574a6880f97be63/docs/reference/community-harness-presets.md) and the [Cursor](https://github.com/mvschwarz/openrig/tree/b6bd4633fbd94634a2b4bdef2574a6880f97be63/docs/reference/community-harness-examples/cursor) or [Hermes](https://github.com/mvschwarz/openrig/tree/b6bd4633fbd94634a2b4bdef2574a6880f97be63/docs/reference/community-harness-examples/hermes) directory. Their presence is source guidance, not an effective-model or complete integration guarantee.
+
+For a structured runner, the [runner chapter](harness-runner-contract.md#2-declare-it-what-the-preset-says-fixed-at-launch) supplies a minimal preset, launch record and activity message. The executable named by the preset must still be implemented and installed by its author.
+
+For a contained worker, the [platform composition](execution-platform-contract.md#a-minimal-composition-using-existing-surfaces) pairs an ordinary coordinator with a worker preset. It uses existing member and preset fields. The wrapper must prepare or attach to the selected environment explicitly; declaring a runtime does not provision it. Its illustrated member list is a fragment, and its launcher is a skeleton, not a ready-to-run provider package.
+
+A rig bundle can carry declarations, roles and supporting scripts. Do not confuse those files with hydrated image layers, installed binaries, registered presets, credentials or live sessions. State what the recipient must supply.
+
+## Lifetime and recovery
+
+**Persistent** and **Non-persistent** describe execution lifetime. A non-persistent invocation has an explicit completion condition tied to its assigned work. **Durable** describes interruption recovery and is independent of that lifetime. An agent can run briefly and still resume its retained conversation later.
+
+Retaining a workspace and reconstruction material can restore useful competence without restoring the same native conversation. Label that as reconstruction. Native resume additionally requires the correct harness-qualified reference and supported native state. The platform chapter's measured example restored copied history into recreated containers on the same host; it does not establish new-host migration or RAM restoration.
+
+## Questions contributors can help resolve
+
+- At a pinned dsh version, which advertised method reopens a session, and can a useful first case run without permission requests?
+- Which additional harness would justify a shared protocol runner rather than a separate wrapper?
+- Which provider effects can use the existing projection, native-reference and lifecycle dispatch, and which need a new maintained interface?
+- What exact state and ownership must travel when compute is recreated or moved?
+
+These questions remain part of the draft. An unanswered contributor thread does not prevent authors from using the implemented interfaces or reviewing the proposed responsibilities. Compatibility claims should name versions, observed behavior and unsupported operations; they should not imply a support guarantee from a manifest alone.

docs/reference/community-harness-examples/cursor/cursor.yaml

@@ -11,3 +11,4 @@ launch_result: true
 resume: true
 startup_assisted: false
 first_input: true
+passes_model: false              # the launcher maps OPENRIG_MODEL for a new session; set true once checked with the real harness

docs/reference/community-harness-examples/cursor/openrig-cursor-launch

@@ -34,6 +34,12 @@ chat_store() {
   printf '%s/chats/%s/%s/store.db' "$CURSOR_CONFIG_DIR" "$sum" "$1"
 }
 
+# The seat's declared model (OPENRIG_MODEL, when the preset sets passes_model) goes to a new chat only. A resumed chat isn't
+# passed one: whether Cursor's --resume keeps the chat's earlier model or takes a new one hasn't been checked.
+if [ -n "${OPENRIG_MODEL:-}" ] && [ -z "${OPENRIG_RESUME_REF:-}" ]; then
+  set -- --model "$OPENRIG_MODEL" "$@"
+fi
+
 # Preserve the file's exact bytes, including trailing newlines. A read error stops before the harness starts.
 if [ -n "${OPENRIG_FIRST_INPUT:-}" ]; then
   first_input=$(cat "$OPENRIG_FIRST_INPUT" && printf '.')

docs/reference/community-harness-examples/dsh/README.md

@@ -0,0 +1,82 @@
+# An example runner for DeepSeek Harness (dsh)
+
+`openrig-dsh-runner.mjs` is a small runner for the draft [harness and runner contract](../../harness-runner-contract.md).
+It runs in the seat's pane, hosts `dsh --profile acp` over ACP on stdio, and does the runner's part of each contract
+section:
+- **identity:** the launch record, written once the conversation is open;
+- **one prompt per block:** each submitted block becomes one `session/prompt`, using bracketed paste in a TTY;
+- **turn boundaries:** reported as activity hooks;
+- **cancel:** `/abort` or Ctrl-C sends `session/cancel`, and a turn reads cancelled only when dsh says so;
+- **permissions:** each request goes to a person through OpenRig;
+- **resume:** it reopens the exact conversation, or says honestly that it couldn't.
+
+It has no dependencies, and it isn't installed by OpenRig. Copy it with [`dsh.yaml`](dsh.yaml) and maintain both as your
+own. [bpamiri/dsh-seat](https://github.com/bpamiri/dsh-seat) is a fuller dsh runner by its own maintainer.
+
+## Set it up
+
+1. **Install dsh and Node.** Install Node 22 or later, then `npm i -g @deepseek-ai/[email protected]`. This is the version
+   it was written for, and it has been checked only against a stand-in agent.
+2. **Install the runner and preset.** Make the runner executable and put it on the seat's `PATH`. Copy `dsh.yaml` to
+   `$OPENRIG_HOME/community-harnesses/`. Then declare a member with `runtime: dsh`.
+3. **Give the seat credentials.** Each seat gets its own dsh home, at `$OPENRIG_HOME/state/dsh/<node>/home`, because dsh
+   rewrites its profile there on every start. Put a `.credentials.yaml` there, or put a provider key in the seat's
+   `env:`. To use a home you prepared yourself, set `OPENRIG_DSH_HOME`. `OPENRIG_DSH` names another `dsh` binary.
+
+The runner's own defaults for dsh:
+- **Uploads:** the seat home gets a `cordis.patch.yml` that turns off dsh's session-log and plugin-list uploads to
+  DeepSeek, and the runner starts dsh with `DSH_TELEMETRY_DISABLED=1`.
+- **Permission mode:** `DSH_PERMISSION_MODE` is `workspace-write` unless you set it.
+- **The hook token:** OpenRig's activity token isn't passed to dsh, whose shell tool inherits the runner's environment.
+
+## Permissions
+
+The runner submits each request to OpenRig, which asks one person:
+- the default person;
+- or the person chosen for the seat with `rig seat set-permissions-to <seat> --to <person> --reason "…"`.
+
+That person answers one of the options dsh offered, and the runner gives that answer to the same live request, once.
+
+**When no one can be asked, the request is permitted** with dsh's own allow option: `allow_once` when dsh offers it,
+else `allow_always`. That's OpenRig's default. The runner does the same when OpenRig can't answer: an older OpenRig
+without this path, an OpenRig it can't reach, or a refusal other than `--deny`.
+
+**`allow_always` creates a standing grant in dsh.** OpenRig doesn't record or revoke it, which is why the runner prefers
+`allow_once`. Where dsh keeps the grant, and for how long, isn't checked.
+
+**The runner refuses** only when the seat's operator opted in to refusal, with `rig seat set-permissions-to <seat>
+--deny`, or when dsh offered no allow option. A refused request is shown in the pane, and the seat reads needs input
+until the next prompt.
+
+**The runner answers cancelled, holding nothing,** when the person says no and dsh offered no reject option, or when
+OpenRig says the request isn't the seat's current occupant's (for example, after a relaunch).
+
+## Before you declare resume
+
+The preset leaves `resume` out. The runner reopens a conversation by the method dsh advertises: `session/resume`, or
+`session/load`. When dsh advertises neither, it tries `session/resume`. It never starts a new conversation on a resume.
+
+**Telling a missing conversation from one that won't reopen.** dsh `0.2.1-rc.2` answers `-32602` both for a
+conversation it doesn't have and for one already active elsewhere, so the runner never reads that error as missing.
+- **Only dsh's own not-found error (`-32002`) counts as missing.** Any other failure leaves the resume unconfirmed.
+- **When dsh advertises `session/list`** (it does), the runner first asks the list, following its pages. A conversation
+  the list names but that won't reopen stays unconfirmed, even on a not-found.
+- **An unlisted conversation is still reopened.** dsh's list can leave out a conversation another running dsh holds
+  unsaved, so it can't show a conversation is gone.
+
+**Limit:** dsh answers `-32602`, not `-32002`, for a conversation it doesn't have, so the runner can't report a missing
+dsh conversation as missing. It reads unconfirmed.
+
+Add `resume: true` after a real dsh has reopened a conversation and recalled what was said in it.
+
+## Check it without a model
+
+Point dsh's `DEEPSEEK_BASE_URL` at a local server that speaks DeepSeek's Anthropic-compatible Messages API. Then run the
+contract's real check:
+- a seat reads ready only after the handshake;
+- one multi-line message becomes one prompt;
+- the seat reads working, then idle;
+- a refused permission reads needs input;
+- `/abort` ends a turn as cancelled.
+
+`packages/daemon/test/community-harness-dsh-example.test.ts` runs the same cases against a stand-in agent.

docs/reference/community-harness-examples/dsh/dsh.yaml

@@ -0,0 +1,20 @@
+# Install as $OPENRIG_HOME/community-harnesses/dsh.yaml, with openrig-dsh-runner.mjs (executable) on the seat's PATH.
+format: 1
+name: dsh
+command: openrig-dsh-runner.mjs
+maintainer: Your name or team
+runner: true
+launch_result: true
+first_input: true
+limits:
+  - written for dsh 0.2.1-rc.2 (npm @deepseek-ai/dsh) over ACP protocol 1, with Node 22 or later, and checked against a stand-in agent only
+  - each seat gets its own dsh home under $OPENRIG_HOME/state/dsh/<node>/home, holding a patch that turns off dsh's session-log and plugin-list uploads to DeepSeek; give it credentials (.credentials.yaml there) or a provider key in the seat's env
+  - DSH_PERMISSION_MODE defaults to workspace-write; a permission request goes to the default person or the seat's override (rig seat set-permissions-to), is permitted with dsh's allow_once (else allow_always) when no one can be asked or OpenRig can't answer, and is refused and held as needs input only after --deny
+  - allow_always creates a standing grant in dsh, which OpenRig doesn't record or revoke; the runner prefers allow_once, and where dsh keeps the grant and for how long isn't checked
+  - resume isn't declared until a real dsh has reopened a conversation and recalled it; the runner reopens by the method dsh advertises and reports missing only on dsh's own not-found error, not when dsh's session list names the conversation; dsh answers -32602 for a missing conversation (and for one active elsewhere), so a missing dsh conversation reads unconfirmed
+  - a prompt whose whole text is /abort can't be sent; a Ctrl-C byte inside a paste also aborts; without a TTY, input is read line by line; this isn't a binary-safe channel
+  - one prompt at a time; a message that arrives mid-turn waits for the next turn
+  - the seat's model isn't passed; dsh uses its own configured model
+  - no guidance, skills or startup files are placed; dsh reads its own AGENTS.md and skill folders
+  - when the pane closes, the runner cancels a running turn and stops dsh; anything dsh's tools started may keep running
+  - fork unsupported

docs/reference/community-harness-examples/dsh/openrig-dsh-runner.mjs

@@ -0,0 +1,565 @@
+#!/usr/bin/env node
+// Example OpenRig runner for DeepSeek Harness (dsh), not installed by OpenRig. Checked against a stand-in agent only;
+// written for dsh 0.2.1-rc.2 (npm @deepseek-ai/dsh), ACP protocol 1 and Node 22 or later. No dependencies.
+//
+// It follows docs/reference/harness-runner-contract.md: it hosts `dsh --profile acp` over ACP on stdio, writes the
+// launch record once the conversation is open, turns each submitted block into one prompt, reports turn boundaries as
+// activity hooks, cancels on /abort or Ctrl-C, and sends permission requests to OpenRig: a person answers, or OpenRig
+// permits when no one can be asked, and so does the runner when OpenRig can't answer (refusing and holding needs input
+// only when the seat opted in to refusal). Copy it and maintain it as your own.
+
+import { spawn } from "node:child_process";
+import { existsSync, mkdirSync, readFileSync, realpathSync, renameSync, writeFileSync } from "node:fs";
+import { homedir } from "node:os";
+import { join } from "node:path";
+import { createInterface } from "node:readline";
+import { pathToFileURL } from "node:url";
+
+const RUNNER = "openrig-dsh-runner 1";
+const NAME = process.env.OPENRIG_RUNTIME || "dsh"; // the preset's name: OpenRig passes it, and checks the record against it
+const LIVENESS_SECONDS = 60;
+const BEAT_MS = 30_000;
+const ACTIVE_EVERY_MS = 15_000;
+const POST_TIMEOUT_MS = 2_000;
+const WAIT_SECONDS = 25;
+const READ_TRIES = 3; // reads in a row that get no answer before the runner permits
+// Refusals that say the request certainly isn't the seat's current occupant's live one: cancel it, and hold nothing.
+const NOT_LIVE = ["stale_generation", "activity_grant_wrong_seat", "activity_grant_stale"];
+// A read's states that end the request, answered cancelled (an `answered` that reaches there names an option dsh didn't
+// offer). Any other state, like an unlisted status, permits.
+const ENDED = ["refused", "withdrawn", "stale", "answered"];
+const CANCELLED = { outcome: { outcome: "cancelled" } };
+const ACP_RESOURCE_NOT_FOUND = -32002;
+// The most pages of dsh's session list read before a resume; a longer list decides nothing.
+const LIST_PAGES = 50;
+
+const PASTE_START = "\x1b[200~";
+const PASTE_END = "\x1b[201~";
+
+/** Turns terminal bytes into submitted blocks. Everything up to an Enter outside a paste is one block, line breaks
+ *  included; a block that is exactly /abort, or a Ctrl-C byte, is the only control. */
+export class InputFramer {
+  #buffer = "";
+  #pasting = false;
+  #partial = "";
+
+  push(chunk) {
+    const events = [];
+    const data = this.#partial + chunk;
+    this.#partial = "";
+    let i = 0;
+    while (i < data.length) {
+      if (data[i] === "\x1b") {
+        const rest = data.slice(i);
+        if (rest.startsWith(PASTE_START) || rest.startsWith(PASTE_END)) {
+          this.#pasting = rest.startsWith(PASTE_START);
+          i += PASTE_START.length;
+          continue;
+        }
+        if (PASTE_START.startsWith(rest) || PASTE_END.startsWith(rest)) { this.#partial = rest; break; }
+        if (!this.#pasting) { i += escapeLength(rest); continue; } // arrow keys and the like
+      }
+      const ch = data[i];
+      i += 1;
+      if (ch === "\x03") { this.#buffer = ""; this.#pasting = false; events.push({ type: "abort" }); continue; }
+      if (this.#pasting) {
+        if (ch === "\r") { if (data[i] === "\n") i += 1; this.#buffer += "\n"; } else this.#buffer += ch;
+        continue;
+      }
+      if (ch === "\r" || ch === "\n") {
+        const text = this.#buffer;
+        this.#buffer = "";
+        if (text.trim() === "/abort") events.push({ type: "abort" });
+        else if (text.trim()) events.push({ type: "block", text });
+        continue;
+      }
+      if (ch === "\x7f" || ch === "\b") {
+        if (this.#buffer) { this.#buffer = this.#buffer.slice(0, -1); events.push({ type: "erase" }); }
+        continue;
+      }
+      if (ch >= " ") { this.#buffer += ch; events.push({ type: "echo", text: ch }); }
+    }
+    return events;
+  }
+}
+
+function escapeLength(rest) {
+  if (rest[1] === "[") {
+    let j = 2;
+    while (j < rest.length && !(rest.charCodeAt(j) >= 0x40 && rest.charCodeAt(j) <= 0x7e)) j += 1;
+    return Math.min(j + 1, rest.length);
+  }
+  return rest[1] === "O" ? Math.min(3, rest.length) : 1;
+}
+
+/** JSON-RPC 2.0 with the agent, one message per line on its stdio. */
+class AgentPeer {
+  #child;
+  #next = 1;
+  #waiting = new Map();
+  #requests = new Map();
+  #notifications = new Map();
+
+  constructor(child, onNoise) {
+    this.#child = child;
+    createInterface({ input: child.stdout }).on("line", (line) => {
+      let message;
+      try { message = JSON.parse(line); } catch { message = null; }
+      if (!message || typeof message !== "object") { if (line.trim()) onNoise(line); return; }
+      if (message.method === undefined) {
+        const waiter = this.#waiting.get(message.id);
+        if (!waiter) return;
+        this.#waiting.delete(message.id);
+        if (message.error) waiter.reject(Object.assign(new Error(message.error.message ?? "error"), { code: message.error.code }));
+        else waiter.resolve(message.result);
+        return;
+      }
+      if (message.id === undefined) { this.#notifications.get(message.method)?.(message.params ?? {}); return; }
+      const handler = this.#requests.get(message.method);
+      if (!handler) { this.#write({ id: message.id, error: { code: -32601, message: `Method not found: ${message.method}` } }); return; }
+      Promise.resolve().then(() => handler(message.params ?? {}, message.id)).then(
+        (result) => this.#write({ id: message.id, result }),
+        (error) => this.#write({ id: message.id, error: { code: error.code ?? -32603, message: error.message } }));
+    });
+  }
+
+  request(method, params) {
+    const id = this.#next++;
+    return new Promise((resolve, reject) => { this.#waiting.set(id, { resolve, reject }); this.#write({ id, method, params }); });
+  }
+
+  notify(method, params) { this.#write({ method, params }); }
+  onRequest(method, handler) { this.#requests.set(method, handler); }
+  onNotification(method, handler) { this.#notifications.set(method, handler); }
+
+  failAll(reason) {
+    for (const waiter of this.#waiting.values()) waiter.reject(new Error(reason));
+    this.#waiting.clear();
+  }
+
+  #write(message) {
+    if (this.#child.stdin.writable) this.#child.stdin.write(`${JSON.stringify({ jsonrpc: "2.0", ...message })}\n`);
+  }
+}
+
+const openrigHome = () => process.env.OPENRIG_HOME || join(homedir(), ".openrig");
+
+/** The daemon's base URL and the seat's token: from the environment, else from activity-endpoint.json. */
+function openrigEndpoint() {
+  const env = process.env;
+  let base = env.OPENRIG_URL || (env.OPENRIG_PORT ? `http://${env.OPENRIG_HOST || "127.0.0.1"}:${env.OPENRIG_PORT}` : "");
+  let token = env.OPENRIG_ACTIVITY_HOOK_TOKEN || "";
+  if (!base || !token) {
+    try {
+      const file = JSON.parse(readFileSync(join(openrigHome(), "activity-endpoint.json"), "utf8"));
+      base ||= file.baseUrl ?? "";
+      token ||= file.token ?? "";
+    } catch { /* no daemon has written one yet */ }
+  }
+  return base && token ? { base: base.replace(/\/+$/, ""), token } : null;
+}
+
+async function callOpenRig(method, path, body, timeoutMs) {
+  const endpoint = openrigEndpoint();
+  if (!endpoint) return { status: 0, body: null };
+  try {
+    const response = await fetch(`${endpoint.base}${path}`, {
+      method, signal: AbortSignal.timeout(timeoutMs),
+      headers: { authorization: `Bearer ${endpoint.token}`, ...(body ? { "content-type": "application/json" } : {}) },
+      ...(body ? { body: JSON.stringify(body) } : {}),
+    });
+    let parsed = null;
+    try { parsed = await response.json(); } catch { /* not JSON */ }
+    return { status: response.status, body: parsed };
+  } catch {
+    return { status: 0, body: null };
+  }
+}
+
+/** Activity posts go one at a time, in order, and never hold up the agent. */
+class Activity {
+  #sequence = 0;
+  #chain = Promise.resolve();
+
+  constructor(identity) { this.identity = identity; }
+
+  post(hookEvent, subtype, extra = {}) {
+    const body = { ...this.identity, sequence: ++this.#sequence, occurredAt: new Date().toISOString(), hookEvent, subtype, ...extra };
+    this.#chain = this.#chain.then(() => callOpenRig("POST", "/api/activity/hooks", body, POST_TIMEOUT_MS)).then(() => {}, () => {});
+  }
+
+  flushed() { return this.#chain; }
+}
+
+const UPLOADS_OFF = [
+  "# Written by openrig-dsh-runner for this seat; safe to edit, and an existing file is left alone.",
+  "# Don't attach this seat's session log, or its plugin package names and versions, to DeepSeek API requests.",
+  "- id: session-log-deepseek",
+  "  config:",
+  "    enabled: false",
+  "- id: plugin-package-inventory-deepseek",
+  "  config:",
+  "    enabled: false",
+  "",
+].join("\n");
+
+/** Each seat gets its own writable dsh home (dsh rewrites its profile there on every start). OPENRIG_DSH_HOME names
+ *  one you prepared yourself, which is used as it is. */
+function seatHome() {
+  if (process.env.OPENRIG_DSH_HOME) return process.env.OPENRIG_DSH_HOME;
+  const seat = process.env.OPENRIG_NODE_ID || process.env.OPENRIG_SESSION_NAME || "default";
+  const home = join(openrigHome(), "state", "dsh", seat, "home");
+  mkdirSync(home, { recursive: true, mode: 0o700 });
+  const patch = join(home, "cordis.patch.yml");
+  if (!existsSync(patch)) writeFileSync(patch, UPLOADS_OFF, { mode: 0o600 });
+  return home;
+}
+
+function writeLaunchRecord(fields) {
+  const path = process.env.OPENRIG_LAUNCH_RESULT;
+  if (!path) return;
+  const record = { format: 1, attempt: process.env.OPENRIG_LAUNCH_ATTEMPT, harness: NAME, ...fields };
+  const temp = `${path}.${process.pid}.tmp`;
+  writeFileSync(temp, `${JSON.stringify(record)}\n`, { mode: 0o600 });
+  renameSync(temp, path);
+}
+
+const sleep = (ms) => new Promise((done) => setTimeout(done, ms));
+const say = (line) => process.stdout.write(`${line}\n`);
+const firstLine = (text) => {
+  const lines = text.split("\n");
+  return lines.length > 1 ? `${lines[0]} … (+${lines.length - 1} lines)` : lines[0];
+};
+
+async function main() {
+  const env = process.env;
+  const cwd = process.cwd();
+  let firstInput = null;
+  if (env.OPENRIG_FIRST_INPUT) {
+    try { firstInput = readFileSync(env.OPENRIG_FIRST_INPUT, "utf8"); } catch (error) {
+      console.error(`openrig-dsh-runner: the first startup turn couldn't be read (${error.message}); not starting`);
+      process.exit(2);
+    }
+  }
+
+  const home = seatHome();
+  const activity = new Activity({ sessionName: env.OPENRIG_SESSION_NAME, nodeId: env.OPENRIG_NODE_ID,
+    runtime: NAME, generation: env.OPENRIG_OCCUPANT_GENERATION });
+  const agentEnv = { ...env, DSH_HOME: home };
+  delete agentEnv.OPENRIG_ACTIVITY_HOOK_TOKEN; // dsh's shell tool inherits this environment
+  agentEnv.DSH_PERMISSION_MODE ??= "workspace-write";
+  if (agentEnv.DSH_TELEMETRY_DISABLED === undefined && agentEnv.DSH_TELEMETRY_MODE === undefined) agentEnv.DSH_TELEMETRY_DISABLED = "1";
+
+  const child = spawn(env.OPENRIG_DSH || "dsh", ["--profile", "acp"], { cwd, env: agentEnv, stdio: ["pipe", "pipe", "inherit"] });
+  child.stdin.on("error", () => {});
+  const peer = new AgentPeer(child, (line) => say(`[dsh] ${line}`));
+  const state = { sessionId: null, turn: 0, running: false, cancelRequestedAt: null, holding: false, refused: false, queue: [], asks: new Set(), lastActive: 0 };
+  const inflight = new Set();
+  const track = (promise) => { inflight.add(promise); promise.finally(() => inflight.delete(promise)); return promise; };
+  let agentExited = false;
+  let stopping = false;
+  let heartbeat;
+  let tty = false;
+
+  const restoreTty = () => {
+    if (!tty) return;
+    tty = false;
+    process.stdout.write("\x1b[?2004l");
+    try { process.stdin.setRawMode(false); } catch { /* already closed */ }
+  };
+  const finish = async (code) => {
+    clearInterval(heartbeat);
+    restoreTty();
+    await Promise.race([Promise.allSettled([...inflight, activity.flushed()]), sleep(2_000)]);
+    process.exit(code);
+  };
+  const fatal = async (message, code) => {
+    stopping = true;
+    say(`[openrig-dsh] ${message}`);
+    if (!agentExited) {
+      child.kill("SIGTERM");
+      await Promise.race([new Promise((done) => child.once("exit", done)), sleep(3_000)]);
+    }
+    await finish(code);
+  };
+  const postIdle = (subtype) => activity.post("Stop", subtype, { livenessSeconds: LIVENESS_SECONDS });
+  const withdraw = (askId, reason) => track(callOpenRig("POST", `/api/activity/permission-requests/${encodeURIComponent(askId)}/withdraw`, { reason }, POST_TIMEOUT_MS));
+
+  child.once("error", (error) => { agentExited = true; void fatal(`dsh couldn't be started (${error.message}); is it on PATH?`, 1); });
+  child.on("exit", (code, signal) => {
+    agentExited = true;
+    peer.failAll("dsh exited");
+    if (stopping) return;
+    for (const ask of state.asks) ask.end("exited");
+    say(`\n[openrig-dsh] dsh exited (code ${code ?? "none"}, signal ${signal ?? "none"})`);
+    activity.post("Stop", `${NAME}_exited`);
+    void finish(code === 0 ? 0 : 1);
+  });
+
+  peer.onNotification("session/update", (params) => {
+    if (params.sessionId !== state.sessionId || !params.update) return;
+    const update = params.update;
+    const reporting = state.running && !state.holding; // while needs input is held, nothing may read working
+    if (update.sessionUpdate === "agent_message_chunk") {
+      if (update.content?.type === "text") process.stdout.write(update.content.text);
+      if (reporting && Date.now() - state.lastActive >= ACTIVE_EVERY_MS) { state.lastActive = Date.now(); activity.post("active", "agent_message"); }
+    } else if (update.sessionUpdate === "tool_call") {
+      say(`\n[tool] ${update.title ?? "tool"} (${update.kind ?? "other"})`);
+      if (reporting) activity.post("PreToolUse", String(update.kind ?? "other"));
+    }
+  });
+
+  const refuse = (why) => {
+    state.holding = true;
+    state.refused = true; // the seat's own refusal holds needs input until the next prompt
+    say(`[permission] refused: ${why}. The seat reads needs input until the next prompt.`);
+    activity.post("Notification", "permission_prompt", { turnId: String(state.turn) });
+    return CANCELLED;
+  };
+
+  // The answered path: OpenRig opens one ask to the default person (or the seat's override, `rig seat
+  // set-permissions-to`), and the runner reads the answer back for this same live request. With no one to ask, OpenRig
+  // permits at once, and so does the runner when OpenRig can't answer. Only the seat's opt-in (`--deny`) refuses; a
+  // request OpenRig says isn't the current occupant's, or that a person said no to with no reject option, is cancelled.
+  const askOwner = async (requestId, toolCall, options) => {
+    const turn = state.turn;
+    const live = () => state.running && state.turn === turn && !state.cancelRequestedAt && !agentExited;
+    const submitted = await callOpenRig("POST", "/api/activity/permission-requests", {
+      sessionName: env.OPENRIG_SESSION_NAME, nodeId: env.OPENRIG_NODE_ID, generation: env.OPENRIG_OCCUPANT_GENERATION,
+      nativeRef: state.sessionId, turnId: String(turn), requestId: String(requestId), toolCall,
+      options: options.map((option) => ({ optionId: option.optionId, name: option.name ?? option.optionId, kind: option.kind })),
+      occurredAt: new Date().toISOString(),
+    }, 10_000);
+    const askId = submitted.body?.ok ? submitted.body.askId : undefined;
+    if (!live()) {
+      if (askId) withdraw(askId, state.cancelRequestedAt ? "turn_cancelled" : agentExited ? "exited" : "turn_ended");
+      return CANCELLED;
+    }
+    const offered = (optionId) => options.some((option) => option.optionId === optionId);
+    // No one can be asked: OpenRig's default is permit, with the agent's own allow option.
+    if (submitted.body?.ok && submitted.body.state === "permitted" && offered(submitted.body.optionId)) {
+      say(`[permission] permitted: ${submitted.body.reason ?? "no one can be asked"} (${submitted.body.optionId})`);
+      return { outcome: { outcome: "selected", optionId: submitted.body.optionId } };
+    }
+    // When OpenRig can't answer, nobody can be asked either, so the same default applies. allow_once is preferred, so
+    // dsh doesn't remember a permission nobody chose.
+    const allow = options.find((option) => option.kind === "allow_once") ?? options.find((option) => option.kind === "allow_always");
+    const permit = (why) => {
+      if (!allow) return refuse(`${why}, and the agent offered no allow option`);
+      state.holding = false;
+      say(`[permission] permitted: ${why} (${allow.optionId})`);
+      return { outcome: { outcome: "selected", optionId: allow.optionId } };
+    };
+    if (!askId) {
+      if (submitted.body?.code === "denied_by_seat") return refuse("the seat's operator chose to refuse its permission requests");
+      if (NOT_LIVE.includes(submitted.body?.code)) {
+        say(`[permission] cancelled: OpenRig says this isn't the seat's current request (${submitted.body.code})`);
+        return CANCELLED;
+      }
+      return permit(submitted.status === 404 ? "this OpenRig has no answered path"
+        : submitted.status === 0 ? "OpenRig couldn't be reached" : `OpenRig couldn't ask anyone (${submitted.body?.code ?? submitted.status})`);
+    }
+    state.holding = true; // OpenRig recorded the seat's permission prompt when it opened the ask
+    say(`[permission] OpenRig asked a person (${askId}); waiting for their answer`);
+    return new Promise((resolve) => {
+      let done = false; // the agent has its answer
+      let ended = false; // OpenRig was told the request is over
+      const ask = { end: (reason) => {
+        if (ended) return;
+        ended = true;
+        state.asks.delete(ask);
+        withdraw(askId, reason);
+        if (done) return;
+        done = true;
+        if (!state.refused) state.holding = false; // the turn ended the person's ask: no one is still being asked
+        resolve(CANCELLED);
+      } };
+      state.asks.add(ask);
+      void (async () => {
+        let unanswered = 0;
+        while (!done) {
+          const read = await callOpenRig("GET", `/api/activity/permission-requests/${encodeURIComponent(askId)}?waitSeconds=${WAIT_SECONDS}`, null, (WAIT_SECONDS + 10) * 1_000);
+          if (done) return;
+          if (read.status === 0 && ++unanswered < READ_TRIES) { await sleep(2_000); continue; }
+          if (read.status === 200 && read.body?.state === "pending") { unanswered = 0; continue; }
+          done = true;
+          const answer = read.body ?? {};
+          // A person's option, or OpenRig's permit when they closed it without one.
+          if ((answer.state === "answered" || answer.state === "permitted") && offered(answer.optionId) && live()) {
+            state.asks.delete(ask);
+            state.holding = false;
+            say(`[permission] ${answer.state}: ${answer.optionId}`);
+            resolve({ outcome: { outcome: "selected", optionId: answer.optionId } });
+          } else if (live() && (read.status === 200 ? !ENDED.includes(answer.state) : !NOT_LIVE.includes(answer.code))) {
+            // OpenRig stopped answering, or answered in a way the contract doesn't list: nobody can be asked. The ask stays
+            // tracked, so the turn's end still withdraws it.
+            resolve(permit(read.status === 0 ? "OpenRig couldn't be reached" : `OpenRig couldn't answer (${answer.code ?? read.status})`));
+          } else {
+            state.asks.delete(ask);
+            state.holding = false; // a person said no, or the request is over: no one is still waiting
+            say(`[permission] not answered (${answer.reason ?? answer.state ?? answer.code ?? read.status}); answered cancelled`);
+            resolve(CANCELLED);
+          }
+        }
+      })();
+    });
+  };
+
+  peer.onRequest("session/request_permission", (params, requestId) => {
+    if (params.sessionId !== state.sessionId) throw Object.assign(new Error(`no turn of conversation ${params.sessionId} runs here`), { code: -32602 });
+    const toolCall = { title: String(params.toolCall?.title ?? "a tool call"), kind: String(params.toolCall?.kind ?? "other") };
+    const options = Array.isArray(params.options) ? params.options : [];
+    say(`\n[permission] dsh asks: ${toolCall.title} (${toolCall.kind}); options ${options.map((option) => option.optionId).join(", ")}`);
+    if (!state.running || state.cancelRequestedAt) return CANCELLED;
+    return askOwner(requestId, toolCall, options);
+  });
+
+  const abort = () => {
+    if (!state.running) { say("[openrig-dsh] no turn is running"); return; }
+    if (state.cancelRequestedAt) return;
+    state.cancelRequestedAt = new Date().toISOString();
+    peer.notify("session/cancel", { sessionId: state.sessionId });
+    say(`\n[openrig-dsh] cancel requested at ${state.cancelRequestedAt}; the turn ends when dsh says so`);
+    if (!state.refused) state.holding = false; // the seat's own refusal holds until the next accepted prompt
+    for (const ask of state.asks) ask.end("turn_cancelled");
+  };
+
+  const runTurn = async (text) => {
+    state.running = true;
+    state.turn += 1;
+    state.holding = false;
+    state.refused = false;
+    state.cancelRequestedAt = null;
+    state.lastActive = 0;
+    const turnId = String(state.turn);
+    say(`\nyou ▸ ${firstLine(text)}`);
+    activity.post("UserPromptSubmit", "accepted", { turnId });
+    let stopReason;
+    try {
+      const result = await peer.request("session/prompt", { sessionId: state.sessionId, prompt: [{ type: "text", text }] });
+      stopReason = String(result?.stopReason ?? "unreported");
+    } catch (error) {
+      if (agentExited) return;
+      say(`\n[openrig-dsh] the turn failed: ${error.message}`);
+      activity.post("Notification", "runtime_error", { turnId });
+      stopReason = "error";
+    }
+    state.running = false;
+    for (const ask of state.asks) ask.end("turn_ended");
+    say(`\n── turn end (${stopReason}) ──`);
+    if (state.holding) say("[openrig-dsh] a permission request was refused, so the seat reads needs input until the next prompt");
+    else postIdle(stopReason);
+    pump();
+  };
+
+  const pump = () => {
+    if (state.running || agentExited || state.queue.length === 0) return;
+    void runTurn(state.queue.shift());
+  };
+
+  const submit = (text) => {
+    if (agentExited) return;
+    state.queue.push(text);
+    if (state.running) say(`[queued: ${state.queue.length}] this goes to dsh after the current turn`);
+    pump();
+  };
+
+  // Handshake, then open or reopen the conversation, then report it. Nothing is read as ready before this.
+  let init;
+  try {
+    init = await peer.request("initialize", { protocolVersion: 1, clientCapabilities: {}, clientInfo: { name: "openrig-dsh-runner", version: "1" } });
+  } catch (error) {
+    return fatal(`dsh didn't complete the ACP handshake (${error.message})`, 1);
+  }
+  const capabilities = init?.agentCapabilities ?? {};
+  const producer = `${RUNNER}; dsh ${init?.agentInfo?.version ?? "(version not reported)"}; ACP ${init?.protocolVersion ?? "?"}`;
+  const resumeRef = env.OPENRIG_RESUME_REF;
+  // Whether dsh's own session list names a conversation, following its pages: true, false when every page was read and
+  // none names it, or undefined when the list fails, answers in another shape, or runs past LIST_PAGES pages. An ACP
+  // cursor is an opaque string, possibly empty, passed back unchanged; an absent or null one ends the list.
+  const listsConversation = async (id) => {
+    let cursor;
+    for (let page = 0; page < LIST_PAGES; page += 1) {
+      let listing;
+      try { listing = await peer.request("session/list", cursor === undefined ? {} : { cursor }); } catch { return undefined; }
+      if (!Array.isArray(listing?.sessions) || listing.sessions.some((session) => typeof session?.sessionId !== "string")) return undefined;
+      if (listing.sessions.some((session) => session.sessionId === id)) return true;
+      if (listing.nextCursor === undefined || listing.nextCursor === null) return false;
+      if (typeof listing.nextCursor !== "string") return undefined;
+      cursor = listing.nextCursor;
+    }
+    return undefined;
+  };
+  if (resumeRef) {
+    // The method dsh advertises; with neither advertised, session/resume, the method dsh 0.2.1-rc.2 answered in its
+    // maintainer's checks. dsh 0.2.1-rc.2 answers -32602 both for a conversation it doesn't have and for one active
+    // elsewhere, so that error can't mean missing. Only dsh's own "not found" counts as missing, and not when dsh's list
+    // names the conversation. The list can't show a conversation is gone: one another running dsh holds unsaved may be
+    // left out, so an unlisted conversation is still reopened. Anything else leaves the resume unconfirmed.
+    const method = capabilities.sessionCapabilities?.resume ? "session/resume" : capabilities.loadSession ? "session/load" : "session/resume";
+    const listed = capabilities.sessionCapabilities?.list ? await listsConversation(resumeRef) === true : false;
+    try {
+      await peer.request(method, { sessionId: resumeRef, cwd, mcpServers: [] });
+    } catch (error) {
+      if (error.code === ACP_RESOURCE_NOT_FOUND && !listed) {
+        writeLaunchRecord({ error: "missing", producer });
+        return fatal(`dsh says conversation ${resumeRef} doesn't exist (${error.message}); no new one was started`, 3);
+      }
+      return fatal(`dsh couldn't reopen conversation ${resumeRef} (${method}: ${error.message}); the resume is unconfirmed and no new conversation was started`, 4);
+    }
+    state.sessionId = resumeRef;
+  } else {
+    try {
+      state.sessionId = (await peer.request("session/new", { cwd, mcpServers: [] }))?.sessionId ?? null;
+    } catch (error) {
+      return fatal(`dsh couldn't open a conversation (${error.message})`, 1);
+    }
+    if (!state.sessionId) return fatal("dsh opened a conversation without naming it", 1);
+  }
+  writeLaunchRecord({ nativeRef: state.sessionId, producer });
+  say(`[openrig-dsh] ready · ${producer} · conversation ${state.sessionId}`);
+  say(`[openrig-dsh] DSH_HOME ${home} · DSH_PERMISSION_MODE ${agentEnv.DSH_PERMISSION_MODE}`);
+  say("[openrig-dsh] each submitted message is one prompt; /abort or Ctrl-C cancels the running turn");
+
+  heartbeat = setInterval(() => { if (!state.running && !state.holding && !agentExited) postIdle("idle_heartbeat"); }, BEAT_MS);
+  for (const signal of ["SIGHUP", "SIGTERM"]) {
+    process.on(signal, () => {
+      if (state.running && !state.cancelRequestedAt) peer.notify("session/cancel", { sessionId: state.sessionId });
+      for (const ask of state.asks) ask.end("exited");
+      child.kill("SIGTERM");
+      setTimeout(() => child.kill("SIGKILL"), 3_000).unref();
+    });
+  }
+  process.on("SIGINT", abort);
+
+  const framer = new InputFramer();
+  if (process.stdin.isTTY) {
+    process.stdin.setRawMode(true);
+    process.stdout.write("\x1b[?2004h");
+    tty = true;
+    process.stdin.setEncoding("utf8");
+    process.stdin.on("data", (chunk) => {
+      for (const event of framer.push(chunk)) {
+        if (event.type === "echo") process.stdout.write(event.text);
+        else if (event.type === "erase") process.stdout.write("\b \b");
+        else if (event.type === "abort") { process.stdout.write("\n"); abort(); }
+        else if (event.type === "block") { process.stdout.write("\n"); submit(event.text); }
+      }
+    });
+  } else {
+    createInterface({ input: process.stdin }).on("line", (line) => {
+      if (line.trim() === "/abort") abort();
+      else if (line.trim()) submit(line);
+    });
+  }
+
+  if (firstInput !== null) submit(firstInput);
+  else postIdle("idle_heartbeat");
+}
+
+const invoked = (() => {
+  try { return process.argv[1] ? pathToFileURL(realpathSync(process.argv[1])).href : null; } catch { return null; }
+})();
+if (invoked === import.meta.url) {
+  main().catch((error) => { console.error(`openrig-dsh-runner: ${error.stack ?? error}`); process.exit(1); });
+}

docs/reference/community-harness-examples/hermes/hermes.yaml

@@ -10,3 +10,4 @@ limits:
   - no guidance, skills or startup files are placed
   - needs-input and approval prompts are read from the screen only
 names_conversation: true
+passes_model: false              # the launcher maps OPENRIG_MODEL for a new session; set true once checked with the real harness

docs/reference/community-harness-examples/hermes/openrig-hermes-launch

@@ -24,6 +24,21 @@ if ! grep -qs OPENRIG_SESSION_NAME "$HERMES_HOME/config.yaml"; then
   echo "openrig-hermes-launch: $HERMES_HOME/config.yaml doesn't pass OPENRIG_SESSION_NAME to Hermes's tools, so rig run from Hermes won't act as this seat; see config.yaml.example" >&2
 fi
 
+# The seat's declared model (OPENRIG_MODEL, when the preset sets passes_model): Hermes takes it only from config.yaml, and
+# Hermes v0.22.4 fixes a session's launch settings when it creates it. So it's written for a new session only, and only
+# into a config.yaml that is absent or empty: a seat's own config is left byte-for-byte as it is, whatever it says. On a
+# resume this launcher applies no OpenRig model.
+if [ -n "${OPENRIG_MODEL:-}" ] && [ -z "${OPENRIG_RESUME_REF:-}" ]; then
+  case "$OPENRIG_MODEL" in
+    ''|*[!A-Za-z0-9._:/@+-]*) echo "openrig-hermes-launch: not writing unusual model id \"$OPENRIG_MODEL\" into $HERMES_HOME/config.yaml" >&2 ;;
+    *) if [ -s "$HERMES_HOME/config.yaml" ]; then
+         echo "openrig-hermes-launch: the seat has its own $HERMES_HOME/config.yaml; OPENRIG_MODEL was not applied; set the model there" >&2
+       else
+         printf 'model: "%s"\n' "$OPENRIG_MODEL" >> "$HERMES_HOME/config.yaml"
+       fi ;;
+  esac
+fi
+
 if [ -n "${OPENRIG_RESUME_REF:-}" ]; then
   exec "$hermes" chat -c "$OPENRIG_RESUME_REF"
 fi

docs/reference/community-harness-examples/openshell/README.md

@@ -0,0 +1,5 @@
+# OpenShell reference driver
+
+Run `rig environment setup --help` to install the maintained OpenShell 0.1.6 / Claude Code reference from the packaged daemon assets. Its README, image recipe, endpoint-specific policy, example RigSpec/AgentSpecs and versioned driver schema are installed together. Setup prepares files and an ordinary scoped work grant; it does not create compute or launch a provider.
+
+This first cut supports explicit owned create or exact attached binding, required destination projection and verified JSON artifact return through authenticated MCP work tools. Explicit owned lifecycle operations retain separate observed receipts. Exact same-host recreate/restore requires verified private history and the selected checkpoint; attached native, storage and compute are preserved. A native process receipt is not a successful model-work claim. Existing unbound host runtimes retain their launch/projection paths.

docs/reference/community-harness-examples/pi-tui/openrig-pi-tui-launch

@@ -0,0 +1,100 @@
+#!/bin/sh
+# A reference launcher for Pi's own terminal UI (interactive `pi`) as an OpenRig community harness. Copy it, own it, and
+# change it: OpenRig doesn't ship or run it. It's written for Pi 1.4.0 from its source and not yet checked with a real Pi.
+# It follows docs/reference/community-harness-presets.md (`names_conversation`, `first_input`):
+#   - a new launch (OPENRIG_CONVERSATION set): Pi creates the session whose id is that name, in the seat's own session
+#     folder, and takes the first input as its first prompt;
+#   - a resume (OPENRIG_RESUME_REF set): the launcher opens the one session file whose id is that name, and starts nothing
+#     when there isn't exactly one. Pi's --session-id creates a new session for an id it doesn't find, and --session with
+#     an id can open another session whose id starts the same way, so neither is used to resume.
+set -eu
+
+pi="${OPENRIG_PI:-pi}"
+state="${OPENRIG_HOME:-$HOME/.openrig}/state/pi-tui/${OPENRIG_NODE_ID:-${OPENRIG_SESSION_NAME:-default}}"
+sessions="$state/sessions"
+
+# Each seat gets its own Pi agent folder (settings, logins, skills) unless PI_CODING_AGENT_DIR is already set, so seats
+# don't share settings and a person's ~/.pi/agent isn't touched. Give the seat its provider key in its environment, or its
+# login and settings in this folder, before its first start.
+PI_CODING_AGENT_DIR="${PI_CODING_AGENT_DIR:-$state/agent}"
+# Pi expands a leading ~ in this folder; do the same, so the skills land where Pi looks.
+case "$PI_CODING_AGENT_DIR" in
+  "~") PI_CODING_AGENT_DIR="$HOME" ;;
+  "~/"*) PI_CODING_AGENT_DIR="$HOME/${PI_CODING_AGENT_DIR#"~/"}" ;;
+esac
+export PI_CODING_AGENT_DIR
+mkdir -p "$PI_CODING_AGENT_DIR/skills" "$sessions"
+
+# Copy the installed OpenRig's coordination skills in at each launch, so an upgrade reaches the seat. Pi lists the skills
+# in <agent folder>/skills/ by itself; the seat's first input should still name queue-handoff (see the recipe).
+skills_from="${OPENRIG_PI_SKILLS_FROM:-${OPENRIG_HOME:-$HOME/.openrig}/plugins/openrig-core/skills}"
+set -f
+for skill in ${OPENRIG_PI_SKILLS:-queue-handoff}; do
+  case "$skill" in
+    *[!a-z0-9-]*|-*) echo "openrig-pi-tui-launch: skipping '$skill': not a skill name" >&2; continue ;;
+  esac
+  if [ -d "$skills_from/$skill" ]; then
+    rm -rf "$PI_CODING_AGENT_DIR/skills/$skill"
+    cp -Rp "$skills_from/$skill" "$PI_CODING_AGENT_DIR/skills/"
+  else
+    echo "openrig-pi-tui-launch: no $skill skill in $skills_from; starting without it" >&2
+  fi
+done
+set +f
+
+# Pi's session ids: letters, digits, '.', '_' and '-', starting and ending with a letter or digit.
+valid_id() {
+  case "$1" in
+    "" | *[!A-Za-z0-9._-]* | [!A-Za-z0-9]* | *[!A-Za-z0-9]) return 1 ;;
+  esac
+}
+
+# The first input goes over as Pi's first prompt, byte for byte, after `--`. OpenRig's starts with its identity-context
+# line; Pi would read one starting with @ as a file even after `--`, and may run one starting with / as a command.
+set --
+if [ -n "${OPENRIG_FIRST_INPUT:-}" ]; then
+  first_input=$(cat "$OPENRIG_FIRST_INPUT" && printf '.')
+  first_input="${first_input%.}"
+  if [ -n "$first_input" ]; then set -- -- "$first_input"; fi
+fi
+
+if [ -n "${OPENRIG_RESUME_REF:-}" ]; then
+  id="$OPENRIG_RESUME_REF"
+  if ! valid_id "$id"; then
+    echo "openrig-pi-tui-launch: '$id' isn't a Pi session id; starting nothing" >&2
+    exit 2
+  fi
+  # Pi names a session file <time>_<id>.jsonl, but the file's first line, its header, is what says which session it is:
+  # its top-level type and id, read as JSON as Pi reads it. Pi writes the file only once the conversation has a message, so
+  # a session that never got one can't be reopened. Pi runs on Node, so node reads the header.
+  node="${OPENRIG_PI_NODE:-node}"
+  if ! command -v "$node" >/dev/null 2>&1; then
+    echo "openrig-pi-tui-launch: $node isn't on PATH to read Pi's session headers; starting nothing" >&2
+    exit 1
+  fi
+  found="" count=0
+  for file in "$sessions"/*_"$id".jsonl; do
+    [ -s "$file" ] || continue
+    if head -n 1 "$file" | "$node" -e '
+      let text = ""; process.stdin.on("data", c => { text += c; }).on("end", () => {
+        try { const h = JSON.parse(text); process.exit(h && h.type === "session" && h.id === process.argv[1] ? 0 : 1); }
+        catch { process.exit(1); }
+      });' "$id"; then
+      found="$file" count=$((count + 1))
+    fi
+  done
+  if [ "$count" -ne 1 ]; then
+    echo "openrig-pi-tui-launch: expected one Pi session $id in $sessions, found $count; starting nothing" >&2
+    exit 1
+  fi
+  exec "$pi" --session-dir "$sessions" --session "$found" "$@"
+fi
+if [ -n "${OPENRIG_CONVERSATION:-}" ]; then
+  if ! valid_id "$OPENRIG_CONVERSATION"; then
+    echo "openrig-pi-tui-launch: '$OPENRIG_CONVERSATION' isn't a Pi session id; starting nothing" >&2
+    exit 2
+  fi
+  exec "$pi" --session-dir "$sessions" --session-id "$OPENRIG_CONVERSATION" "$@"
+fi
+echo "openrig-pi-tui-launch: OpenRig passed no conversation name; is names_conversation set in the preset?" >&2
+exit 2

docs/reference/community-harness-examples/pi-tui/pi-tui.yaml

@@ -0,0 +1,15 @@
+# Install as $OPENRIG_HOME/community-harnesses/pi-tui.yaml, with openrig-pi-tui-launch on the seat's PATH.
+# "pi" itself is the built-in managed Pi runtime, so this preset for Pi's own terminal UI takes another name.
+format: 1
+name: pi-tui
+command: openrig-pi-tui-launch    # no model or provider here: they go in the seat's Pi settings or environment
+maintainer: Your name or team
+limits:
+  - checked once with Pi 1.4.0 and OpenRig 0.7.4 (one run, one model); other Pi versions aren't checked
+  - resume not declared until reopening a named session, and a missing one, have been checked with a real Pi
+  - with the default typing guard, a wake submits a person's half-typed draft in Pi's editor together with the wake; run the seat unattended, not where a person also types
+  - a send while a Pi dialog OpenRig doesn't recognize is open, such as /settings, types into the dialog
+  - how Pi takes a wake that arrives while it's busy isn't shown
+  - only the listed openrig-core skills are copied into the seat's Pi agent folder; no guidance or startup files are placed
+names_conversation: true
+first_input: true

docs/reference/community-harness-presets.md

@@ -4,6 +4,9 @@ A community harness is a coding harness an outside contributor maintains. You in
 rig member runs it by naming it as its `runtime`. OpenRig doesn't test these harnesses. It shows them as community
 harnesses, and it reads their activity only from what the screen shows.
 
+For what an integration author owns end to end, including a runner's duties and running a harness somewhere else, see
+the draft [integration author contracts](author-contracts.md).
+
 ## What a community harness gets
 
 What OpenRig does for a community-harness seat, capability by capability. **Partly** and **no** mean OpenRig shows that it doesn't know;
@@ -15,14 +18,14 @@ it never reads an unsupported capability as idle, working, ready or verified.
 | Delivery | yes | `rig send` and queue wakes paste the text into the pane and press Enter, as for any seat. A pane that's back at a shell is refused (`target_runtime_unverified`) and nothing is typed. OpenRig doesn't confirm the harness took the message; `rig send --verify` checks the pane afterwards. |
 | Identity | when declared | With `launch_result`, the conversation is shown as reported by its launcher; with `names_conversation`, as named by OpenRig. Neither is observed by OpenRig. Without either, the identity is unavailable. |
 | Resume | when declared | `resume: true` with one of the two above (see [Reopening a conversation](#reopening-a-conversation-resume-true)). Without it, a resume says `Resume is unsupported: community harness "<name>" is launch-only`, and a restore stops and asks. |
-| Activity | partly | Read from the screen only; there are no activity hooks. Work and attention the screen shows are kept, but a screen that looks idle reads `unknown`, never idle, because a quiet prompt can be an approval or a pause. |
+| Activity | partly | Read from the screen only; there are no activity hooks. Work and attention the screen shows are kept, but a screen that looks idle reads `unknown`, never idle, because a quiet prompt can be an approval or a pause. A [runner preset](#runner-presets-runner-true)'s own reports decide it instead. |
 | Attention | partly | Read from the screen: a permission question or numbered choice that OpenRig recognizes reads as needing attention, and `rig send` won't type into it. A prompt it doesn't recognize reads `unknown`. |
-| Approval | no | Approvals stay in the harness's own pane. OpenRig relays none, so a person answers them there. |
+| Approval | no | Approvals stay in the harness's own pane. OpenRig relays none, so a person answers them there. A [runner preset](#runner-presets-runner-true) is the exception: its runner submits the agent's permission requests to OpenRig, which asks a person, and the default is permit (section 7 of the [harness and runner contract](harness-runner-contract.md)). |
 | Cancel | no | OpenRig has no verb that cancels a running turn. `rig seat stop` and `rig down` end the seat's tmux session; anything the harness runs outside its pane, such as a gateway, is its own, and the preset's `limits` should say what keeps running. Restarting in place and switching harness are refused. |
 | Transport | yes | The seat is a tmux pane like any other, so `rig capture`, `rig send` and `--host` work as they do for any seat. |
 
 **Not placed:** guidance, skills and startup files (a launch warns that they weren't delivered), and the model: set it in the harness's own
-configuration.
+configuration, or let a launcher that opts in take the seat's declared model (`passes_model`, below).
 
 **Check that the harness answers on its own first.** Before writing a preset, start the harness in a terminal outside OpenRig, with the
 same configuration and credentials your launcher will give it, and send it one prompt. Use the folder your launcher gives the seat, not
@@ -48,6 +51,8 @@ launch_result: true             # optional: the launcher reports the conversatio
 resume: true                    # optional, needs one of those two; declare it only once your harness's resume is checked
 startup_assisted: true           # optional: startup submission needs a person's check (below)
 first_input: true                # optional: launcher takes the first turn from OPENRIG_FIRST_INPUT (below)
+# runner: true                  # optional, needs launch_result: the launcher reports the seat's activity (below)
+# passes_model: true            # optional: the launcher takes the seat's declared model from OPENRIG_MODEL (below)
 ```
 
 - **`command`** is a single program and its arguments. Shell syntax such as `|`, `&&`, `;`, redirections or `$(…)`
@@ -144,6 +149,45 @@ For a reported conversation, OpenRig also says so plainly when:
 - **the saved conversation isn't this harness's own** (a former runtime's id, say): it isn't passed to the launcher. A restore stops and asks, and elsewhere the resume says it isn't possible;
 - **the record's folder can't be made:** the launcher couldn't be told what to reopen, so nothing starts.
 
+## Runner presets (`runner: true`)
+
+A plain preset's seat reads its activity from the screen. A **runner** preset is for a launcher that hosts the agent itself
+and reports what it's doing, as OpenRig's own Pi runner does. The seat's activity then follows those reports.
+
+- **It needs `launch_result: true`.** A runner reports the conversation it opened in its launch record, so `runner: true`
+  without `launch_result` is refused.
+- **The declaration is fixed at launch.** Each launch, a restore's included, records on the session what the preset
+  declares at that moment. Editing the preset changes the next launch, not a running seat. A session recorded before this
+  field existed reads as plain.
+- **No record means attention.** When a runner's launch record doesn't arrive within the wait, or its folder can't be
+  made, the seat needs attention instead of reading ready. A plain `launch_result` preset still starts, with its identity
+  unavailable.
+- **Input is as for any seat.** A send pastes the text and presses Enter, so a runner should submit everything up to that
+  Enter as one prompt.
+
+**What the runner posts.** The runner posts each event to `POST /api/activity/hooks` with the local activity hook token,
+one at a time and in order, without making the agent wait. It finds the address and token in `OPENRIG_URL` (or
+`OPENRIG_PORT`) and `OPENRIG_ACTIVITY_HOOK_TOKEN`, or else in `$OPENRIG_HOME/activity-endpoint.json`. Every post names:
+- the seat, as `sessionName` and `nodeId`;
+- `runtime`, the preset's name;
+- `generation`, from `OPENRIG_OCCUPANT_GENERATION`;
+- an increasing `sequence`, and `occurredAt`;
+- `hookEvent` and `subtype`, as in the table.
+
+| The runner posts | When | The seat reads |
+|---|---|---|
+| `UserPromptSubmit` | it gave the agent a prompt | working |
+| `active` or `PreToolUse` | the agent is working | working |
+| `Notification` / `permission_prompt` | the agent asked for a permission | needs input |
+| `Stop` / the turn's stop reason | the turn ended | idle |
+| `Stop` / `<preset name>_exited` | the agent's process ended | unknown, never idle |
+
+- **The screen still counts,** but only as evidence of work: on its own it never reads the seat idle.
+- **The generation attributes a post; it isn't authentication.** A post with an earlier occupant's generation doesn't
+  count for the live seat. The hook token is one local token, so any process that holds it can post for any seat.
+- **A runner's full duties** (liveness, input, cancellation and permissions) are in the draft
+  [harness and runner contract](harness-runner-contract.md).
+
 ## An example: Hermes
 
 [`community-harness-examples/hermes/`](community-harness-examples/hermes/) has a preset and a small launcher for Hermes
@@ -214,8 +258,59 @@ CLI.
 
 Copy both and maintain them as your own.
 
+## An example: Pi's terminal UI
+
+[`community-harness-examples/pi-tui/`](community-harness-examples/pi-tui/) has a preset and a small launcher that run Pi's
+own terminal UI as a community harness. `pi` is the built-in managed Pi runtime, so the preset is named `pi-tui`. OpenRig
+delivers to this seat as to any community seat (the Delivery row above), not as to a plain `runtime: terminal` seat, which
+queue wakes don't reach. Checked once, on Pi 1.4.0 with OpenRig 0.7.4 and one model: a parked row's wake reached Pi in
+the same session (the same session file) and Pi resumed the work, and a handoff reached a second `pi-tui` seat, which
+picked it up unattended. The first seat's task named the commands to use, so this shows a wake reaching Pi and Pi acting
+on it, not Pi choosing by itself to park or hand work on.
+- **The launcher:** each seat gets its own Pi agent folder (`$OPENRIG_HOME/state/pi-tui/<node>/agent`, unless
+  `PI_CODING_AGENT_DIR` is set) and its own session folder. At each launch it copies the installed `openrig-core` skills
+  it's given (`queue-handoff` by default) into that agent folder. A new launch starts `pi --session-id <the name OpenRig
+  gave>` and passes the first input as Pi's first prompt, byte for byte.
+- **Resume isn't declared.** The launcher can reopen a named session: it opens the one session file whose header has that
+  id, and starts nothing when there isn't exactly one, because Pi's own `--session-id` creates a session for an id it
+  doesn't find. Declare `resume: true` only after a real Pi has reopened a named session and refused a missing one.
+- **Its limits:** run a `pi-tui` seat unattended, not where a person also types into it.
+  - With the seat's [typing guard](seat-delivery.md) at its default, a wake submits a person's half-typed draft in Pi's
+    editor together with the wake. The guard's draft-aware mode isn't a known fix for this.
+  - A send while a Pi dialog OpenRig doesn't recognize is open, such as `/settings`, types into the dialog.
+  - Not shown: how Pi takes a wake that arrives while it's busy (a message sent during a busy turn was answered after
+    it), and other Pi versions.
+- **Give the seat a model** before its first start: its provider key in its environment, or its login and settings in its
+  agent folder.
+
+Copy both and maintain them as your own.
+
+## An example runner: dsh
+
+[`community-harness-examples/dsh/`](community-harness-examples/dsh/) has a runner preset (`runner: true`) and a small runner
+for DeepSeek Harness, which speaks ACP on stdio and has no terminal UI of its own. It's written for dsh 0.2.1-rc.2 and
+checked only against a stand-in agent.
+- **The runner:**
+  - it opens the conversation and writes the launch record;
+  - it turns each submitted block into one prompt;
+  - it reports turn boundaries as activity hooks;
+  - it cancels on `/abort` or Ctrl-C, as the [runner contract](harness-runner-contract.md) describes.
+- **Permissions:** it sends each request to OpenRig, which asks the default person or the seat's override.
+  - **When no one can be asked, or OpenRig can't answer,** it permits with dsh's own allow option (`allow_once`, else
+    `allow_always`), which is OpenRig's default. `allow_always` creates a standing grant in dsh.
+  - **It refuses and holds needs input** only after `--deny`. A person's no when dsh offered no reject option, or a
+    request OpenRig says isn't the current occupant's, is answered cancelled.
+- **Each seat gets its own dsh home,** with dsh's session-log and plugin-list uploads to DeepSeek turned off. Give that
+  home credentials, or put a provider key in the seat's environment.
+- **Resume isn't declared.** The runner reopens a conversation by the method dsh advertises, and reports one missing only
+  on dsh's own not-found error. Declare `resume: true` only after a real dsh has reopened a conversation and refused a
+  missing one.
+
+Its [README](community-harness-examples/dsh/README.md) has the setup and a check you can run without a model. Copy both
+and maintain them as your own.
+
 The launcher is the contributor's code, in the contributor's package. OpenRig runs no code from a preset and loads no
-module into its daemon. A preset that sets `launch_result`, `names_conversation`, `startup_assisted` or `first_input` is refused by an older
+module into its daemon. A preset that sets `launch_result`, `names_conversation`, `startup_assisted`, `first_input`, `runner` or `passes_model` is refused by an older
 OpenRig that doesn't know the field.
 
 ## Pass the first startup input at launch
@@ -228,6 +323,87 @@ OpenRig records `node.startup_first_input_passed` with the path and `submission:
 
 Presets without `first_input` and built-in runtimes keep their existing behavior. Older OpenRig versions reject this new field as unknown. The private input file remains for diagnostics; it can contain private startup context. A launcher that turns its contents into one argument is subject to the platform’s argument-size limit; the launcher must report a launch error rather than silently truncate the turn. The examples’ argument transport is covered with stand-ins; native unattended startup still requires qualification with the installed harness.
 
+## Pass the seat's declared model at launch
+
+A format-1 preset may set `passes_model: true` (a boolean, default false). Then a model declared for the seat (in the rig
+spec, or set later with `rig seat set-model`) reaches its launcher as `OPENRIG_MODEL`, on new and resumed launches alike. A seat that declares no model gets none,
+and an inherited `OPENRIG_MODEL` is cleared. The launcher maps it into its harness's own setting; Core knows no harness's
+model flag or config key.
+
+OpenRig keeps three things apart:
+- **requested:** the declared model, as `rig ps --nodes` shows it ("configured", or "not-declared");
+- **passed:** the launcher was handed the model. OpenRig records `node.startup_model_passed` with the model and whether the
+  launch was a resume only when the launch command's send succeeded, and a later failure of the launch doesn't undo that.
+  It never says passed from the declaration alone or for a launch that failed before anything was typed. When typing
+  began and the send then failed, the model's dispatch is uncertain: it isn't recorded as passed, and it isn't retyped;
+- **effective:** the model that actually runs. OpenRig doesn't observe it for a community harness, so it reads UNKNOWN.
+
+Passing a model never makes OpenRig pick a default, substitute another model, or retry with one. A model the harness
+refuses is the harness's own error.
+
+The examples apply the request for a new session only; on a resume, the example launcher doesn't apply it, and which
+model a resumed session runs stays unknown unless the harness's own evidence says otherwise:
+- **Cursor:** `--model "$OPENRIG_MODEL"` for a new chat. What a resumed chat runs hasn't been checked: the launcher leaving
+  out `--model` doesn't show that it keeps its earlier model.
+- **Hermes:** Hermes takes its model only from the seat's `config.yaml`, so for a new session the launcher writes
+  `model: "<id>"` there only when that file is absent or empty. A seat that has its own `config.yaml` keeps it
+  byte-for-byte, and the launcher says the model wasn't applied: set it in that file. A seat set up from
+  `config.yaml.example` already has a config, so set its model there. Hermes v0.22.4 fixes a session's
+  launch settings when it creates it (its own source), so a resumed Hermes session runs with what it was created with.
+
+Both example presets ship with `passes_model: false` until that mapping has been checked with the real harness. Presets
+without the field and built-in runtimes are unchanged, and older OpenRig versions reject it as unknown.
+
+## Interactive Pi in a terminal seat: the coordination skills
+
+This is for a seat that runs interactive `pi` from a `runtime: terminal` member's startup, or from a community preset's
+launcher. It isn't for a managed `runtime: pi` seat, which already copies its selected skill directories into the seat's own
+Pi agent directory (`$OPENRIG_HOME/state/pi/<session>/agent/skills/`), or for Workshop, whose Pi profiles carry pinned skill
+copies (openrig-world's `PI-SKILLS.md`). The [`pi-tui` example](#an-example-pis-terminal-ui) above does step 1 for its
+seats in its launcher.
+
+OpenRig installs no skills for a terminal or community seat: it skips their resources and startup files, and `rig skill
+loadout` covers Claude Code and Codex only. Without the coordination skills, the agent there doesn't know the queue's habits,
+such as parking a row that waits on CI instead of leaving it in progress, or handing work on so the next row names its
+source. The seat's setup does two things.
+
+1. **Make the released skills readable.** The installed OpenRig's own plugin carries them, matching the installed release:
+   `rig plugin list` shows `openrig-core` and its path (usually `$OPENRIG_HOME/plugins/openrig-core`), and the skills are in
+   its `skills/` folder. Copy whole skill directories, keeping their helper files and modes, into the `skills/` folder of the
+   Pi agent directory that seat uses (`PI_CODING_AGENT_DIR`, or `~/.pi/agent` by default). At least `queue-handoff`:
+
+   ```bash
+   CORE="$(rig plugin list | awk '$1 == "openrig-core" {print $NF}')"
+   PI_AGENT_DIR="${PI_CODING_AGENT_DIR:-$HOME/.pi/agent}"
+   mkdir -p "$PI_AGENT_DIR/skills"
+   cp -Rp "$CORE/skills/queue-handoff" "$PI_AGENT_DIR/skills/"
+   ```
+
+   Copy again after each OpenRig upgrade: a copy doesn't follow the plugin.
+2. **Point the agent at it when it starts.** Don't count on Pi finding the skill by itself; name it in the seat's first turn.
+   For example:
+
+   > Before you work your queue, read `<Pi agent directory>/skills/queue-handoff/SKILL.md` in full: it's how this team hands
+   > work on, parks it and waits. When you wait on CI or a long job, park the row with `rig queue block <id> --on
+   > external:<what> --wake-after <duration> --continuation "<what resumes>"`.
+
+   - A community preset with `first_input: true` carries it in the seat's first input when the file is declared with
+     `delivery_hint: send_text` (see "Pass the first startup input at launch" above). A `.md` file left at the `auto` hint
+     counts as guidance, which a community seat doesn't take.
+   - A plain terminal seat can pass it as Pi's initial message in the command its startup runs. Check `pi --help` for that
+     argument in your Pi version.
+   - `first_input` carries text, not files: it doesn't install the skills.
+
+**In a plain terminal seat, a queue wake doesn't reach Pi.** OpenRig doesn't treat a `runtime: terminal` seat as an agent
+seat. When a parked row's wake fires, OpenRig doesn't type it into Pi, and a handoff to that seat records a failed wake
+(`failed:terminal-seat`). Parking still records the wait, but `--wake-after` won't bring Pi back. Check the seat's parked
+rows yourself (`rig queue whoami --session <seat>`, or `rig queue show <id>`), or tell Pi to continue the named row.
+
+**What this shows, and what it doesn't.** A copied folder and a startup line put the text in front of the agent. They don't
+show that Pi discovered the skill by itself, or that the agent follows it. Check those separately: that the agent read the
+whole file (its tool trace), and what its rows record, such as a CI wait parked on `external:` with a wake, and a handoff
+made with `handoff-and-complete`, whose next row names it as its source.
+
 ## Write your own
 
 Start from the example closest to your harness, and change it. First check that the harness answers a prompt on its own

docs/reference/edge-types.md

@@ -1,7 +1,8 @@
 # Edge Types Reference
 
-Last validated against code: 2026-10-07, at the 0.6.6 cut `2620dea8` (whole document at `fcaf1f8e`; its sources unchanged in edge behaviour since then)
-Source of truth: `packages/daemon/src/domain/rigspec-schema.ts`, `packages/daemon/src/domain/rigspec-instantiator.ts`, `packages/daemon/src/domain/queue-owner.ts`
+Historical whole-document validation: 2026-10-07, at the 0.6.6 cut `2620dea8` (whole document at `fcaf1f8e`).
+The fallback escalation section and deferred-target exceptions below describe #605.
+Source of truth: `packages/daemon/src/domain/rigspec-schema.ts`, `packages/daemon/src/domain/rigspec-instantiator.ts`, `packages/daemon/src/domain/queue-owner.ts`, `packages/daemon/src/domain/escalation-owner.ts`
 
 ---
 
@@ -16,8 +17,7 @@ See `docs/reference/rig-spec.md` for the YAML syntax.
 
 ## Edge Kinds
 
-Five edge kinds are accepted by the pod-aware validator (the legacy flat-node format accepts only `delegates_to`,
-`spawned_by` and `can_observe`):
+Five edge kinds are accepted by both the pod-aware and legacy flat-node validators:
 
 | Kind | Accepted | Has Runtime Behavior | Description |
 |------|----------|---------------------|-------------|
@@ -25,13 +25,13 @@ Five edge kinds are accepted by the pod-aware validator (the legacy flat-node fo
 | `spawned_by` | yes | **yes — affects launch order** | Source was spawned by target |
 | `can_observe` | yes | no | Source can observe target's output |
 | `collaborates_with` | yes | no | Peer collaboration |
-| `escalates_to` | yes | no | Source escalates to target |
+| `escalates_to` | yes | **yes — fallback queue escalation** | Source escalates to target |
 
 ## What Edges Actually Do Today
 
 ### Launch ordering
 
-Only `delegates_to` and `spawned_by` affect runtime behavior. They constrain the order in which nodes are launched:
+Only `delegates_to` and `spawned_by` constrain the order in which nodes are launched:
 
 - **`delegates_to`**: source launches BEFORE target. The delegator must be up before the delegate.
 - **`spawned_by`**: target (parent) launches BEFORE source (child). The parent must be up before the child it spawned.
@@ -46,13 +46,42 @@ All other edge kinds (`can_observe`, `collaborates_with`, `escalates_to`) do NOT
 wake ladder escalates an unacknowledged row to that orchestrator, and stuck-row findings and recovery go to it. A seat
 with no incoming `delegates_to` edge has no default orchestrator. `spawned_by` plays no part here.
 
+### Fallback escalation across rigs
+
+A seat with **no declared `delegates_to` parent** may declare one `escalates_to` destination. The wake ladder and
+stuck sweep use it for new recovery rows; they do not move the original obligation or reassign an already-open recovery. A stopped or unbound declared
+parent still takes precedence, preserving the existing parent route and its fallback.
+
+```yaml
+edges:
+  - kind: escalates_to
+    from: work.lead
+    to: review-advisor@coord
+```
+
+The target is a canonical same-daemon seat address (`pod-member@rig`), or an existing local member using the usual
+edge syntax. Named targets are stored without requiring the other rig to exist and resolved when needed. A matching
+physical session name takes precedence over a derived canonical address. Deferred declarations survive export and
+snapshots; they don't constrain launch order or make the rigs depend on each other's lifecycle.
+
+Delivery is one hop. Recovery rows carry `escalation-route:one-hop`, so a destination's own `escalates_to` declaration
+does not forward the recovery again. A missing or stopped destination, multiple matches or declarations, self route,
+or declared cycle preserves the existing fallback: the wake ladder proceeds to its operator rung; the sweep retains
+its existing owner fallback. The recovery body/transition explains the result. Cycle inspection includes stopped
+seats and delegation parents; an unresolved downstream endpoint is reported as unknown, never as proof of no cycle.
+Queue address admission is unchanged. Cross-host and synthetic destinations are not supported.
+
+On an existing rig, use `rig edge add <rig> work.lead review-advisor@coord --kind escalates_to`; `--plan` writes nothing.
+The receipt's target node ID is null for a deferred address. Remove it with the exact edge ID returned by add.
+A rig-local graph omits deferred targets, since those seats are outside that graph; the raw topology and export retain them.
+
 ### Graph visualization
 
-All edges are rendered in the UI topology graph. They create visual connections between nodes, helping the operator understand the team structure. Pod-local edges appear within the pod group; cross-pod edges connect across groups.
+Local-node edges are rendered in the UI topology graph; deferred cross-rig targets are omitted. They create visual connections between nodes, helping the operator understand the team structure. Pod-local edges appear within the pod group; cross-pod edges connect across groups.
 
 ### Identity projection
 
-All edges appear in `rig whoami --json` output under `edges.outgoing` and `edges.incoming`, with their kind and the connected peer's identity. This lets an agent know its relationship to other members — but the agent interprets this information, it's not enforced by the runtime.
+Local-node edges appear in `rig whoami --json` output under `edges.outgoing` and `edges.incoming`, with their kind and the connected peer's identity. This lets an agent know its relationship to other members — but the agent interprets this information, it's not enforced by the runtime.
 
 ### Attach hint heuristic
 
@@ -64,7 +93,7 @@ node, in creation order, whose session is running, or else the first node with a
 Edges do NOT currently:
 
 - **Route messages** — `rig send` can target any session, regardless of edges (queue escalation, above, is the one
-  daemon path that follows `delegates_to`)
+  daemon path that follows `delegates_to` or fallback `escalates_to`)
 - **Enforce delegation** — an agent can communicate with any peer, not just its edge targets
 - **Control permissions** — there's no edge-based access control
 - **Affect transport** — `rig capture`, `rig broadcast`, etc. work based on session identity, not edge topology
@@ -128,7 +157,7 @@ If none fit, `can_observe` is the safest default — it documents the relationsh
 ## Validation Rules
 
 1. Edge kinds must be one of: `delegates_to`, `spawned_by`, `can_observe`, `collaborates_with`, `escalates_to`
-2. Pod-local edges use unqualified member IDs — both `from` and `to` must exist in the same pod
-3. Cross-pod edges use `pod.member` format — both endpoints must resolve to actual members
+2. Pod-local edges use unqualified member IDs — both endpoints must exist in the same pod, except a named `escalates_to` target as above.
+3. Cross-pod edges use `pod.member` format — both endpoints must resolve to actual members, except a named `escalates_to` target as above.
 4. Cross-pod edges must reference different pods (same-pod edges should use the pod-local syntax)
 5. Self-edges are not validated at the schema level but are not meaningful

docs/reference/execution-platform-contract.md

@@ -0,0 +1,138 @@
+# Execution and platform integration contract — draft
+
+This chapter describes the placement and resource responsibilities of an integration author. The harness/runner chapter owns native protocol, model input, activity, attention, permission exchange and turn cancellation. This chapter owns where that harness runs, which files and credentials it can reach, how native state survives an environment change, and how results and resources return to their owners.
+
+**Status:** a design contract for authors, with an existing-compatible composition and one measured private implementation. The measured baseline used OpenRig 0.7.4, OpenShell 0.1.6 and Claude Code 2.1.310. 0.7.6 published this draft; publishing it does not implement a generic environment driver, a `DeploySpec` loader or a cross-host native-resume guarantee. Proposed operations below are not commands or accepted configuration keys. Capability support must be stated for the actual integration and version.
+
+## Keep the identities and responsibilities separate
+
+RigSpec composes members, their AgentSpecs/profiles, runtimes, working directories and relationships. AgentSpec supplies reusable role, startup and context declarations. A built-in runtime or registered community preset supplies a launch path. An environment integration supplies placement and provider-specific resource handling around that path. Existing declarations do not automatically provision a remote filesystem, credentials or compute.
+
+| Identity or handle | Meaning and responsibility |
+|---|---|
+| Stable seat and work identity | The member's canonical address/node and owned work records. OpenRig retains this identity across executions; an environment IP or display name does not replace it. |
+| Native session identity | The harness's conversation token/UUID, qualified by harness and resume-reference type. The harness owns its format and resume semantics. |
+| Environment binding | The integration's durable record linking placement, immutable provider resource IDs, native-state location, projection and ownership. It is separate from the native session token. |
+| Execution generation | A particular launch/process instance, with observable process/start identity when available. A new process can resume an old conversation. |
+| Viewer/client | A terminal attachment or connection to an execution. Closing it need not stop either the native process or compute. |
+
+An application service bundled with a rig is a separate responsibility. An artifact browser or database in `RigSpec.services` serves the team's application; its presence does not place an agent inside that service. Bundling service definitions also does not bundle image bytes, native binaries, authenticated homes or live runtime state.
+
+## A minimal composition using existing surfaces
+
+The following is a **fragment of one pod** (`pods[].members`), using existing RigSpec fields, not a complete standalone RigSpec:
+
+```yaml
+pods:
+  - id: dev
+    label: Development
+    members:
+      - id: coordinator
+        agent_ref: local:agents/coordinator
+        profile: default
+        runtime: claude-code
+      - id: worker
+        agent_ref: local:agents/worker
+        profile: default
+        runtime: example-contained-claude
+```
+
+The worker runtime names an author-installed community preset. This preset shape uses existing fields exercised on the 0.7.4 baseline:
+
+```yaml
+format: 1
+name: example-contained-claude
+maintainer: Integration author
+command: ./scripts/launch-contained-worker
+launch_result: true
+resume: true
+startup_assisted: true
+limits:
+  - Requires an explicitly prepared environment binding and guest projection.
+  - Native resume is qualified only for the documented retained-state recipe.
+  - Viewer closure does not implement turn cancellation or resource deletion.
+```
+
+This is an integration skeleton, not an installed or qualified preset. The author must supply the executable, resolve its path against the actual launch working directory, register the preset in the existing OpenRig home, and prepare the member's host working directory. The launcher implements the preset's declared launch-result and typed-resume behavior; the runner chapter defines those mechanics. `startup_assisted` requires a startup-submission check; it does not hold back submission or establish successful context delivery or ingestion. A member's model setting is not automatically honored by a custom launcher: its resolved native model/effort must be explicit in that launcher's supported configuration. With `passes_model: true`, Core passes the declared model through `OPENRIG_MODEL`; the launcher must map and apply it. The effective model remains unknown to Core.
+
+For a contained worker, the launcher may resolve a private binding, create or attach to the selected resource, project explicit inputs, then connect to the native process inside it. In the measured recipe, OpenRig ran on the host, the plain coordinator used the built-in native runtime, and the worker used a private community launcher. The wrapper performed projection, provider-reference delivery, resume validation and artifact export explicitly. Those effects were not supplied by the preset declarations themselves. The integration must not assume host guidance/skills, hook paths or environment variables automatically reach a community runtime's guest.
+
+A downstream author can implement this composition today around a qualified provider API and native harness. It remains an author-maintained integration until maintained platform support is separately implemented and documented. Validate against the target OpenRig schema and effective launch configuration; a parsed field or successful process start alone does not establish its effect.
+
+## Placement and launch receipts
+
+Before mutation, resolve the selected host/provider, workspace, image/native versions, reachability, credential references, resource request and ownership. Record existing resources and the selected resource scope. Host headroom includes supervisors, gateway and other team members; a workload memory request is not proof of their combined cap. Report requested, admitted and observed resources separately. Do not infer enforced CPU/memory or isolation policy from accepted flags.
+
+Persist a launch intent before a create operation. Use an attempt/retry identity where the provider supports it; otherwise reconcile the original operation before another create. A public binding receipt carries only nonsecret values. If the harness's native resume reference works like a bearer credential (anyone holding it can reopen the conversation), keep it in secret-held receipt data, never in the public receipt. A useful public receipt identifies:
+
+- canonical seat, harness version and native resume-reference type;
+- provider/host/workspace, immutable environment ID and execution generation;
+- image/policy versions, requested resources and observed readiness;
+- inner working directory, private native-state location and projection manifest;
+- a public-safe native reference (never a bearer-like resume token), creator or attached-consumer disposition, and exact owned resource handles;
+- lifecycle operations supported, completion observations, pending effects and errors.
+
+A resource name or IP can be reused. Reconcile immutable ID, workspace and any available generation/fingerprint before acting. A lost response is **unconfirmed**, not success or absence. Record provider failures without exposing credentials. Readiness should identify what is ready: provider reachable, filesystem projected, native process started, team identity authenticated, or task completed.
+
+## Project files and credential references deliberately
+
+Stage only declared inputs: role/startup material, selected skills, current work context, schemas and required native settings. Preserve unmanaged files and report conflicts. Review every path-bearing setting for its destination: host-only hook scripts, interpreters, status collectors and loopback URLs may not exist or be reachable in the guest. Missing activity or permission transport must remain unsupported rather than being inferred from the base harness name.
+
+Distinguish three observations: host staging, destination delivery and native ingestion. A destination manifest can record relative path, byte count and hash; a useful task using the current context revision/nonce can show consumption. Neither observation substitutes for the other. Current working input may change between tasks while an earlier artifact stays immutable.
+
+Retained native history is a different data class from ordinary project context. Preserve its exact bytes through the harness's supported resume path; do not edit records or synthesize a conversation. Preserve a complete checkpoint prefix with byte length/hash if the harness may append after the checkpoint. Record which native-state files are required and which remain unavailable. Never substitute a fresh conversation seeded with a report while calling it native resume.
+
+Keep secret values out of RigSpec, preset YAML, bindings, argv and public receipts. The provider/credential owner supplies scoped references or native-supported credential helpers and controls revocation. A successful login, provider registration or model-list request is not evidence of usable inference credit or an authorized spend envelope. Do not infer subscription portability from a provider name or copy an authenticated owner home into a guest.
+
+Authenticate guest coordination through the actual granted member route. A shared source IP is not a seat identity. Record the grant-bound principal and scope separately from resource placement; distinct principals do not imply per-seat row ACLs or isolation between co-located secrets.
+
+## Own only the effects assigned to the integration
+
+Record **created** versus **attached** for each launch receipt. The creator may stop/delete the exact resource it created under the selected policy. An attached consumer normally closes only its own client and any separately identified process it owns; it must preserve the external environment, native session, workspace and shared services. Attachment must not silently transfer provider custody. An explicit owner-authorized transfer is a separate recorded action.
+
+Describe these effects independently:
+
+| Operation | Required distinction |
+|---|---|
+| Cancel a turn | Ask the runner to stop current work and observe its result. Unsupported cancellation must not be replaced silently by killing compute. |
+| Close a viewer | Release that client/connection; report whether native execution continues or is unknown. |
+| Stop native execution | End the identified native process and observe its exit. This is separate from task success. |
+| Stop compute | Release the documented provider allocation; identify retained storage and actual provider/process state. |
+| Delete an owned environment | Remove the exact creator-owned resource and reconcile final absence, including provider-managed dependents. |
+
+Use graceful shutdown first where supported, then only explicitly owned, identified handles for stronger termination. Never equate pane disappearance with process or resource release. Persistent/non-persistent lifetime describes when execution should stop; durability describes recovery. An explicit non-persistent completion condition must name its invocation/work scope. An unrelated queue closure or an empty queue is not permission to terminate a member, and no new lifecycle YAML syntax is specified here.
+
+## Resume exactly, or return an honest disposition
+
+A resume request identifies the seat, harness-qualified native reference, prior binding, expected environment boundary and retained state. Validate those before launching the native harness. Reject mismatched reference types, missing/changed history, wrong immutable resources, unavailable required files and duplicate active executions. Never use “latest” as a replacement for an exact requested conversation or silently start fresh on failure.
+
+Separate the outcomes: same native conversation resumed; new conversation rebuilt from artifacts; fresh launch; unsupported; refused; or unconfirmed. Environment continuity has its own outcome: same retained environment, recreated compute with copied state, different host, or unknown. A matching UUID alone does not show that native history was consumed.
+
+If a provider stops exposing file transfer after canonical-process exit, capture artifacts/history and publish any retained resume descriptor **while transfer is still available**. Verify destination bytes before shutdown. After completion, use host/provider observations instead of assuming guest exec still works. For recreated compute, explicitly transfer the retained checkpoint and required artifact into the successor environment and validate them before publishing its resume descriptor. A new environment ID is expected and must be recorded as a successor, not disguised as the old environment.
+
+The measured private recipe demonstrated this latter boundary: the original sandbox was deleted; copied native history and task bytes were restored into new containers on the same physical host; the native harness launched with `--resume` and the retained UUID. Its first new reply recalled a private conversation-only marker before tools, with no marker in the new input and no native-history tool read. It then verified artifact 1, used changed task context, produced artifact 2 and returned it for independent coordinator verification. This establishes one bounded native-return/work result. It does **not** establish new-host migration, RAM restoration, same-container stop/start, cgroup enforcement, general portability or shipped platform support.
+
+## Export results before teardown
+
+A guest path/hash note means bytes exist in a guest; it is not delivery to the consumer. Identify the actor holding the provider transfer handle. In the measured recipe that actor was the host wrapper, not the guest worker.
+
+The worker publishes its report once and records the task/path/hash on its existing child. The wrapper validates the binding, task, allowed path, file type, size and expected hash; transfers to a fresh host staging path; checks the returned bytes/schema/context; and publishes without overwriting an existing result. It records an actual consumer-accessible destination, hash, byte count and actor through the authenticated work route. The worker closes its child after the matching export receipt; the coordinator independently reads/verifies those bytes before closing its parent. An ambiguous transfer is reconciled by existing bytes/hash, not a new report or duplicate task.
+
+Complete exports before resource deletion. If output cannot be returned, preserve the exact resource/owner and failure state. Cleanup then reconciles the exact owned workload, supervisor, volumes and provider records to the intended stopped/absent state. Keep reused engines, caches, authenticated environments and attached resources. Revoke scoped credentials/grants through their owners and record remaining state. “Delete requested” and “absence verified” are different receipts.
+
+## Proposed generic platform seam — not implemented
+
+The existing composition need not acquire another YAML format to describe these responsibilities. If a maintained platform seam is selected, the following **design pseudocode** names the smallest useful operations; it is not a loader or API available in this release:
+
+```text
+prepare(placement, ownership, resourceRequest, credentialRefs) -> bindingReceipt
+project(binding, explicitFiles, nativeCheckpoint?) -> deliveryReceipt
+launch(binding, harnessLaunch, exactResumeRequest?) -> executionReceipt
+observe(binding, execution) -> observedState | unsupported | unconfirmed
+export(binding, artifactRequest) -> publicationReceipt
+shutdown(binding, execution, selectedEffect) -> effectReceipt
+```
+
+The provider implementation interprets its image/network/compute configuration. The runner interprets native protocol and control. Core would retain typed provenance and route the selected effects; it must not reinterpret arbitrary provider keys, leak secrets or collapse ownership into a stop boolean. Each mutating operation needs an attempt identity and reconciliation route; partial receipts remain useful after failure. Capability reporting must expose unsupported projection, cancellation, resume or cleanup rather than letting method names imply support.
+
+The most useful author evidence follows the real work path: exact resolved placement and native launch; delivered current context; authenticated work ownership; returned artifact bytes; exact resume boundary and consumed retained state; and final resource disposition. A parser pass, launcher success, viewer reconnect or same display name leaves the corresponding effect unconfirmed. These distinctions let an author implement and diagnose an integration; they do not create a new release gate or require another runtime campaign.

docs/reference/getting-started.md

@@ -191,11 +191,11 @@ You can also read [`skills/rigs/SKILL.md`](../../skills/rigs/SKILL.md) directly.
 script will do and changes nothing:
 
 ```sh
-curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.0/scripts/install.sh | sh -s -- --dry-run
+curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.5/scripts/install.sh | sh -s -- --dry-run
 ```
 
 The same command without `--dry-run`,
-`curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.0/scripts/install.sh | sh`, installs the
+`curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.5/scripts/install.sh | sh`, installs the
 latest published `@openrig/cli` with `npm install -g`,
 runs the Node.js and SQLite check, then `rig setup --dry-run` and `rig setup`. `rig setup` checks both Claude Code and
 Codex and may install a missing one, as described below. A failed step prints
@@ -331,6 +331,16 @@ every seat while those seats still run the sessions it started, reports that
 restore's outcome instead of starting another. Explicitly repeating the saved
 `--non-interruptive` or `--no-non-interruptive` choice also joins that restore.
 
+`rig start --last` restores a rig unless every seat it launched was stopped on
+purpose with `rig down` or `rig seat stop`. A rig whose stop was never recorded
+counts as last running: one taken down with `rig down` before OpenRig 0.6.12, or
+one whose seats exited or were lost without `rig down`. So does a rig with no snapshot
+whose saved state can be restored. To keep such a rig down, run `rig down <rig>`
+once, by rig id when names are shared; it records the stop, and `--last` then
+leaves the rig down. `--last` never selects the kernel. Under `--no-kernel`
+(`OPENRIG_NO_KERNEL=1`) the daemon skips the kernel's automatic restore too;
+bring it back with `rig up kernel --existing`.
+
 An explicit `rig up kernel --existing --fresh <seat...>` waits for startup's
 restore to finish, then replaces only the named seats that still hold the exact
 sessions startup launched. Other seats keep running. An operator's conversation
@@ -499,7 +509,12 @@ rig terminal open saved:kernel --provider herdr --json
 rig terminal open saved:kernel --provider cmux --json
 ```
 
-Use `rig terminal status --json` to inspect provider availability and liveness.
+Use `rig terminal status --json` to inspect provider availability, liveness and the
+daemon's effective tmux server. To keep managed seats out of your personal tmux
+session list, opt into [an isolated server](tmux-servers.md) with
+`rig config set terminal.tmux_server isolated`; existing live seats defer the
+switch until they are stopped and the daemon restarts.
+
 To keep rigs in separate Herdr sessions, select one explicitly:
 `rig terminal open future-prod-rig --provider herdr --session future-prod`.
 `--session default` selects Herdr's default session; omitting the option keeps
@@ -838,13 +853,47 @@ the lifecycle API without booting a new kernel, then stop the remaining rigs.
 | "Claude resume did not finish within Ns; Claude may still be starting, so the launch was kept" | A restored Claude seat was not interactive when the wait ended. Its pane and resume token are kept. Look at the pane; once Claude is up, run `rig seat clear-attention <seat>`. The wait is about `runtime.readiness_timeout_seconds` in total for each seat (default 30 s), counting the first 3.2 s, and the time adds up seat by seat in a restore. Raising the setting for slow fresh launches lengthens it. If `rig up` or `rig launch` times out at 120 s, that is not a rollback: check `rig ps` before running it again. |
 | Daemon is healthy, kernel is still starting | Read `rig status` and `rig ps --nodes --rig kernel`; kernel readiness is separate. |
 | Kernel seats fail within a second of the first start, and tmux reports "server exited unexpectedly" | Check that your tmux can capture a pane, then start the failed seats again; see the tmux check below this table. |
-| Shared terminal is absent | Inspect the existing kernel binding and recovery state; use standalone `rig tui` while resolving it. |
+| Shared terminal is absent | Check the kernel's `operator.human` terminal separately from its agents; recover only that seat as described below. Standalone `rig tui` remains available while resolving it. |
 | Viewing terminal was closed | Open the conversations again with `rig terminal open saved:kernel --window`; `rig tui --shared` is only the dashboard fallback if that window cannot open. Do not relaunch the team. |
 | Daemon restarted but tmux survived | Re-read `rig status` and the existing queue; a daemon restart is not a fresh project. |
-| Host reboot lost tmux sessions | Starting the daemon restores the kernel by itself when all of its seats were lost; check `rig status`, and if it reports the restore failed, run `rig up kernel --existing`. For other rigs, open `rig`, start the daemon if needed (press **S** if it opens on the work views), and select the existing rig and seats. Resume is the default; a fresh conversation needs a separate decision. |
+| Host reboot lost tmux sessions | Starting the daemon restores the kernel by itself when that start finds all of its seats lost; a kernel that was already down before the reboot stays down, and `rig start` names `rig up kernel --existing`. Check `rig status`, and if it reports the restore failed, run `rig up kernel --existing`. For other rigs, open `rig`, start the daemon if needed (press **S** if it opens on the work views), and select the existing rig and seats. Resume is the default; a fresh conversation needs a separate decision. |
 | Launch reports no usable snapshot | Inspect the existing rig and retained project files, then follow the same-seat recovery below. |
 | Work is waiting on a prompt or decision | Read the row, transition and named prompt; preserve the obligation until the missing decision arrives. |
 
+**When the shared kernel dashboard is missing.** The dashboard runs in the
+kernel's `operator.human` terminal seat. Restoring the operator, advisor and
+queue worker does not also restore an unselected terminal. Read
+`rig ps --nodes --rig kernel --json` and use the terminal's registered
+`canonicalSessionName` for the `operator.human` row. Replace `<dashboard-session>`
+in the commands below with that name. Check the actual pane with
+`rig capture "<dashboard-session>"`: a persisted running row alone does not prove
+that its tmux session is alive.
+A daemon or transport error is not proof that the terminal is absent.
+
+- If only the viewing window closed and the TUI is still running, reopen it
+  with `rig terminal open saved:kernel --window`.
+- If the TUI exited but its terminal shell remains, run `rig tui` once in that
+  existing pane. A fresh seat is unnecessary.
+- If the terminal session itself is absent, start only that existing seat:
+
+  ```sh
+  rig seat launch "<dashboard-session>" --fresh --reason "recover missing shared kernel dashboard"
+  ```
+
+  The terminal has no agent conversation to resume. Its fresh-start action
+  runs `rig tui` after the shell is ready; sibling agent sessions stay
+  unchanged. This command intentionally omits `--stop`, so it refuses to
+  replace a live occupant. Check the result before repeating it, then run
+  `rig capture "<dashboard-session>"` to confirm the TUI and
+  `rig ps --nodes --rig kernel --full` to check the seats. Once the dashboard
+  is running, open its view with `rig terminal open saved:kernel --window`.
+
+Opening `saved:kernel` places a view of existing sessions; it cannot create a
+missing terminal. Whole-kernel recovery is for the all-seats-lost case in the
+table, and `rig up kernel --existing` refuses a kernel with live sessions.
+A retained `excluded_from_subset` reason may come from an earlier subset
+restore; inspect its event before attributing it to the latest reboot.
+
 **When the kernel's seats all fail at the first start.** A report on Oracle Linux 10
 ([#980](https://github.com/mvschwarz/openrig/issues/980)) traced this to the distribution's packaged tmux, whose
 `tmux -V` prints `next-3.4`: its server exited on every `capture-pane -p`, also outside OpenRig. `rig setup` (its tmux

docs/reference/harness-runner-contract.md

@@ -0,0 +1,389 @@
+# Harness and runner contract (DRAFT, published with 0.7.6)
+
+> **Status.** This is a draft for contributors.
+> - **Released in 0.7.6:** the `runner` field and OpenRig's side of declaration, identity and activity
+>   (sections 2-4).
+> - **Proposed runner-author behavior:** the input, cancellation and permission recipe in sections 5-7, and the
+>   runner-side shutdown obligation in section 9. A dsh port implementing this recipe is not qualified.
+> - **Existing interfaces with author obligations:** launch-result identity, exact resume, first input and model
+>   handoff. These interfaces do not establish that a particular runner implements them correctly.
+> - **Unchanged:** plain community presets and the built-in runtimes (Claude Code, Codex, Pi, OMP, terminal).
+>
+> No harness has been run end to end under this contract yet. The first candidate is dsh over ACP.
+>
+> Release status and pinned references are in the [joined author guide](author-contracts.md). The
+> [community presets reference](https://github.com/mvschwarz/openrig/blob/b6bd4633fbd94634a2b4bdef2574a6880f97be63/docs/reference/community-harness-presets.md)
+> describes the implemented fields at that base; this chapter also labels proposed author behavior.
+
+This chapter covers **which agent runs in a seat and how OpenRig talks to it**. *Where* that agent runs (a host, a
+sandbox, another machine) is the [execution platform chapter](execution-platform-contract.md). The two are separate on purpose:
+the same harness can run in many places, and one place can host many harnesses.
+
+## 1. Plain preset or runner preset?
+
+| | Plain community preset | Runner preset (`runner: true`) |
+|---|---|---|
+| **What runs in the pane** | The harness's own terminal app | A small program you maintain, *the runner*, which hosts the agent (for example an ACP agent over stdio) |
+| **How OpenRig reads activity** | From the screen: a changing screen reads working; a quiet one reads **unknown**, never idle | From the runner's reports: working, idle, needs input |
+| **How input arrives** | Typed into the app's prompt | Typed into the runner, which turns each submitted message into one agent prompt |
+| **Choose it when** | The harness has a usable terminal UI (Cursor, Hermes, OpenClaw examples) | The harness speaks a protocol rather than a screen (ACP, JSON-RPC, stream-JSON), or you can report turn boundaries reliably |
+
+**If your harness works well as a terminal app, a plain preset is enough. Use a runner only when you can report state
+honestly.** A runner that guesses is worse than a quiet screen read as unknown.
+
+**Vocabulary.** "Preset", "runner preset", `launch_result`, `resume`, `first_input` and `passes_model` are existing
+terms in 0.7.6. Broader names you may see in design discussion, such as *HarnessSpec*, are **proposals**, not file
+formats or loaders that exist.
+
+## 2. Declare it: what the preset says, fixed at launch
+
+```yaml
+# $OPENRIG_HOME/community-harnesses/dsh.yaml
+format: 1
+name: dsh
+command: dsh-seat              # your runner; it starts the agent itself
+maintainer: <you>
+runner: true                   # since 0.7.6
+launch_result: true            # required with runner
+resume: true                   # only after you've checked a real reopen and a refused missing one
+limits:
+  - permission requests go to a person through OpenRig (section 7), and are permitted when no one can be asked
+  - a prompt whose whole text is /abort can't be sent
+  - fork unsupported
+```
+
+- **`runner: true` requires `launch_result: true`,** and can't be combined with `names_conversation`.
+- **The declaration is fixed at launch.** Each launch, a restore's included, records what the preset declares at that
+  moment. Editing the preset changes the next launch, not a running seat.
+- **A preset without `runner` behaves exactly as today,** and so does a session recorded before the field existed.
+- **The other fields keep their existing meanings:**
+  - `first_input: true` passes the seat's first startup turn in a file named by `OPENRIG_FIRST_INPUT`;
+  - `passes_model: true` passes the seat's declared model as `OPENRIG_MODEL`.
+
+## 3. Identity: report the conversation once, in the launch record
+
+- **When:** after the agent has opened or reopened its conversation (for ACP: after `session/new`, `session/load` or
+  `session/resume`).
+- **How:** write the attempt's launch record to the path in `OPENRIG_LAUNCH_RESULT`, with the attempt from
+  `OPENRIG_LAUNCH_ATTEMPT`. The record is either the conversation:
+
+  ```json
+  {"format": 1, "attempt": "<OPENRIG_LAUNCH_ATTEMPT>", "harness": "dsh", "nativeRef": "<conversation id>", "producer": "dsh-seat 0.3.0; dsh 0.2.1-rc.2; ACP 1"}
+  ```
+
+  or the agent's own word that it doesn't exist:
+
+  ```json
+  {"format": 1, "attempt": "<OPENRIG_LAUNCH_ATTEMPT>", "harness": "dsh", "error": "missing", "producer": "dsh-seat 0.3.0"}
+  ```
+
+- **`producer` is free text** for your versions. There are no structured version or capability fields.
+- **The launch record is the only identity channel.** Don't report identity on the activity endpoint.
+- **No record means attention.** For a runner preset:
+  - **a new launch** with no usable record within OpenRig's wait leaves the seat needing attention, not ready. So does a
+    new launch where OpenRig can't create the record's folder: the runner still starts, without `OPENRIG_LAUNCH_RESULT`;
+  - **on a resume,** a folder OpenRig can't create means nothing starts, because the launcher can't be told what to
+    reopen. When the folder exists but no usable record arrives, the resume command has already been sent, and the
+    resume reads unconfirmed (section 8).
+- **OpenRig shows this identity as launcher-reported.** It doesn't observe the conversation itself.
+
+## 4. Activity: report turn boundaries
+
+Post each event to `POST <base>/api/activity/hooks` with `Authorization: Bearer <token>`.
+- **Finding the base and token:** take them from `OPENRIG_URL` (or `OPENRIG_PORT`, with `OPENRIG_HOST` defaulting to
+  `127.0.0.1`) and `OPENRIG_ACTIVITY_HOOK_TOKEN`, or else from `$OPENRIG_HOME/activity-endpoint.json`.
+- **Posting:** post one at a time, in order, with a short timeout. Never make the agent wait on a post.
+
+```json
+{"sessionName": "dev-builder@my-rig", "nodeId": "<node id>", "runtime": "dsh",
+ "generation": "<OPENRIG_OCCUPANT_GENERATION>", "sequence": 42, "occurredAt": "2026-10-29T12:00:00Z",
+ "hookEvent": "UserPromptSubmit", "subtype": "accepted", "turnId": "7"}
+```
+
+| When | `hookEvent` / `subtype` | The seat reads |
+|---|---|---|
+| You gave the agent the prompt for a submitted message | `UserPromptSubmit` / `accepted`, `turnId` = your prompt number | working |
+| The agent is working (message chunks, tool calls) | `active` / `agent_message`; `PreToolUse` / `<tool kind>` | working |
+| You refused a permission request (section 7) | `Notification` / `permission_prompt` | needs input |
+| The turn ended | `Stop` / its stop reason (`end_turn`, `cancelled`, `max_tokens`, `refusal`) | idle. While you hold attention (section 7), don't post it: OpenRig reads any non-exit `Stop` as idle and clears needs input |
+| Optional idle heartbeat | `Stop` / `idle_heartbeat`, with `livenessSeconds` | idle while beats keep coming |
+| The agent process ended | `Stop` / `<preset name>_exited` | unknown, **never idle** |
+| Your runner or the agent hit an error | `Notification` / `runtime_error` | recorded, but not used as activity: the prior activity evidence keeps its usual lifetime. It doesn't itself request or clear attention |
+
+- **Liveness:** state `livenessSeconds` (a whole number of seconds from 1 to 3600; any other value is ignored) on every
+  idle `Stop`, the turn's end as well as each beat. OpenRig stops reading the seat idle after 2 × `livenessSeconds` without
+  a newer idle report. Without it, an idle report holds until your next post, even if your runner has died. Never send a
+  non-exit `Stop` or a beat while holding permission attention.
+- **Exit:** post `Stop` / `<preset name>_exited` only when you observe the agent's process exit.
+
+- **The screen still counts,** but only as evidence of work. On its own it never reads a runner seat idle.
+- **"Accepted" means your runner submitted its Nth prompt.** It isn't the agent's own acknowledgement, and it isn't
+  tied to a particular `rig send`.
+- **The generation attributes a post; it isn't authentication.**
+  - A post carrying an earlier occupant's generation never counts for the live seat.
+  - The hook token is one local token, so any process holding it can post for any seat.
+- **A seat activity grant** is what a runner in another environment, such as a sandbox guest, receives instead of the
+  local token. It arrives in the same `OPENRIG_ACTIVITY_HOOK_TOKEN`, with an `OPENRIG_URL` the guest can reach, and it
+  speaks only for its own seat and launch generation:
+  - with a grant, the generation is authenticated, and you may leave out `nodeId` and `generation`;
+  - a post for another seat is refused with `403 activity_grant_wrong_seat`;
+  - a post claiming another generation, or from an occupant that has since been replaced, is refused with
+    `409 activity_grant_stale`;
+  - a post before the launch has registered its occupant is refused with `409 activity_grant_pending`;
+  - a post with a revoked grant is refused with `401 activity_grant_revoked`, including a post whose grant was revoked
+    while its body was arriving.
+
+  A refused post records nothing, so the seat's activity stays as it was, or unknown.
+
+  **The guest's boundary is its listener, not the grant.** A guest reaches OpenRig only through a guest-only listener that
+  serves this one route and accepts only a grant. It must never reach the daemon's main port. The grant scopes activity
+  posts only, and the main port's other routes don't check it.
+
+## 5. Input: one submitted message is one prompt
+
+Sections 5-7 specify the proposed first dsh/ACP runner recipe. They do not change permission or input handling in
+existing built-in runtimes or require every possible runner protocol to use ACP.
+
+- **Set up the pane:** put your TTY in raw mode and turn on bracketed paste (`\e[?2004h`).
+- **What arrives:** `rig send` and queue reminders paste the whole message, then press Enter. Everything up to that Enter
+  is **one block**, newlines included. Send it to the agent as **one** prompt (for ACP, one `session/prompt`).
+- **Never split a block,** and never act on lines inside it.
+- **Control input:** the only control is a block that is exactly `/abort`, or Ctrl-C typed outside a paste.
+  - Ordinary sends and reminders arrive inside a From/To envelope, so they never equal `/abort`, whatever lines they
+    contain.
+- **A message that arrives mid-turn** waits for the next turn. Report it as queued, never as steering the current turn.
+- **Limits to keep and to list under `limits`:**
+  - a prompt whose whole text is `/abort` can't be sent as a prompt;
+  - a Ctrl-C byte inside a paste also aborts;
+  - without a TTY, input is read line by line;
+  - this isn't a binary-safe channel.
+
+## 6. Cancel: requested isn't the same as done
+
+- **On `/abort` or Ctrl-C:** send the agent's cancel (for ACP, the `session/cancel` notification, which has no reply),
+  and record when you sent it.
+- **Report "cancelled" only when the agent says so:** for ACP, when `session/prompt` returns `stopReason: "cancelled"`.
+  Then post `Stop` / `cancelled`.
+- **Whether the provider stopped work or spend is unknown.** Don't claim it.
+
+## 7. Permissions: a person answers, and the default is permit
+
+### The first version: refuse every request (history; still what `--deny` does)
+
+- **Refuse at once:** answer every permission request immediately with a refusal (for ACP,
+  `{"outcome": {"outcome": "cancelled"}}`).
+- **Make it visible:** show what was asked in the pane, and post `Notification` / `permission_prompt`.
+- **Hold that attention** as your last post through the end of the turn, so the seat reads needs input. Don't post the
+  turn's `Stop` or idle beats while holding it.
+- **It clears** when your next prompt is accepted, or when an operator clears it
+  (`rig seat clear-attention <session> --reason "…"`).
+- **Never leave a request pending.**
+- **What it couldn't do:**
+  - answer from the seat's permission setting;
+  - take answers typed into the pane;
+  - route an approval to another seat.
+
+  So a job that needed an edit or a command approved couldn't finish. The answered path below, added in 0.7.8,
+  replaces this as the default.
+- **Unknown requests from the agent:** answer with JSON-RPC `-32601` and carry on.
+
+### The answered path (0.7.8)
+
+**Only requests the agent actually makes reach anyone.** The harness's own permission mode decides that, and a harness
+that runs without asking gains no approval step. Ordinary conversation and `rig send` are unchanged.
+
+**What happens to a request, in the person's terms:**
+
+| Situation | What happens |
+|---|---|
+| **There's a default person** (chosen with `rig gateway default`, or the only registered person), and no override | They get an ordinary ask in the Feed and on Slack. No setup is needed. |
+| **The seat's operator chose someone** (`rig seat set-permissions-to <seat> --to <person> --reason "…"`) | That person gets the ask instead. |
+| **No one can be asked** | **Permitted at once**, with the agent's own allow option. No ask is opened. That covers: no registered person; several and none the default; the chosen person no longer registered; the people unreadable; or options a person can't be shown (bad ids, or not 2-4). The agent's `allow_once` is used, else its `allow_always`. If the agent offers only `allow_always`, the harness then remembers a permission nobody chose. |
+| **The seat's operator opted in to refusal** (`rig seat set-permissions-to <seat> --deny --reason "…"`) | Refused at once and held as needs input, as in the first version. |
+
+**What the ask shows:**
+- the seat and the conversation;
+- the tool call, as the agent described it;
+- why this person was asked;
+- every option the agent offered, each with its native meaning. `allow_always` means the agent's harness remembers
+  the choice and allows similar requests without asking. OpenRig doesn't record or revoke that.
+
+**How a person answers, and what the agent gets:**
+
+| The person… | The agent gets |
+|---|---|
+| clicks an option: the asked person, any other registered person, or the primary user in the TUI | that option |
+| replies with just a yes ("yes", "ok", "allow", "go ahead" …) | the agent's allow (`allow_once`, else `allow_always`) |
+| replies with just a no ("no", "deny", "reject", "stop" …), or denies it in the Feed | the agent's reject (`reject_once`, else `reject_always`), or, with no reject option, cancelled. A person's no is never a grant |
+| routes or hands it to another person | the ask waits for that person, who gets the same options; their answer counts as if they'd been asked. The handoff itself never permits |
+| routes or hands it to an agent seat | that seat gets no options. The request waits until that row is closed, by the seat or a person. Closed as done without a choice (including a drop), or as failed, is no answer, so it's permitted. Closed as denied (the Feed's deny, `--closure-reason denied`, or the state `denied`) is a no. A row set to the state `canceled` cancels the request, and a further handoff is followed; a person it reaches gets the same options. Or the request waits until the agent's turn ends |
+| closes it without choosing: drop, approve, or other typed words | no answer, so the default applies: **permitted** with the agent's allow |
+| doesn't answer | the request **stays pending and visible** (the seat reads needs input) until the agent's turn ends. It's never timed out into a grant. |
+
+- **An answer reaches only the live request it was asked for,** once. A request from an earlier occupant reads
+  `stale`, and one withdrawn at the turn's end reads `withdrawn`. Neither can take a later answer.
+- **Text typed into the runner's pane, an agent's own text, and `rig send` are never an answer.** While a request waits
+  on a person, the seat reads needs input, and sends are normally refused. A send that does get through reaches the
+  agent as a prompt.
+
+**The runner's protocol:**
+1. **Submit** the agent's live request to `POST <base>/api/activity/permission-requests`, with the same token or seat
+   grant as section 4:
+
+   ```json
+   {"sessionName": "dev-builder@my-rig", "nodeId": "<node id>", "generation": "<OPENRIG_OCCUPANT_GENERATION>",
+    "nativeRef": "<the conversation>", "turnId": "7", "requestId": "<the agent's JSON-RPC id>",
+    "toolCall": {"title": "Edit src/app.ts", "kind": "edit"},
+    "options": [{"optionId": "allow", "name": "Allow", "kind": "allow_once"},
+                {"optionId": "reject", "name": "Reject", "kind": "reject_once"}],
+    "occurredAt": "2026-10-30T12:00:00Z"}
+   ```
+
+   - **A person is asked:** the reply is `{"ok": true, "askId": "…", "created": true}`. OpenRig records the seat's
+     `permission_prompt`, so it reads needs input. Don't post it yourself.
+   - **No one can be asked:** the reply is `{"ok": true, "state": "permitted", "optionId": "…", "reason": "…"}`.
+     Deliver that option.
+   - **`requestId` and `turnId`** may be strings or numbers. Pass the agent's id unchanged: a number is read as its
+     decimal string, so `7` and `"7"` are the same request.
+   - **Resubmitting** the same generation, conversation and request id returns the same `askId`.
+2. **Read** the answer with `GET <base>/api/activity/permission-requests/<askId>?waitSeconds=25` (25 at most), and read
+   again until the agent's turn ends.
+   - **The `state`** is `pending` (read again, don't cancel), `answered` or `permitted` (each with `optionId`),
+     `refused` (with `reason`: `said_no` when the person said no and the agent offered no reject option, or
+     `no_allow_offered` when it was closed without a choice and the agent offered no allow option), `withdrawn` or
+     `stale`.
+   - **A seat grant reads and withdraws only its own launch's requests.** Once its occupant is replaced, the grant is
+     refused (`409 activity_grant_stale`), even during a read's wait.
+3. **Deliver** `{"outcome": {"outcome": "selected", "optionId": "<optionId>"}}` to the agent once, and only while that
+   request is still its live one. For `refused`, `withdrawn` or `stale`, answer `cancelled` and hold no attention: no
+   person is still waiting. A refusal of the read itself follows the refusals below.
+4. **Withdraw** when the turn is cancelled or ends, or the agent exits:
+   `POST <base>/api/activity/permission-requests/<askId>/withdraw` with
+   `{"reason": "turn_cancelled" | "turn_ended" | "exited"}`.
+   - A later answer isn't sent.
+   - A relaunch makes the earlier occupant's open requests stale on its own.
+
+**Refusals, at submit or while reading.** One rule covers both. Cancel only a request that certainly isn't the current
+occupant's live one, and hold no attention. Anything that may still be a live request permits, as when OpenRig can't
+answer. Only the seat's opt-in refuses and holds attention.
+
+| Code | Status | Meaning | The runner |
+|---|---|---|---|
+| `denied_by_seat` | 409 | The seat's operator opted in to refusal (`--deny`). | refuses and holds attention |
+| `options_unsupported` | 422 | No one can be asked, and the agent offered no allow option to permit. | refuses and holds attention |
+| `stale_generation` | 409 | With the activity token: the post isn't from the seat's current occupant. | answers `cancelled` |
+| `activity_grant_wrong_seat` | 403 | The grant speaks for another seat. | answers `cancelled` |
+| `activity_grant_stale` | 409 | The grant's launch generation no longer occupies the seat. | answers `cancelled` |
+| `activity_grant_revoked` | 401 | The grant was revoked while its launch generation still occupies the seat, or before it has registered, so this may still be the live request; a replaced occupant's grant reads `activity_grant_stale` instead, even once revoked. | permits |
+| `activity_grant_pending` | 409 | The grant's launch hasn't registered its occupant yet. | permits |
+| `duplicate_request` | 409 | This request id was already submitted in another turn; an agent may reuse an id. | permits |
+
+**When the runner can't get an answer from OpenRig,** nobody can be asked, so it permits with the agent's own allow
+option: `allow_once` when offered, else `allow_always`. Prefer `allow_once`, so the harness doesn't remember a
+permission nobody chose. That covers:
+- an OpenRig without this path, which answers 404;
+- an OpenRig the runner can't reach, at submit or while reading (a runner may retry a read that got no answer a few
+  times first);
+- a refusal the table marks "permits", and any status or code this section doesn't list.
+
+**Limits in 0.7.8:**
+- **The answer owner is a registered person.** Choosing a seat (another agent) to answer is a later change.
+- **A colleague's answer is followed on this host.** If the ask is handed to another host, the request waits, shown as
+  needs input, until the agent's turn ends.
+- **An override or `--deny` is set with the command above, never in a rig spec.** A spec field is a possible later
+  change.
+- **Permits and answers aren't recorded beyond the runner's pane and the ask itself.** A permit with no one to ask
+  opens no ask.
+
+## 8. Resume: exactly the same conversation, or say honestly that you couldn't
+
+- On a resume, OpenRig passes the conversation to reopen in `OPENRIG_RESUME_REF`. Use the method your agent
+  advertises: ACP's `session/load` (capability `loadSession`), ACP's `session/resume` (capability
+  `sessionCapabilities.resume`, which dsh `0.2.1-rc.2` uses), or your agent's own reopen method, named with its
+  capability and pinned version in your `limits`.
+- **Telling "missing" from "won't reopen":** an agent may answer one error for both. dsh `0.2.1-rc.2` answers `-32602`
+  for a conversation it doesn't have and for one already active elsewhere. So don't read an error's code or text as
+  missing unless the agent defines it that way, as ACP's not-found `-32002` is.
+  - **Report missing only on the agent's own not-found,** and leave anything else unconfirmed.
+  - **When the agent advertises `sessionCapabilities.list`,** its `session/list` can show that a conversation exists:
+    one it lists but won't reopen is unconfirmed, even on a not-found. Follow `nextCursor` (an opaque string, possibly
+    empty), and treat a page you can't read as deciding nothing.
+  - **A list doesn't show that a conversation is gone,** unless your agent guarantees it lists every conversation. An
+    agent may leave out one that another running process holds unsaved; dsh `0.2.1-rc.2` can. So still try the reopen
+    for a conversation the list doesn't name.
+
+| What happened | Write in the launch record | What OpenRig does |
+|---|---|---|
+| The same conversation reopened | the same `nativeRef` | resumed |
+| The agent says it doesn't exist | `error: "missing"` | reports it missing; nothing new starts |
+| Anything else: a timeout, an error, or the working folder no longer matches the conversation's | no record, or an error other than `missing` | **unconfirmed**: the seat keeps its conversation and needs attention |
+| A different conversation opened | the other id | a mismatch; OpenRig doesn't switch to it |
+| OpenRig couldn't create the record's folder | (nothing to write) | nothing starts |
+
+- **Never start a new conversation on a resume.**
+- A conversation saved under another harness's runtime is never handed to your launcher.
+
+## 9. Context, model and shutdown: whose job is what
+
+| Responsibility | OpenRig does | Your preset or runner does |
+|---|---|---|
+| **First startup turn** | Writes it to a private file and passes `OPENRIG_FIRST_INPUT` (with `first_input: true`) | Gives those exact bytes to the agent through its own first-prompt route |
+| **Guidance, skills, startup files** | Doesn't place them for community harnesses; a launch warns that they weren't delivered | Puts whatever the agent needs where *your* harness reads it, or says in `limits` that it isn't supported |
+| **Model** | Passes the seat's declared model as `OPENRIG_MODEL` (with `passes_model: true`); never substitutes one | Maps it to the harness's own setting. OpenRig shows the effective model as unknown |
+| **Placing the agent somewhere else** (a sandbox, another machine) | See the [execution platform chapter](execution-platform-contract.md) | Keeps the runner independent of where it runs |
+| **Shutdown** | `rig seat stop` and `rig down` end the seat's tmux session at once: no graceful stop handshake reaches the runner, its pane just closes. OpenRig ends nothing outside that session, and does no external provider cleanup | **(Proposed)** When the runner's pane closes (a hangup) or it's asked to stop, it attempts where possible to cancel any running turn and end the agent process it started, and lists in `limits` anything that may keep running. Cleanup of external compute needs the platform owner to reconcile it |
+
+## 10. What counts as evidence
+
+- **Keep three things separate:**
+  - **`rig send --verify`** checks the pane before and after a send. It's pane evidence, not proof the agent received
+    or finished anything. Mirror each accepted prompt in your pane (for example `you ▸ …`); OpenRig doesn't yet promise
+    that `--verify` recognises it.
+  - **A turn ending isn't the work being done.** Task completion is whatever the work's own record says: a queue task
+    marked done, or an artifact written.
+  - **Your own tests** should run against a fake agent that covers each section above. The
+    [runner conformance check](runner-conformance/README.md) does that for you: point it at your runner, and it reports
+    which promises your runner keeps.
+- **The useful real check** for a maintainer:
+  - a seat starts and reads ready only after the handshake;
+  - one multi-line message becomes one prompt;
+  - the seat reads working, then idle;
+  - a refused permission reads needs input;
+  - `/abort` ends a turn as cancelled;
+  - a resume reopens the same conversation, and a missing one is refused, never replaced.
+
+  That's the same promotion test the presets reference describes, applied to these declarations. Passing it is
+  compatibility evidence for the versions you name, not certification.
+
+## Contributor checklist
+
+- [ ] The preset sets `runner: true` and `launch_result: true`, and not `names_conversation`.
+- [ ] After the agent opens or reopens its conversation, the runner writes the launch record, with versions in `producer`.
+- [ ] Activity posts carry `sessionName`, `nodeId`, `runtime`, `generation`, `sequence`, `occurredAt`, `hookEvent` and
+      `subtype`. They're sent in order and never block the agent.
+- [ ] One submitted message becomes exactly one prompt. Only a bare `/abort` block, or Ctrl-C outside a paste, is control.
+- [ ] Cancel records "requested" when sent, and "cancelled" only on the agent's own stop reason.
+- [ ] Every permission request goes to OpenRig's answered path, and the runner delivers the option that comes back: the
+      person's, or OpenRig's permit. When OpenRig can't answer (404, unreachable, or a refusal section 7's table marks
+      "permits"), the runner permits with the agent's `allow_once`, else `allow_always`. A refusal that says the request
+      isn't the current occupant's, or a `refused` answer, means answer `cancelled` and hold nothing. Only
+      `denied_by_seat`, or no allow option to permit, means refuse at once, show it, and hold needs input until the next
+      accepted prompt.
+- [ ] An agent exit posts `Stop` / `<preset name>_exited`, never a plain idle.
+- [ ] A resume reopens the exact conversation. It reports `missing` only on the agent's word, and otherwise leaves it
+      unconfirmed.
+- [ ] Unknown agent requests get `-32601`.
+- [ ] `limits` names the input limits, the permission limit, what keeps running after a stop, and anything not supported.
+
+## Open questions
+
+1. **Answered:** which method and capability does dsh advertise for reopening a session, at a pinned version? dsh
+   `0.2.1-rc.2` advertises `sessionCapabilities` `{close, list, resume}`, without `loadSession`, and reopens with
+   `session/resume` (section 8). This was observed with a stand-in model, not a real one.
+2. **For the dsh maintainer:** can dsh's ACP profile run with tools off? A first journey that needs no permission
+   answers is the simplest to check.
+3. **For OpenRig:** should a later version route permission requests to a chosen seat, as `rig acp run` already does
+   with queue tasks? 0.7.8 routes them to the default person, or to a person chosen for the seat (section 7). A chosen
+   seat is still open.
+4. **For OpenRig:** once a second ACP harness exists, should one shared ACP runner live in OpenRig rather than one per
+   harness?

docs/reference/hosted-members.md

@@ -3,6 +3,10 @@
 A hosted member is a tool that lives outside a terminal, such as an assistant, a chat bot or a cloud routine, and works as a member of a rig. You
 address work to it like any seat. It reaches the team through OpenRig's MCP server under a grant you issue and can revoke. It never gets a pane.
 
+Assistants that run on your machine can reach OpenRig's MCP server locally. Many assistants live in a vendor's cloud, and those can only reach a server on the internet.
+
+> **Experimental and unsupported. Off by default.** Nothing in OpenRig listens on the internet unless you set it up yourself. You serve `rig mcp serve --http`, put it behind your own TLS tunnel or proxy, and issue a token. Doing that is like opening a port in your firewall. Anyone with the URL and a valid token can act as that member within its grant. Treat the token like a password, give the member the narrowest grant, and revoke it when you're done (`rig mcp token revoke <name>`). We don't support, and take no responsibility for, an OpenRig reachable from the internet, just as with running OpenRig itself on the open internet. Use at your own risk.
+
 ## Set one up
 
 1. **Pick its address** on an existing rig: `<member>@<rig>`, for example `grok@research`. No node is added; the queue takes rows for it.
@@ -52,7 +56,7 @@ member, so it checks its rows again.
 
 The server tells the member this in its MCP instructions. The text is versioned, so you can check which version a tool received:
 
-> Member workflow (member-workflow/v1). Your inbox is the items addressed to your seat: rig_queue_list with owned set. It's yours, not the team's queue. A wake, a ring or a message only points at it: list your inbox, then act. For each item: claim it (rig_queue_claim), do it, then close it done with the answer in resolution (rig_queue_update), or hand it on (rig_queue_handoff). Listing or reading an item doesn't take it. If the person connected you, that's basically authorized: claim the items in your inbox and do them without asking whether you may. When an item needs a person's approval before you can finish it, don't close it and don't take the action: keep the item and ask. If it came from a person (an address ending in @external that isn't a shared channel; the item's 'Shared channel:' line names one), set it blocked with blockedOn that person and, as its summary, the question with a short form of what you'll do: the person sees the summary and the item, not your note, which keeps the record. Link a longer draft as an https evidenceRef. Their answer returns the item to you in progress. Otherwise ask as a new item, to the seat it came from or, for a shared channel's item, to your rig's lead (rig_team shows it), with your question and draft. Then set yours blocked with blockedOn that new qitemId: when that item closes, yours returns to pending, to claim again. Nothing promises to wake you, so check your inbox again. Then finish the same item; don't open a new one.
+> Member workflow (member-workflow/v2). Your inbox is the items addressed to your seat: rig_queue_list with owned set. It's yours, not the team's queue. A wake, a ring or a message only points at it: list your inbox, then act. For each item: claim it (rig_queue_claim), do it, then close it done with closure reason no-follow-on and the answer in resolution (rig_queue_update), or hand it on (rig_queue_handoff). Listing or reading an item doesn't take it. Say what you can reach from your grant, not from the tool list: a listed tool isn't access. rig_artifact_read's answer for an item names the shared roots you may read, and none means no shared files. If the person connected you, that's basically authorized: claim the items in your inbox and do them without asking whether you may. When an item needs a person's approval before you can finish it, don't close it and don't take the action: keep the item and ask. If it came from a person (an address ending in @external that isn't a shared channel; the item's 'Shared channel:' line names one), set it blocked with blockedOn that person and, as its summary, the question with a short form of what you'll do: the person sees the summary and the item, not your note, which keeps the record. Link a longer draft as an https evidenceRef. Their answer returns the item to you in progress. Otherwise ask as a new item, to the seat it came from or, for a shared channel's item, to your rig's lead (rig_team shows it), with your question and draft. Then set yours blocked with blockedOn that new qitemId: when that item closes, yours returns to pending, to claim again. Nothing promises to wake you, so check your inbox again. Then finish the same item; don't open a new one.
 
 ## Ring it when work lands
 

docs/reference/pi-tool-selection-recovery.md

@@ -0,0 +1,83 @@
+# Pi tool-selection recovery
+
+Use this when a live Pi conversation has fewer active tools than its operator
+currently intends. This does not establish why the selection changed or that
+stock Pi/OpenRig lost tools. The motivating report used a customized Pi 1.4.0
+deployment; it is not a stock cross-platform reproduction.
+
+## Diagnose before changing the selection
+
+Confirm the exact seat, occupant and conversation, then use an idle boundary.
+Preserve pending work and reconcile unknown writes before any replacement.
+Keep four things separate:
+
+- **Configured:** the effective merged `defaultTools` setting. Do not infer it
+  from one settings file; retain any unresolved patterns or modifiers as such.
+- **Currently intended:** the tools the operator wants enabled now, including
+  deliberate exclusions. Configuration alone does not establish that intention.
+- **Registered:** names the current Pi runtime knows about.
+- **Active:** names returned by `getActiveTools`, which can differ from both.
+
+A live RPC connection does not prove the tools are usable. Record observation
+time and conversation binding; an unavailable query stays unknown. Reloading
+unchanged settings can retain session-disabled tools by design, so a reload
+that restores nothing is not itself evidence of accidental loss.
+
+## Choose the interface the seat actually has
+
+Pi 1.4.0 includes an opt-in selector at `examples/extensions/tools.ts`. Its
+`/tools` command remembers selection per branch and needs Pi's own TUI. Use it
+where that extension and native TUI are available. **Do not send `/tools` to an
+agent OpenRig drives over RPC**: that interface cannot operate the selector.
+
+For RPC seats, an optional [community-maintained helper][helper] provides
+read-only status and an explicit command that adds only named intended tools.
+It is maintained outside OpenRig, is not bundled, and this link is not a support
+claim. Loading it also registers the agent-callable `openrig_runtime_status` and
+`openrig_slots` tools; the latter reads a snapshot of the bound rig's agents and
+active queue items. It can add a prompt section when a guidance file is set.
+Review its [shared implementation][runtime] and
+[setup/capability limits][setup] before opting in through Pi's per-seat extension
+configuration. Keep the two helper files together; do not install the separate
+RepoStew application just to use the neutral bridge. Verify what the target
+session actually loaded after the supported activation path.
+
+With that helper already loaded, these are **Pi extension commands**, not new
+`rig` commands:
+
+```text
+/openrig-status
+/openrig-restore-tools read
+```
+
+For an agent OpenRig drives over RPC, send each extension command with
+`rig send <agent> --raw "<command>"` so it arrives without a messaging envelope;
+the usual interactive-prompt guard still applies.
+
+The second command means that `read` is intended now. It does not restore every
+configured default. If `bash` was deliberately disabled, leave it disabled
+unless the operator explicitly intends to add it too. With no names, this helper
+refuses to change the selection.
+
+## Verify the recovery
+
+1. Establish the explicit current intended names. Reject unregistered or hidden
+   requested tools before changing anything; do not infer replacement names.
+2. Preserve current active extras: `setActiveTools` **replaces** the selection,
+   so pass the union of current active names and only the requested names. It can
+   ignore names it cannot activate; sending the command is not proof of recovery.
+3. Read `getActiveTools` back. Retain a before/after receipt with the intended
+   names and any missing names, then make a harmless actual call to a recovered
+   tool within the existing task authority. Listed-active is not proof of usable
+   permissions or working execution.
+
+This changes tool selection, not permission policy, task authority or conversation
+identity. It does not automatically persist a new branch selection; verify again
+after branch navigation or reload. If the runtime cannot support or verify this
+in place, report that capability gap and follow its supported replacement path
+with retained work. Do not manufacture a fresh conversation or claim that another
+harness has Pi's capabilities.
+
+[helper]: https://github.com/dajiaohuang/repostew/blob/adf9f36c23e27cf40cf8dd2dae887972a0dc776b/skill/integrations/pi-openrig.js
+[runtime]: https://github.com/dajiaohuang/repostew/blob/adf9f36c23e27cf40cf8dd2dae887972a0dc776b/skill/integrations/pi-runtime.js
+[setup]: https://github.com/dajiaohuang/repostew/blob/adf9f36c23e27cf40cf8dd2dae887972a0dc776b/skill/references/harnesses/pi.md

docs/reference/publishing-a-rig-bundle.md

@@ -120,6 +120,11 @@ or **"Not tested by OpenRig"**, which means exactly that. You're welcome to run
 
 ## 6. Submit it to openrig.dev/rigs
 
+Before you submit, run `rig bundle check` on the folder at the commit you'll submit. openrig-registry's registry check
+runs the same check, with the latest published OpenRig, on the folder at your `ref`, and any finding fails it. A team
+made with `rig export` has no README of its own: add one and declare it in `rig.yaml` as `docs: [{ path: README.md }]`
+(see [section 1](#1-lay-out-the-folder)).
+
 Open a pull request to [mvschwarz/openrig-registry](https://github.com/mvschwarz/openrig-registry) that adds one file,
 `registry/submissions/<your-team>.yaml`, with three fields: `repository` (your GitHub repository URL), `folder` (the
 folder holding `rig.yaml`, or `.` for the repository root) and `ref` (a branch, tag or commit). No pull requests? Open

docs/reference/quickstart.md

@@ -17,7 +17,7 @@ npx skills add https://github.com/mvschwarz/openrig/tree/<release-tag>/skills/ri
 npx skills add https://github.com/mvschwarz/openrig/tree/<release-tag>/skills/rigs --skill rigs --agent codex
 ```
 
-For a new install, use the latest release's tag, for example `v0.7.0`. If OpenRig is already installed, use the tag
+For a new install, use the latest release's tag, for example `v0.7.5`. If OpenRig is already installed, use the tag
 that matches `rig --version`, or upgrade first. If the command fails with `npx: command not found`, Node.js isn't
 installed or isn't on your PATH yet (see [What you need](#what-you-need)). Then type `/rigs` in Claude Code, or `$rigs` in Codex, and say what
 you want built.

docs/reference/rig-spec.md

@@ -511,9 +511,9 @@ pods:
 **Where the declared names reach, by runtime.** Declared values are handed to the seat's terminal as tmux environment, so most seats take them as given:
 
 - Plain Claude Code (no permission mode selected), Codex and terminal seats take a spec `HOME`, `PATH` or `CLAUDE_CONFIG_DIR` as given.
-- The cleared environment (`env -i`) belongs only to a managed Claude launch, which is the one used when a permission mode is selected for the seat (for example `permission_policy: builtin:auto`). It forwards the declared names through the cleared environment, except `PATH`, `HOME` and `CLAUDE_CONFIG_DIR`, which OpenRig sets for that launch itself, and `CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN`, for which it sets its own renderer setting. A spec value for any of those four is not forwarded as such; an absolute `CLAUDE_CONFIG_DIR` becomes the config dir OpenRig sets (see "A seat's own account" below). A seat that selected the shipped `openrig-core` plugin also has `CLAUDE_CODE_PLUGIN_DIRS` (the vendored plugin, added after a list the seat declares itself) among its declared names, so it is forwarded like the others and the plugin's hooks load in a seat with a permission mode as in one without; a seat that didn't select the plugin has no such name added. Beyond the declared names, the launch also forwards, by name and only when the seat's terminal has them set, the terminal and locale variables (`TERM`, `COLORTERM`, `TERM_PROGRAM`, `TERM_PROGRAM_VERSION`, `LANG` and the `LC_*` names) and the person's editor and display session: `EDITOR`, `VISUAL`, `DISPLAY`, `XAUTHORITY` and `WAYLAND_DISPLAY`. They let Claude's external-editor action open the person's editor and let image paste reach the clipboard, without a value in a spec. `WAYLAND_DISPLAY` is passed as an absolute path: an absolute value is kept, and a relative one is joined to the terminal's `XDG_RUNTIME_DIR`; if the terminal has no `XDG_RUNTIME_DIR` for it, the name is left out. The runtime directory and its `disabled:` bus address (see "The runtime directory, on Linux" above) are forwarded together, and the runtime directory is also forwarded beside a declared `DBUS_SESSION_BUS_ADDRESS`. A terminal's own `XDG_RUNTIME_DIR` or `DBUS_SESSION_BUS_ADDRESS`, from a shell startup file or tmux's `update-environment`, is not, because D-Bus clients look for the session bus at `$XDG_RUNTIME_DIR/bus` when no address is set; declare it in `env:` to forward it. `XDG_SESSION_TYPE`, `XDG_CURRENT_DESKTOP` and `SSH_AUTH_SOCK` are not forwarded. A running seat picks the names up at its next managed Claude launch, not before: `rig seat launch <seat> --fresh --stop --reason "<why>"` for that seat alone (a new conversation), or a restore of the whole rig (`rig down <rig>`, then `rig up <rig> --existing`, which resumes each conversation but restarts every seat in the rig), or a handover. A seat that is up keeps the environment it started with.
+- The cleared environment (`env -i`) belongs only to a managed Claude launch, which is the one used when a permission mode is selected for the seat (for example `permission_policy: builtin:auto`). It forwards the declared names through the cleared environment, except `PATH`, `HOME` and `CLAUDE_CONFIG_DIR`, which OpenRig sets for that launch itself, and `CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN`, for which it sets its own renderer setting. A spec value for any of those four is not forwarded as such; an absolute `CLAUDE_CONFIG_DIR` becomes the config dir OpenRig sets (see "A seat's own account" below). A seat that selected the shipped `openrig-core` plugin also has `CLAUDE_CODE_PLUGIN_DIRS` (the vendored plugin, added after a list the seat declares itself) among its declared names, so it is forwarded like the others and the plugin's hooks load in a seat with a permission mode as in one without; a seat that didn't select the plugin has no such name added. Beyond the declared names, the launch also forwards, by name and only when the seat's terminal has them set, the terminal and locale variables (`TERM`, `COLORTERM`, `TERM_PROGRAM`, `TERM_PROGRAM_VERSION`, `LANG` and the `LC_*` names) and the person's editor and display session: `EDITOR`, `VISUAL`, `DISPLAY`, `XAUTHORITY` and `WAYLAND_DISPLAY`. They let Claude's external-editor action open the person's editor and let image paste reach the clipboard, without a value in a spec. `WAYLAND_DISPLAY` is passed as an absolute path: an absolute value is kept, and a relative one is joined to the terminal's `XDG_RUNTIME_DIR`; if the terminal has no `XDG_RUNTIME_DIR` for it, the name is left out. The runtime directory and its `disabled:` bus address (see "The runtime directory, on Linux" above) are forwarded together, and the runtime directory is also forwarded beside a declared `DBUS_SESSION_BUS_ADDRESS`. A terminal's own `XDG_RUNTIME_DIR` or `DBUS_SESSION_BUS_ADDRESS`, from a shell startup file or tmux's `update-environment`, is not, because D-Bus clients look for the session bus at `$XDG_RUNTIME_DIR/bus` when no address is set; declare it in `env:` to forward it. `XDG_SESSION_TYPE` and `XDG_CURRENT_DESKTOP` are not forwarded. The SSH agent socket (`SSH_AUTH_SOCK`) is forwarded when OpenRig itself was started with one, so git over SSH works in the seat; a socket only the seat's terminal has stays out. To check, run `ssh-add -l` inside the seat. If it can't reach an SSH agent, start the OpenRig daemon from a terminal where `ssh-add -l` works. A running seat picks the names up at its next managed Claude launch, not before: `rig seat launch <seat> --fresh --stop --reason "<why>"` for that seat alone (a new conversation), or a restore of the whole rig (`rig down <rig>`, then `rig up <rig> --existing`, which resumes each conversation but restarts every seat in the rig), or a handover. A seat that is up keeps the environment it started with.
 - Codex seats run OpenRig's own Codex executable whatever a spec `PATH` says. They use the seat's own `CODEX_HOME` when its `env:` declares an absolute one, else the daemon's when the daemon has one.
-- On Pi and OMP seats the values reach the terminal. The agent process gets only what the runner passes on, so a spec or one-off value reaches it only for the names the runner always passes: `PATH`, `HOME`, `USER`, `LOGNAME`, `TERM`, `LANG`, `LC_ALL`, `SHELL` and `TMPDIR`. Pi's agent process receives the runner's `HOME`, so a spec or one-off `HOME` value that reaches the runner is passed to Pi; OMP's agent process gets its seat state directory (`$OPENRIG_HOME/state/omp/<seat>`, see Oh My Pi below) as `HOME`, so a `HOME` value does not reach it. The selected provider's variables are passed separately, as described under each runtime.
+- On Pi and OMP seats the values reach the terminal. The agent process gets only what the runner passes on, so a spec or one-off value reaches it only for the names the runner always passes: `PATH`, `HOME`, `USER`, `LOGNAME`, `SSH_AUTH_SOCK`, `TERM`, `LANG`, `LC_ALL`, `SHELL` and `TMPDIR`. Pi's agent process receives the runner's `HOME`, so a spec or one-off `HOME` value that reaches the runner is passed to Pi; OMP's agent process gets its seat state directory (`$OPENRIG_HOME/state/omp/<seat>`, see Oh My Pi below) as `HOME`, so a `HOME` value does not reach it. The selected provider's variables are passed separately, as described under each runtime.
 
 **A seat's own account.** Claude Code keeps its login in its config dir and Codex in its home, so a seat whose `env:`
 declares an absolute `CLAUDE_CONFIG_DIR` (Claude Code) or `CODEX_HOME` (Codex) runs on whatever login that directory
@@ -526,6 +526,105 @@ only the stored `env:`. Seats may share a home. Live usage refresh (`rig provide
 daemon's own logins; reading each seat's home would need a native reader per home. OpenRig doesn't copy logins or conversation history between homes, so a conversation saved under a seat's earlier home
 doesn't resume after the seat moves to another one. The `openrig-user` skill has the setup steps.
 
+**A seat's own subscription token.** A Claude member can start on a prepared subscription token from its first launch,
+with no conversation to switch first. Name the token file beside its config dir:
+
+```yaml
+- id: dev
+  runtime: claude-code
+  env:
+    CLAUDE_CONFIG_DIR: /abs/claude-dev
+  claude_token: { label: dev-sub, file: /abs/tokens/dev.token }
+```
+
+The label and the absolute path are stored on the seat when the rig is created, as `rig provider switch --seat <seat>
+--to-account <label> --token-file <path>` stores them. Its launches run the same token step, which reads the file as
+Claude starts; a launch path that can't run it (a pane shell other than a POSIX one, for example) starts without it and
+warns. The file's contents never pass through OpenRig: validation doesn't read it, and they don't reach the spec, the
+seat's record or the launch command. A missing, empty or unreadable file doesn't refuse the spec; the seat starts on its
+config dir's stored login (or a token the daemon gives every seat) and says so. A launch also warns when a setting Claude
+Code ranks above the token would be used instead: an `apiKeyHelper`, or an outranking `env` entry in the user, project
+or local settings. Managed settings aren't inspected. `rig export` writes the seat's current reference, except for a seat
+with an account home or an API-key selection (select those again after import), so a later switch shows there, and `rig provider switch --stored-login` removes it. Only
+`claude-code` members and legacy flat nodes accept it, and there's no rig-level default.
+
+**Switching to an isolated account home.** A running Codex or Claude Code seat can change its native account home
+while staying in its rig:
+
+```sh
+rig provider switch --seat dev-impl@my-rig --to-account api-account \
+  --account-home /home/me/accounts/codex-api --model gpt-5.5 --reason "Use the API budget" --dry-run
+rig seat set-runtime dev-impl@my-rig --runtime claude-code \
+  --account-home /home/me/accounts/claude-api --model claude-sonnet-4-6 --reason "Codex limit"
+```
+
+The directory must already exist and contain the runtime's login or API settings. You choose the account and budget;
+OpenRig does not choose or create credentials. `--account-home` sets `CODEX_HOME` for Codex or `CLAUDE_CONFIG_DIR` for
+Claude Code before the successor starts.
+
+This switch starts a fresh conversation with a handover, even when the runtime stays the same. The existing
+subscription-login switch preserves the conversation. Once the successor confirms, OpenRig commits its home, runtime,
+model, profile and permission selection together. The seat keeps its address, edges, queue rows and claims. If OpenRig
+refuses the switch before cutover, the predecessor's configuration stays unchanged.
+
+For a Claude plan authenticated with `setup-token`, supply the token file with the home:
+
+```sh
+rig provider switch --seat dev-impl@my-rig --to-account claude-plan \
+  --account-home /home/me/accounts/claude-plan --token-file /home/me/credentials/claude-plan-token \
+  --reason "The plan reset"
+# Across runtimes, use set-runtime with --account-home, --token-file and --account-label.
+```
+
+To return to the seat's configured login:
+
+```sh
+rig provider switch --seat dev-impl@my-rig --clear-account-home --reason "The plan reset" --dry-run
+rig provider switch --seat dev-impl@my-rig --clear-account-home --reason "The plan reset"
+# Across runtimes: rig seat set-runtime dev-impl@my-rig --runtime claude-code --clear-account-home --reason "Return to configured login"
+```
+
+Clearing the selection also starts a successor with a fresh conversation and a handover. OpenRig removes the saved
+account-home override and its token reference only after the successor confirms. The successor starts with the member's
+or rig's configured home, falling back to the daemon's selection and then the native default, as an ordinary launch does.
+If there was no account-home override, OpenRig preserves any unrelated legacy per-seat token. Clear uses the current
+configuration; it does not remember a prior login or restore old credentials. Use clear on its own, without
+`--to-account`, `--stored-login`, `--account-home` or a new token reference.
+
+**Anthropic profile limitation.** An active or default profile in Anthropic's configuration directory can override a
+stored Claude login. `CLAUDE_CONFIG_DIR` does not isolate that directory, and OpenRig does not change `ANTHROPIC_CONFIG_DIR`
+or rewrite global profiles. Check or deselect the profile before using a stored-login account home, and verify the
+running account privately.
+
+An isolated Claude launch removes `ANTHROPIC_PROFILE`, `ANTHROPIC_FEDERATION_RULE_ID` and `ANTHROPIC_ORGANIZATION_ID`
+from its inherited environment. It also refuses profile selectors in the selected home's or project's settings `env`
+blocks. An explicit subscription token takes precedence over these selectors, including in the legacy
+subscription-token switch.
+
+OpenRig receives only the token's label and file path. The launch reads the token; if the required file is unusable, no agent
+starts. Selecting an API home without `--token-file` clears an older seat token reference when the selection commits.
+The home's API helper can stay in its user settings. For a selected plan token, OpenRig refuses a home whose API helper
+or credential environment would take precedence over that token. Both modes refuse project credential overrides.
+OpenRig leaves those settings unchanged.
+
+The selected home persists across managed launches and restores. These launches exclude inherited Claude/OpenAI
+credentials and routing variables so credentials from the daemon or shell cannot override the selected home. OpenRig
+refuses to launch if it cannot read the saved selection or the shell cannot enforce isolation. Launches without a home
+selection and existing subscription-login switches keep their behavior. A Claude seat using account-home mode must
+use that mode for later account changes too.
+
+User-managed provider settings can still affect authentication. After a real tool turn, check which account and billing
+identity the runtime used. Do not print tokens or the process environment. The configured home alone cannot verify the
+running account.
+
+Before downgrading to a version without account-home switching, stop the managed seats. Select their accounts using
+that version's supported setup before resuming. Older versions ignore the saved selection and may launch with inherited
+or default credentials. Rolling back the binary alone does not preserve the selected billing account.
+
+OpenRig leaves the imported team file unchanged. Exporting and reimporting a team does not preserve account-home
+isolation, so configure it explicitly for each new rig. If a switch is interrupted, inspect it with `rig seat status`
+and follow the printed recovery command. Wait until the outcome is known before starting another switch.
+
 **Do not put secrets in `env:` or `--env`.** The values are not secret, and nothing here encrypts or hides them (see #306):
 
 - Spec values are stored in OpenRig's database and in rig snapshots, and `rig export` and bundles copy them, so a secret put there by mistake travels with the rig.
@@ -534,6 +633,52 @@ doesn't resume after the seat moves to another one. The `openrig-user` skill has
 
 Use `env:` for settings only. Credentials belong in the provider's own login or a secret store the seat reads at run time.
 
+### Switching credentials while keeping the conversation
+
+`rig provider switch` resumes a running seat in its existing home and pane. A stopped Claude Code
+seat isn't relaunched: the switch saves the new reference for its next launch, but only when the
+launch it predicts resumes the seat's conversation. That prediction covers the snapshot that
+launch would use and restore's automatic periodic carry. It then prints the launch that resumes:
+`rig launch <rig> <seat> --snapshot-id <id>` for a saved conversation, or a plain
+`rig launch <rig> <seat>` when the resume comes through the periodic carry, which naming a
+snapshot would turn off. Otherwise it refuses and changes nothing. `rig provider precheck` gives
+the same answer.
+For Claude Code, select an API-key file, a subscription-token file, or the home's stored login:
+
+```sh
+rig provider precheck --seat dev-a@my-rig --to-account api-budget --api-key-file /secure/claude-api-key
+rig provider switch --seat dev-a@my-rig --to-account api-budget --api-key-file /secure/claude-api-key
+rig provider switch --seat dev-a@my-rig --to-account plan --token-file /secure/claude-plan-token
+rig provider switch --seat dev-a@my-rig --stored-login
+```
+
+The API-key file must be private (0600 or 0400), nonempty, and at most 8192 bytes, containing
+one key with optional surrounding whitespace. Only the reference is stored. The child reads the
+key at launch; key bytes never enter command arguments, snapshots or switch reports. Classic
+launches require a supported POSIX pane shell. Missing/invalid API-key files prevent launch.
+Settings that could override the key cause a refusal; OpenRig does not edit those settings.
+Native managed settings and a signed-in Claude gateway cannot be ruled out; Claude may require API-key approval before its
+first turn. Check `/status` after a real tool turn to confirm the authentication method.
+
+For Codex, a saved native API-key login is selected with the existing `--to-account <name>`
+path, exactly like a saved subscription login. Prepare it using native `codex login --with-api-key`
+with the key on stdin, then `rig auth save --codex-home <home> <name>`; never put a key in argv.
+Use the seat's existing home. Preparing a login replaces that home's active credentials, so do it
+before running seats there, or prepare/import a native auth profile separately. Switches preserve
+an outgoing credential copy and report peers sharing the home. A native keyring or administrator
+policy can override `auth.json`; the report remains unverified until native account verification.
+
+Select or clear an account home in a separate operation: `--api-key-file` cannot be combined with
+`--account-home` or `--clear-account-home`. A credential switch inside an already selected home
+keeps that home.
+
+Both paths retain the conversation and launch settings, interrupt in-flight work, and report
+uncertain restarts without retrying. They do not move account homes or start a fresh conversation.
+The caller chooses subscription-first order and configures/enforces the API budget; this command
+neither selects accounts automatically nor imposes a spending limit. Downgrade only after switching
+API-key Claude seats back to subscription credentials: older OpenRig versions cannot read the new
+credential kind correctly.
+
 ### Choosing the Claude instruction file
 
 OpenRig writes its instructions for Claude Code members into managed blocks in
@@ -597,12 +742,14 @@ directory's `CLAUDE.md`, including blocks written by other rigs.
 | `profile` | string | yes | — | Profile name from the referenced AgentSpec. Use `default` for the default profile. Exception: `none` for terminal nodes. |
 | `codex_config_profile` | string | no | — | Codex-only native profile passed as `-p <name>`; letters, numbers, `_`, `.`, `-`. Separate from the AgentSpec `profile`. With the normal launch mode, this replaces OpenRig's explicit workspace-write sandbox flag. A full-bypass policy instead emits danger-full-access and omits this profile argument. |
 | `codex_permissions` | mapping | no | — | Codex-only sandbox and approval envelope, emitted as explicit flags on every managed launch and resume; see [per-member Codex permissions](#per-member-codex-permissions). |
+| `claude_token` | mapping | no | — | Claude-only `{ label, file }`: the seat's own subscription token file, by absolute path, used from its first launch; see "A seat's own subscription token" under [launch environment](#launch-environment). |
 | `runtime` | string | yes | — | Agent runtime. Current supported values: `claude-code`, `codex`, `pi`, `omp`, `terminal`, `stub`. `stub` is a deterministic test harness, not an agent: a Node runner in the seat's tmux pane goes through the normal launch, skill projection, startup-file and readiness path, follows `<cwd>/.openrig/stub/script.json` when present, and calls no model. See "What a stub can and cannot prove" in `docs/as-built/test-layers.md`. |
 | `cwd` | string | yes | — | Working directory for the agent. A relative path resolves against the rig root (the directory containing the rig spec). If that path exists as written, it is used. Otherwise a leading `~` or `~/` expands to the home directory (`~user` does not) and `$VAR` / `${VAR}` expand from the daemon host's environment; a relative result still resolves against the rig root. A reference to an unset variable stays as written, so the cwd is that literal path and works only if it exists. Use `"."` for the rig root itself. Can be overridden at launch time with `rig up --cwd`. A cwd inside the OpenRig installation fails preflight unless `--cwd` is given. Editing this declaration does not update an existing seat; use [`rig seat set-cwd`](agent-startup-guide.md#change-an-existing-seats-working-directory) to save its future launch directory. |
 | `label` | string | no | — | Human-readable member name. Shown in UI when present. |
 | `model` | string | no | — | Model override. Runtime-specific (e.g., `claude-opus-4-6` for Claude Code). |
 | `effort` | string | no | — | Reasoning effort override, trimmed. Claude gets `--effort`, Codex `-c model_reasoning_effort=…`, Pi `--thinking` (see "Pi" below). A non-string value is ignored with an advisory. |
 | `advisor` | string | no | — | Claude-only Advisor model for this seat. Overrides the AgentSpec profile/default; blank or non-string values are ignored with an advisory. |
+| `claude_mcp` | `none` | no | normal loading | Claude members only. Disable MCP server loading for this agent on every launch and resume. Omit the field to keep Claude's normal configuration; see [Claude MCP loading](#claude-mcp-loading-runtime-claude-code). |
 | `permission_policy` | string | no | — | Member-level permission policy; overrides the rig's. Not valid on terminal members. |
 | `env` | map | no | — | Launch environment for this seat, laid over the rig's `env`. Same rules as the rig-level field. Valid on terminal members too. See "Launch environment" below. |
 | `agent_file` | string | no | — | Claude members only. A Claude agent definition (a `.md` file with `name` in its frontmatter) for this seat. It's copied into the seat's own storage and launched as `--agent openrig:<name>`. A path beside the spec: no `..`, not absolute. |
@@ -616,6 +763,53 @@ directory's `CLAUDE.md`, including blocks written by other rigs.
 | `starter_ref` | object | no | — | `{ name }` of an agent starter (`^[a-zA-Z0-9][a-zA-Z0-9_-]*$`) whose files lead the seat's startup. Not valid on terminal members or with a fork. |
 | `startup` | StartupBlock | no | — | Member-level startup files and actions. Applied only to this member. |
 
+### Claude MCP loading (`runtime: claude-code`)
+
+Set `claude_mcp: none` on an individual member to launch it without MCP servers:
+
+```yaml
+members:
+  - id: reviewer
+    agent_ref: local:agents/reviewer
+    profile: default
+    runtime: claude-code
+    cwd: .
+    claude_mcp: none
+```
+
+OpenRig passes `--strict-mcp-config --mcp-config '{"mcpServers":{}}'` to
+Claude for this seat. Strict mode selects only the supplied empty server set,
+excluding servers from user, project, team and plugin configuration. It does
+not disable plugin skills or change the seat's permission mode. The choice is
+saved on the seat and carried through fresh starts, forks, resumes (including
+saved startup context) and handover successors. Export and bundle round trips
+preserve it. The same field is supported on legacy `nodes` declarations.
+
+Other agents in the same directory keep their own choices; OpenRig does not
+rewrite shared MCP configuration files. Omit the field for Claude's normal
+loading. A later launch from an authored spec without the field clears the
+seat's saved opt-out once the launch is accepted for startup; a rejected
+launch leaves the choice unchanged. Restarting from saved context keeps it.
+Only `none` is accepted, and the field is invalid on other runtimes. If the
+seat switches to another runtime, the Claude choice stays dormant and applies
+again if it switches back. A currently running Claude process is unaffected.
+Export describes the current runtime and omits a dormant Claude choice. YAML
+exports name each affected agent in a `# Not carried:` comment, which `rig export`
+also prints after writing the file; JSON exports keep their existing shape. If
+you rebuild from that export and switch back to Claude, restore `claude_mcp: none`
+in the Claude declaration. Re-check this setting after downgrading
+OpenRig too: older versions can ignore it.
+
+Shutdown resume probes and the displayed native recovery commands (including
+the conversation picker) also carry the saved MCP choice. If shutdown cannot
+read that choice, it skips the probe and retains the token; shutdown proceeds.
+
+This opt-out is incompatible with an active enterprise `managed-mcp.json`:
+[Claude rejects strict overrides of that managed server set](https://code.claude.com/docs/en/managed-mcp#servers-passed-with---mcp-config-or---strict-mcp-config).
+OpenRig keeps the requested opt-out flags, so startup and resume remain subject
+to that native refusal; it does not fall back to loading servers. Use a host
+without that managed configuration, or omit the opt-out to use the managed set.
+
 ### Claude Advisor (`runtime: claude-code`)
 
 Set `advisor` independently of the primary `model`:
@@ -653,7 +847,7 @@ own defaults/settings apply. Other runtimes ignore this field.
 - **State:** Each seat uses `$OPENRIG_HOME/state/pi/<session>/agent` as its Pi agent directory, with `sessions/` beside it, instead of your default `~/.pi/agent`. The runner sets `PI_CODING_AGENT_DIR` to that directory. `<session>` is the seat's session name, unchanged; for a pod member it is `{podId}-{memberId}@{rigName}`. `$OPENRIG_HOME` defaults to `~/.openrig`. OpenRig creates these directories at launch and does not copy anything from `~/.pi/agent`.
 - **Thinking level:** A Pi seat's effort (the choice saved with `rig seat set-effort`, else the spec's `effort`) is sent to Pi as `--thinking <level>` on every managed launch and resume: fresh launch, successor and restore. Pi accepts `off`, `minimal`, `low`, `medium`, `high`, `xhigh` and `max`; OpenRig passes the value as written and does not translate it, so Codex's `none` is not mapped to `off`. Only a value of the shape `set-effort` accepts (a letter, then up to 31 letters, digits, `-` or `_`) is sent; any other spec value, such as one with a space or one that starts with `--`, is left out, and `rig seat effort` says so. An unfamiliar level gets an advisory (a warning from `set-effort`), never a refusal, and Pi itself ignores an unknown level with a warning and keeps the level it would use without the flag: a model suffix, the resumed session's level, or its default. With no effort resolved, no flag is sent. The effort wins over a model suffix, and the model string is never changed. `rig seat effort` reports the level sent at launch and, on Pi 0.99.0 or later, the `thinkingLevel` Pi records on this launch's latest successful assistant turn (error and aborted turns are skipped); that is the session's level after Pi's own cap, not proof the provider applied it, and it is shown apart from the level sent when they differ. On older Pi it is NOT VERIFIED. Without an effort, you can still set the level as a model suffix, for example `model: openrouter/x-ai/grok-4:high`. Pi first looks for a model whose id is the whole pin, and uses it when there is one, suffix included. Only when there is none does Pi read the suffix as the level and record the model without it. Any other suffix, such as OpenRouter's `:free`, stays part of the model id. The model-divergence check reads the pin literally when the seat's `models.json` defines a model with that exact id; otherwise it accepts the model with or without the suffix. If the seat's `models.json` can't be read, a pin with a suffix stays unchecked, with the usual warning, when the recorded model matches the whole pin or the pin without its suffix. A record that matches neither is reported.
 - **Custom models:** Pi reads `models.json` from its agent directory, so a seat reads `$OPENRIG_HOME/state/pi/<session>/agent/models.json`, not `~/.pi/agent/models.json`. Custom provider and model definitions for a seat go in that file. A seat that uses only providers Pi already includes does not need one. OpenRig does not create or write `models.json`.
-- **Credentials:** The Pi process receives only `PATH`, `HOME`, `USER`, `LOGNAME`, `TERM`, `LANG`, `LC_ALL`, `SHELL`, `TMPDIR`, a fixed set of OpenRig seat and instance variables, `PI_CODING_AGENT_DIR`, `PI_CODING_AGENT_SESSION_DIR`, and at most one provider key. That key is `OPENROUTER_API_KEY`, `ZAI_API_KEY`, `KIMI_API_KEY` or `MINIMAX_API_KEY`, passed only when it is set in the seat's environment and the seat's `model` is written as `openrouter/<id>`, `zai/<id>`, `kimi-coding/<id>` or `minimax/<id>`. Naming that variable in `recovery.provider_auth_env_allowlist` (empty by default) makes the daemon add it, when set in the daemon's environment, to the seat's launch environment. No other variable reaches Pi, including other providers' keys and a custom provider's own key variable.
+- **Credentials:** The Pi process receives only `PATH`, `HOME`, `USER`, `LOGNAME`, `SSH_AUTH_SOCK`, `TERM`, `LANG`, `LC_ALL`, `SHELL`, `TMPDIR`, a fixed set of OpenRig seat and instance variables, `PI_CODING_AGENT_DIR`, `PI_CODING_AGENT_SESSION_DIR`, and at most one provider key. That key is `OPENROUTER_API_KEY`, `ZAI_API_KEY`, `KIMI_API_KEY` or `MINIMAX_API_KEY`, passed only when it is set in the seat's environment and the seat's `model` is written as `openrouter/<id>`, `zai/<id>`, `kimi-coding/<id>` or `minimax/<id>`. Naming that variable in `recovery.provider_auth_env_allowlist` (empty by default) makes the daemon add it, when set in the daemon's environment, to the seat's launch environment. No other variable reaches Pi, including other providers' keys and a custom provider's own key variable.
 
 - **Extension dialogs:** The RPC runner shows extension `select`, `confirm`, `input` and `editor` requests in the pane and cancels them because the managed pane cannot answer them. The seat needs attention through the end of the turn, until a new agent run starts. Review the cancelled request and send an instruction the extension can complete without a dialog. Extension `notify` text is shown without raising attention or sending a response.
 
@@ -718,7 +912,11 @@ This is human-authored (you choose the pod/member IDs) and system-validated (the
 | `spawned_by` | Source was spawned by target; the target launches first. Constrains launch order. | Child → parent in hierarchical topologies |
 | `can_observe` | Source can observe target's output. Does NOT constrain launch order. | Reviewer → implementer, monitor → worker |
 | `collaborates_with` | Peer collaboration relationship. Does NOT constrain launch order. | Co-equal peers working together |
-| `escalates_to` | Source escalates to target for decisions. Does NOT constrain launch order. | Worker → lead for escalation |
+| `escalates_to` | Fallback queue escalation for a seat without a declared delegation parent. Does NOT constrain launch order. | Local member or same-daemon `pod-member@rig` address |
+
+A named `escalates_to` target may be in another rig on the same daemon and need not exist at import time.
+See [fallback escalation across rigs](edge-types.md#fallback-escalation-across-rigs) for precedence, single-hop delivery,
+cycle detection and visible fallback outcomes.
 
 ### Pod-Local Edges
 
@@ -801,19 +999,21 @@ the modified topology, run `rig snapshot <rig-id>` and use its returned ID with
 
 Startup blocks can appear at three levels: rig, pod, and member. They are merged additively via the startup layering model (see [the layering model](agent-startup-guide.md#the-layering-model)). The rig's `culture_file` joins as a required file with the `auto` hint in the culture layer, on both fresh start and restore.
 
-`rig export` includes the rig's saved culture file and startup block, and now also
-writes each pod's saved startup files under that pod. The file order, delivery
-hints, required flags, orientation and `applies_on` selections are retained.
-Previously the saved pod file lists were omitted. Pods without a recorded file
-list, including older rigs, keep their existing export behavior; the exporter
-does not reconstruct declarations from a member's combined startup context.
+For member `startup:`, export uses that seat's retained launch recipe (or its pre-launch materialization) rather than guessing from projected files. If that source is unavailable or a member's files came from a different rig root, the YAML names the member's omitted startup; JSON export names omissions in the `X-OpenRig-Export-Warnings` response header. An older seat without a recipe may have had no member startup or an unretained one: inspect its source before re-importing. When member startup files are present in the export, `rig export` checks required paths against the output folder and warns for missing files. The export does not copy member startup files.
+
+`rig export` includes the rig's saved culture file and startup block, and writes
+saved pod startup files and actions under their pod. The file and action order,
+delivery hints, required flags, orientation and `applies_on` selections are retained.
+Pod actions are saved for export when a new rig is created; older rigs and pods
+added from a separate expansion may not have those declarations recorded. Export
+does not reconstruct them from a member's combined startup context; keep the source
+spec when those declarations are needed. This does not change which startup actions
+are inherited when a seat is added later.
 
 File paths remain relative to the exported spec's folder. When exporting to a
 different folder, supply the referenced startup files there or import with the
-original `--rig-root`. The CLI warns once per missing required rig or pod file;
-missing optional files do not trigger that warning. Pod startup actions and
-member startup blocks remain outside this export; keep the source spec when you
-need those declarations.
+original `--rig-root`. The CLI warns once per missing required rig, pod or member
+file; missing optional files do not trigger that warning.
 
 ### Files
 

docs/reference/runner-conformance/README.md

@@ -0,0 +1,75 @@
+# Runner conformance check
+
+A runnable check for runner authors. Point it at your runner and it reports which promises of the
+[harness and runner contract](../harness-runner-contract.md) your runner keeps. It needs Node and tmux, and no OpenRig
+daemon, model, credentials or network.
+
+```sh
+node check-runner.mjs --runner 'my-runner --agent-command {agent}' --name my-harness --resume --first-input
+```
+
+- **`{agent}`** is replaced with the path of the fake ACP agent beside this file. Your runner starts it where it would
+  start your real agent. It accepts any arguments your runner adds.
+- **`--name`** is your preset's name, which your posts carry as `runtime`.
+- **`--resume` and `--first-input`** say your preset declares `resume: true` and `first_input: true`. Without them,
+  those cases are skipped, because OpenRig wouldn't pass that input.
+- **`--cases 2,4`** runs only some cases. `--json` prints a machine-readable report. `--keep` keeps each launch's
+  folder: the agent's message log, the launch record and the scenario.
+- **`--wait-ms 20000`** waits longer for each step (default 8000), for a runner whose agent takes a while to start. The check's shorter settling times, and the fake agent's own delays and holds, grow in proportion.
+- **`--pass-env NAME`** passes one more variable through. Otherwise the runner gets only `PATH`, `HOME`, `USER`,
+  `SHELL`, `LANG` and `TMPDIR`, plus the OpenRig variables.
+- **The exit status** is 0 when nothing fails, 1 when a case fails, and 2 for a usage or setup problem. An advisory result (7c) never changes it.
+
+In an installed OpenRig, the check is at `$(npm root -g)/@openrig/cli/daemon/docs/reference/runner-conformance/`.
+
+## How it runs
+
+- **Your runner** starts in a private tmux server, with the environment a runner-preset launch gives it:
+  `OPENRIG_URL`, `OPENRIG_ACTIVITY_HOOK_TOKEN`, `OPENRIG_SESSION_NAME`, `OPENRIG_NODE_ID`, `OPENRIG_RUNTIME`,
+  `OPENRIG_OCCUPANT_GENERATION`, `OPENRIG_LAUNCH_RESULT` and `OPENRIG_LAUNCH_ATTEMPT`, plus `OPENRIG_RESUME_REF` or
+  `OPENRIG_FIRST_INPUT` when a case needs them. `OPENRIG_HOME` is an empty folder, so nothing reaches a real OpenRig.
+- **The fake agent** speaks ACP v1 over stdio. Each case gives it a scenario, such as a slow handshake, a permission
+  request, a turn that ends cancelled, or an exit mid-turn. It logs every message in both directions. Its own request
+  ids are numbers, as JSON-RPC agents commonly use, so a runner passes them on unchanged.
+- **A local endpoint** stands in for `POST /api/activity/hooks`. It records each post, and refuses one without the
+  launch's bearer token, as OpenRig does.
+  - **In cases 4 and 4b,** it also stands in for the answered path's three routes: submit, read with `waitSeconds`, and
+    withdraw. It checks a submit the way OpenRig does: the seat and generation, `nativeRef` as a non-empty string, `turnId`
+    and `requestId` as a string or a number (a number reads as its decimal string). Options a person can't be shown (not
+    2-4 of them, ids that repeat or use other characters, or other kinds) are permitted with the agent's allow option, as OpenRig does. Each launch scripts the answer,
+    the permit or the refusal it gives, or closes the connection unanswered.
+  - **Where a launch scripts none of that,** those routes answer 404, as an OpenRig without the answered path does, so
+    the runner permits.
+- **Input arrives the way `rig send` delivers it:** the text is pasted as one bracketed paste with raw newlines, then
+  Enter is pressed. Ordinary messages carry a From/To envelope.
+
+## The cases
+
+| Case | Contract section | What it checks |
+|---|---|---|
+| 1 | 3 | No launch record until `session/new` is answered. Then a record holding format 1, this attempt, the preset's name, the agent's conversation id and a `producer` |
+| 1b | 3, 4 | A handshake that fails at `initialize` or `session/new` (an error or an exit) writes no conversation in a launch record, and posts no accepted prompt or idle report |
+| 2 | 5, 6 | One pasted block becomes exactly one `session/prompt`, byte for byte. A line reading `/abort` inside a block isn't control. A block sent mid-turn waits for the next turn. A block that is exactly `/abort`, or Ctrl-C outside a paste, sends one `session/cancel`. `Stop/cancelled` is posted only when the agent itself returns `cancelled` |
+| 3 | 4 | `UserPromptSubmit/accepted` with a `turnId`, then `Stop/end_turn`. Every post carries the contract's fields, the hook token, this seat's identity and a rising `sequence`. `livenessSeconds`, when given, is from 1 to 3600 |
+| 4 | 7 | When no one can be asked, the request is permitted: an OpenRig without the answered path (404) at once, and one the runner can't reach, each get the agent's `allow_once` option, never its `allow_always` while `allow_once` is offered. The turn then reports as usual, with no `permission_prompt`. Only `denied_by_seat`, or a request with no allow option, is answered at once with a refusal, never an allow option. `Notification/permission_prompt` is then posted, and for `denied_by_seat` the pane shows what was asked (the tool call's title). No idle `Stop` or beat is posted while that attention is held, even when `/abort` then cancels the turn. The next prompt is accepted as usual |
+| 4b | 7 | The live request is submitted with the hook token, its `requestId` as the agent's JSON-RPC id, passed unchanged, the accepted prompt's `turnId`, this conversation and generation, and the agent's options as offered. The runner doesn't post `permission_prompt` itself for a created ask, and each read waits at most 25 s. The chosen option, allow or deny, reaches the agent once, and the turn then reports as usual. OpenRig's `permitted`, at submit or at a read, reaches the agent the same way. A refusal is read by its code, never its status: one section 7's table marks "permits" (`duplicate_request`, `activity_grant_pending`, `activity_grant_revoked`), a read that gets no answer, and a status or read state the section doesn't list (500, `deferred`) each give the agent its `allow_once` option. A refusal saying the request isn't the current occupant's (`stale_generation` at submit, `activity_grant_stale` on a read), and a `refused` (`said_no`), stale or withdrawn read, give the agent `cancelled`, and the turn then ends as usual, with nothing held. With no answer, the request stays pending with no idle report, and a pasted message isn't an answer. `/abort` withdraws it with `turn_cancelled`, and an answer after that isn't delivered. A turn's end or the agent's exit while it waits withdraws it with `turn_ended` or `exited`. Once the ask is withdrawn no person is still waiting, so the turn's end is reported as usual. An ask the runner permitted after a read got no answer, or an unlisted one, may still be open, so it's withdrawn at the turn's end too |
+| 5 | 4 | Every post carries `OPENRIG_OCCUPANT_GENERATION` |
+| 6 | 8 | A resume asks the agent to reopen `OPENRIG_RESUME_REF` in the launch's working folder, and records that conversation. The agent's own not-found becomes `error: "missing"`. Any other failure is neither missing nor resumed. No case starts a new conversation |
+| 7 | 4 | An agent that dies mid-turn gets `Stop/<preset>_exited`, and no idle report after it |
+| 7b | 4 | A failed `session/prompt` posts `Notification/runtime_error`, not a permission request |
+| 7c | 9 | **Advisory.** Closing the pane ends the agent process. Section 9 marks this duty as proposed, and lets a runner list in `limits` what may keep running, so this result is shown as met or not met and never counts as a pass or a fail, or toward the exit status. The fake agent outlives a hangup here, so the runner has to end it |
+| 8 | 7 | An unknown request from the agent gets JSON-RPC `-32601`, and the next prompt still goes through |
+| 9 | 9 | `OPENRIG_FIRST_INPUT`'s exact bytes, newlines and spaces included, are the first prompt |
+
+## What it doesn't check
+
+- **How OpenRig reads your posts:** working, idle, needs input, unknown after an exit, a send refused while a
+  permission waits, and an earlier generation not counting. That's OpenRig's side of the contract, and its own tests
+  cover it.
+- **Your real agent:** a passing check isn't a run with your agent. Section 10's real check, with your agent under a
+  runner preset, is still the useful evidence. Passing this is evidence for the versions you ran, not certification.
+- **The model handoff:** `OPENRIG_MODEL` is mapped to your harness's own setting, with no wire this check can see.
+- **A resume for a preset without `resume`:** OpenRig doesn't pass `OPENRIG_RESUME_REF` to it.
+- **Restarting the agent after a failed handshake:** the contract doesn't rule on that.
+- **The chosen person's side of an ask:** who may answer, the Feed and Slack are OpenRig's side, and its own tests
+  cover them. Routing an ask to a seat is a later change.

docs/reference/runner-conformance/check-runner.mjs

@@ -0,0 +1,721 @@
+#!/usr/bin/env node
+// OpenRig runner conformance check: point it at your runner and see which promises of the runner contract
+// (docs/reference/harness-runner-contract.md) it keeps. It needs Node and tmux, and no OpenRig daemon:
+//   - your runner starts in a private tmux server, with the environment a runner-preset launch gives it;
+//   - the fake ACP agent beside this file stands in for your agent (your runner starts it where it would start the real one);
+//   - a small local endpoint stands in for OpenRig's activity route, and records what your runner posts;
+//   - input arrives the way `rig send` delivers it: one bracketed paste with raw newlines, then Enter.
+//
+//   node check-runner.mjs --runner '<command>' [--name <preset name>] [--resume] [--first-input]
+//                         [--cases 1,2,4] [--wait-ms 8000] [--pass-env NAME] [--json] [--keep]
+//
+// `{agent}` in the command is replaced with the fake agent's path, an executable that takes any arguments your runner
+// adds. For example: --runner 'my-runner --agent-command {agent}'. --resume and --first-input say your preset declares
+// `resume: true` and `first_input: true`; without them those cases are skipped, since OpenRig wouldn't pass that input.
+// The runner gets a clean environment: PATH, HOME, USER, SHELL, LANG and TMPDIR, the OpenRig variables, and any
+// --pass-env names. Exit status: 0 when nothing fails, 1 when a case fails, 2 for a usage or setup problem.
+//
+// It checks the runner's side only. How OpenRig reads those posts (working, idle, needs input) is OpenRig's own tests'
+// job, and one real run with your agent is still the useful check (section 10 of the contract). Passing this is
+// evidence for the versions you ran, not certification.
+import { spawnSync } from "node:child_process";
+import { randomBytes } from "node:crypto";
+import { chmodSync, existsSync, mkdirSync, mkdtempSync, readFileSync, realpathSync, rmSync, statSync, writeFileSync } from "node:fs";
+import { createServer } from "node:http";
+import { tmpdir } from "node:os";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+
+const here = dirname(fileURLToPath(import.meta.url));
+const usage = "usage: node check-runner.mjs --runner '<command with {agent}>' [--name <preset>] [--resume] [--first-input] [--cases 1,2] [--wait-ms 8000] [--pass-env NAME] [--json] [--keep]";
+
+function parseArgs(argv) {
+  const opts = { name: "runner", cases: undefined, passEnv: [], resume: false, firstInput: false, json: false, keep: false, waitMs: 8000 };
+  for (let i = 0; i < argv.length; i += 1) {
+    const flag = argv[i];
+    const value = () => { if (i + 1 >= argv.length) throw new Error(`${flag} needs a value`); return argv[++i]; };
+    if (flag === "--runner") opts.runner = value();
+    else if (flag === "--name") opts.name = value();
+    else if (flag === "--cases") opts.cases = value().split(",").map((c) => c.trim()).filter(Boolean);
+    else if (flag === "--pass-env") opts.passEnv.push(value());
+    else if (flag === "--wait-ms") opts.waitMs = Number(value());
+    else if (flag === "--resume") opts.resume = true;
+    else if (flag === "--first-input") opts.firstInput = true;
+    else if (flag === "--json") opts.json = true;
+    else if (flag === "--keep") opts.keep = true;
+    else throw new Error(`unknown option ${flag}`);
+  }
+  if (!opts.runner) throw new Error("--runner is required");
+  if (!opts.runner.includes("{agent}")) throw new Error("--runner must contain {agent}, where your runner takes its agent command");
+  if (!/^[A-Za-z0-9._-]+$/.test(opts.name)) throw new Error("--name must be a preset name (letters, digits, '.', '_' or '-')");
+  if (!(opts.waitMs > 0)) throw new Error("--wait-ms must be a positive number");
+  return opts;
+}
+
+let opts;
+try { opts = parseArgs(process.argv.slice(2)); } catch (error) { console.error(`${error.message}\n${usage}`); process.exit(2); }
+if (spawnSync("tmux", ["-V"]).status !== 0) { console.error("tmux is required: the check drives your runner in a tmux pane, as OpenRig does."); process.exit(2); }
+
+const socket = `openrig-runner-check-${process.pid}`;
+const tmux = (...args) => spawnSync("tmux", ["-L", socket, ...args], { encoding: "utf8" });
+const sh = (value) => `'${String(value).replaceAll("'", "'\\''")}'`;
+const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
+// Settling times and per-step bounds grow with --wait-ms on a slow machine; they never shrink below these values. The
+// fake agent's own delays and holds grow by the same factor, so each settle keeps its place against what it measures.
+const scale = (ms) => Math.round(ms * Math.max(1, opts.waitMs / 8000));
+async function waitFor(read, ms = opts.waitMs) {
+  const deadline = Date.now() + ms;
+  for (;;) {
+    const value = read();
+    if (value) return value;
+    if (Date.now() > deadline) return undefined;
+    await sleep(50);
+  }
+}
+
+// The stand-in for OpenRig's activity route and, for a launch that scripts one, its answered permission routes (section 7).
+// Each launch has its own token; a request without one is refused, as OpenRig does.
+const posts = [];
+const tokens = new Map();
+const OPTION_ID = /^[A-Za-z0-9_-]{1,40}$/;
+const ANSWERABLE = new Set(["allow_once", "allow_always", "reject_once", "reject_always"]);
+const text = (value) => typeof value === "string" && value.trim() ? value.trim() : null;
+// A JSON-RPC id or prompt number as the runner passed it: a number reads as its decimal string, so 7 and "7" match.
+const idText = (value) => typeof value === "number" && Number.isFinite(value) ? String(value) : text(value);
+const server = createServer((req, res) => {
+  let raw = "";
+  req.on("data", (chunk) => { raw += chunk; });
+  req.on("end", () => {
+    const token = (req.headers.authorization ?? "").replace(/^Bearer /, "");
+    let body;
+    try { body = raw ? JSON.parse(raw) : undefined; } catch { body = undefined; }
+    const url = new URL(req.url, "http://stand-in");
+    posts.push({ at: Date.now(), method: req.method, path: url.pathname, query: Object.fromEntries(url.searchParams), token, body, raw });
+    const send = (status, payload) => res.writeHead(status, { "content-type": "application/json" }).end(JSON.stringify(payload));
+    const h = tokens.get(token);
+    if (req.method === "POST" && url.pathname === "/api/activity/hooks") {
+      if (!h) return send(401, { ok: false });
+      return body ? send(200, { ok: true }) : send(400, { ok: false, code: "invalid_json" });
+    }
+    const ask = url.pathname.match(/^\/api\/activity\/permission-requests(?:\/([^/]+)(\/withdraw)?)?$/);
+    if (!ask) return send(404, { ok: false, code: "not_found" });
+    if (!h) return send(401, { ok: false });
+    // Without a scripted ask, the stand-in is an OpenRig that predates the answered path: 404, so the runner permits.
+    if (!h.asks) return send(404, { ok: false, code: "not_found" });
+    // A scripted `drop` closes the connection unanswered: an OpenRig the runner can't reach.
+    const drop = req.method === "POST" && !ask[1] ? h.asks.script.submit?.drop : req.method === "GET" && h.asks.script.read?.drop;
+    if (drop) return void res.socket.destroy();
+    if (req.method === "POST" && !ask[1]) return submitAsk(h, body, raw, send);
+    const entry = h.asks.byId.get(decodeURIComponent(ask[1] ?? ""));
+    if (!entry) return send(404, { ok: false, code: "not_found" });
+    if (req.method === "GET" && h.asks.script.read) return send(h.asks.script.read.status, { ok: false, code: h.asks.script.read.code });
+    if (req.method === "POST" && ask[2]) {
+      if (!["turn_cancelled", "turn_ended", "exited"].includes(body?.reason)) return send(400, { ok: false, code: "invalid_permission_request" });
+      entry.withdrawn = body.reason;
+      entry.wake();
+      return send(200, { ok: true, ...askState(h, entry) });
+    }
+    if (req.method === "GET" && !ask[2]) {
+      const waitMs = Math.min(Number(url.searchParams.get("waitSeconds") ?? 0), 25) * 1000;
+      const reply = () => send(200, { ok: true, ...askState(h, entry) });
+      if (askState(h, entry).state !== "pending" || !(waitMs > 0)) return reply();
+      const timer = setTimeout(() => { entry.waiters.delete(reply); reply(); }, waitMs);
+      entry.waiters.add(() => { clearTimeout(timer); reply(); });
+      return undefined;
+    }
+    return send(404, { ok: false, code: "not_found" });
+  });
+});
+
+// What the scripted ask reads as now: an answer once its time has come, `withdrawn` after a withdraw, otherwise pending. A
+// `late` answer is what the stand-in reports after a withdraw, so a runner that keeps reading and delivers it is caught.
+function askState(h, entry) {
+  const script = h.asks.script;
+  if (entry.withdrawn) return script.late ?? { state: "withdrawn" };
+  if (script.answer && Date.now() >= entry.created + (script.answer.afterMs ?? 0)) {
+    const { state, optionId, reason } = script.answer;
+    return { state, ...(optionId ? { optionId } : {}), ...(reason ? { reason } : {}) };
+  }
+  return { state: "pending" };
+}
+
+// Submit, checked as OpenRig checks it: the seat and generation, nativeRef as a string, turnId and requestId as a string
+// or a number. When no one can be asked, OpenRig permits with the agent's allow_once, else allow_always: here when the
+// script says so, or for options a person can't be shown (2-4 with distinct ids). With no allow option it refuses.
+function submitAsk(h, body, raw, send) {
+  const script = h.asks.script;
+  const problems = [];
+  if (!body) return send(400, { ok: false, code: "invalid_json" });
+  if (body.sessionName !== h.session) return send(404, { ok: false, code: "session_not_found" });
+  if (!text(body.nativeRef)) problems.push(`nativeRef must be a non-empty string (got ${JSON.stringify(body.nativeRef)})`);
+  for (const key of ["turnId", "requestId"]) if (!idText(body[key])) problems.push(`${key} must be a non-empty string or a number (got ${JSON.stringify(body[key])})`);
+  const options = Array.isArray(body.options) ? body.options : null;
+  if (!options) problems.push("options must be an array");
+  else if (options.some((o) => !text(o?.optionId) || !text(o?.name) || !text(o?.kind))) problems.push("each option needs optionId, name and kind");
+  h.asks.submitted.push({ body, problems });
+  if (problems.length) return send(400, { ok: false, code: "invalid_permission_request", error: problems.join("; ") });
+  if (body.generation !== h.generation) return send(409, { ok: false, code: "stale_generation" });
+  if (script.submit?.status) return send(script.submit.status, { ok: false, code: script.submit.code });
+  const allow = options.find((o) => o.kind === "allow_once") ?? options.find((o) => o.kind === "allow_always");
+  const permit = (reason) => allow ? send(200, { ok: true, state: "permitted", optionId: allow.optionId, reason })
+    : send(422, { ok: false, code: "options_unsupported" });
+  if (script.submit?.permitted) return permit("no one can be asked");
+  if (options.length < 2 || options.length > 4 || new Set(options.map((o) => o.optionId)).size !== options.length
+    || options.some((o) => !OPTION_ID.test(o.optionId) || !ANSWERABLE.has(o.kind))) return permit("these options can't be shown to a person");
+  const requestId = idText(body.requestId), turnId = idText(body.turnId);
+  const key = JSON.stringify([body.generation, text(body.nativeRef), requestId]);
+  const earlier = h.asks.byKey.get(key);
+  if (earlier) return send(200, { ok: true, askId: earlier.askId, created: false });
+  if ([...h.asks.byKey.values()].some((e) => e.requestId === requestId && e.turnId !== turnId)) return send(409, { ok: false, code: "duplicate_request" });
+  const entry = { askId: `ask-${h.label}-${h.asks.byKey.size + 1}`, requestId, turnId, created: Date.now(), waiters: new Set() };
+  entry.wake = () => { for (const w of [...entry.waiters]) w(); entry.waiters.clear(); };
+  if (script.answer) setTimeout(entry.wake, script.answer.afterMs ?? 0);
+  h.asks.byKey.set(key, entry);
+  h.asks.byId.set(entry.askId, entry);
+  return send(200, { ok: true, askId: entry.askId, created: true });
+}
+await new Promise((resolve) => server.listen(0, "127.0.0.1", resolve));
+const base = `http://127.0.0.1:${server.address().port}`;
+
+const launches = [];
+async function launch(label, { scenario = {}, env = {}, asks } = {}) {
+  const dir = mkdtempSync(join(tmpdir(), `runner-check-${label}-`));
+  const cwd = join(dir, "work");
+  mkdirSync(cwd);
+  mkdirSync(join(dir, "home"));
+  mkdirSync(join(dir, "launch"));
+  const id = randomBytes(4).toString("hex");
+  const h = {
+    label, dir, cwd, token: randomBytes(16).toString("hex"), session: `check-${label}@runner-check`, nodeId: `node-${label}-${id}`,
+    generation: `gen-${id}`, attempt: `attempt-${id}`, recordPath: join(dir, "launch", "record.json"), agentLog: join(dir, "agent.jsonl"),
+    target: `check-${label}`,
+  };
+  if (asks) h.asks = { script: asks, byKey: new Map(), byId: new Map(), submitted: [] };
+  tokens.set(h.token, h);
+  writeFileSync(join(dir, "scenario.json"), JSON.stringify(scenario));
+  const agent = join(dir, "agent");
+  writeFileSync(agent, `#!/bin/sh\nexec ${sh(process.execPath)} ${sh(join(here, "fake-acp-agent.mjs"))} --scenario ${sh(join(dir, "scenario.json"))} --log ${sh(h.agentLog)} "$@"\n`);
+  chmodSync(agent, 0o755);
+  const vars = {
+    PATH: process.env.PATH, HOME: process.env.HOME, USER: process.env.USER, SHELL: process.env.SHELL, LANG: process.env.LANG,
+    TMPDIR: process.env.TMPDIR, TERM: "xterm-256color",
+    ...Object.fromEntries(opts.passEnv.map((name) => [name, process.env[name]])),
+    OPENRIG_HOME: join(dir, "home"), OPENRIG_URL: base, OPENRIG_ACTIVITY_HOOK_TOKEN: h.token, OPENRIG_SESSION_NAME: h.session,
+    OPENRIG_NODE_ID: h.nodeId, OPENRIG_RUNTIME: opts.name, OPENRIG_OCCUPANT_GENERATION: h.generation,
+    OPENRIG_LAUNCH_RESULT: h.recordPath, OPENRIG_LAUNCH_ATTEMPT: h.attempt, ...env,
+  };
+  const assignments = Object.entries(vars).filter(([, value]) => value !== undefined).map(([key, value]) => `${key}=${sh(value)}`).join(" ");
+  const command = opts.runner.replaceAll("{agent}", sh(agent));
+  writeFileSync(join(dir, "launch.sh"), `#!/bin/sh\ncd ${sh(cwd)} || exit 1\nexec /usr/bin/env -i ${assignments} /bin/sh -c ${sh(command)}\n`);
+  chmodSync(join(dir, "launch.sh"), 0o755);
+  const started = tmux("new-session", "-d", "-s", h.target, "-x", "200", "-y", "50", join(dir, "launch.sh"));
+  if (started.status !== 0) throw new Error(`tmux couldn't start the runner: ${started.stderr.trim()}`);
+  Object.assign(h, {
+    log: () => !existsSync(h.agentLog) ? [] : readFileSync(h.agentLog, "utf8").split("\n").filter(Boolean).map((line) => JSON.parse(line)),
+    received: (method) => h.log().filter((e) => e.dir === "in" && e.msg.method === method),
+    sent: (method) => h.log().filter((e) => e.dir === "out" && e.msg.method === method),
+    answerTo: (method) => { const ask = h.sent(method)[0]; return ask && h.log().find((e) => e.dir === "in" && e.msg.id === ask.msg.id && !e.msg.method); },
+    replyTo: (entry) => h.log().find((e) => e.dir === "out" && e.msg.id === entry.msg.id && !e.msg.method),
+    events: (event) => h.log().filter((e) => e.dir === "event" && e.event === event),
+    prompts: () => h.received("session/prompt").map((e) => (e.msg.params?.prompt ?? []).filter((b) => b.type === "text").map((b) => b.text).join("")),
+    posts: () => posts.filter((p) => p.path === "/api/activity/hooks" && (p.token === h.token || p.body?.sessionName === h.session)),
+    permissionCalls: (method, suffix = "") => posts.filter((p) => p.token === h.token && p.method === method
+      && p.path.startsWith("/api/activity/permission-requests") && p.path.endsWith(suffix)),
+    hooks: (event, subtype) => h.posts().filter((p) => p.body?.hookEvent === event && (subtype === undefined || p.body?.subtype === subtype)),
+    record: () => { if (!existsSync(h.recordPath)) return undefined; try { return JSON.parse(readFileSync(h.recordPath, "utf8")); } catch { return { unreadable: true }; } },
+    paste: (text) => {
+      const file = join(dir, `paste-${randomBytes(3).toString("hex")}`);
+      writeFileSync(file, text);
+      const buffer = `b-${randomBytes(3).toString("hex")}`;
+      tmux("load-buffer", "-b", buffer, file);
+      tmux("paste-buffer", "-t", h.target, "-b", buffer, "-d", "-r", "-p");
+      tmux("send-keys", "-t", h.target, "Enter");
+    },
+    keys: (...keys) => tmux("send-keys", "-t", h.target, ...keys),
+    stop: () => tmux("kill-session", "-t", h.target),
+    screen: () => tmux("capture-pane", "-p", "-t", h.target).stdout ?? "",
+  });
+  launches.push(h);
+  return h;
+}
+
+// An ordinary send arrives inside an envelope, so a line in it that reads /abort is still just text.
+const message = (h, body) => `From: checker@runner-check\nTo: ${h.session}\n---\n${body}`;
+const nonExitStops = (h, after = 0) => h.hooks("Stop").filter((p) => p.at >= after && !String(p.body?.subtype ?? "").endsWith("_exited"));
+const ready = (h, ms) => waitFor(() => h.record()?.nativeRef && h.record(), ms);
+
+function result(id, title, { proposed = false } = {}) {
+  const r = { id, title, status: "pass", checks: [], ...(proposed ? { proposed: true } : {}) };
+  r.check = (name, ok, detail) => { r.checks.push({ name, ok: Boolean(ok), ...(ok ? {} : { detail }) }); if (!ok) r.status = "fail"; return ok; };
+  r.skip = (reason) => { r.status = "skip"; r.reason = reason; return r; };
+  return r;
+}
+
+const cases = {
+  async "1"() {
+    const r = result("1", "Handshake: the launch record comes only after the conversation is open");
+    const h = await launch("1", { scenario: { newDelayMs: scale(2000), sessionId: "fake-session-1" } });
+    const asked = await waitFor(() => h.received("session/new")[0]);
+    if (!r.check("the runner opens a conversation with session/new", asked, "no session/new arrived")) return r;
+    // Until the agent's answer is logged, there must be no record. The agent logs each message before it sends it, so a
+    // record read before a log that still holds no answer was written before the answer went out. Reading against the
+    // agent's log, not after a fixed settle, keeps a busy machine from reading a correct runner's record as early.
+    let early;
+    await waitFor(() => {
+      const record = h.record();
+      if (h.replyTo(asked)) return true;
+      if (record) early = record;
+      return early;
+    }, scale(2000) + opts.waitMs);
+    r.check("no launch record while session/new is still unanswered", !early, `a record was written early: ${JSON.stringify(early)}`);
+    const record = await ready(h);
+    r.check("a launch record arrives after session/new is answered", record, `no record with a nativeRef at ${h.recordPath}`);
+    if (!record) return r;
+    const answered = h.replyTo(asked);
+    r.check("the record is written after the answer", answered && statSync(h.recordPath).mtimeMs >= answered.at - 5, "the record's time precedes the answer");
+    r.check("format 1, this attempt, this preset", record.format === 1 && record.attempt === h.attempt && record.harness === opts.name,
+      `format=${record.format} attempt=${record.attempt} harness=${record.harness}`);
+    r.check("nativeRef is the agent's conversation id", record.nativeRef === "fake-session-1", `nativeRef=${record.nativeRef}`);
+    r.check("producer names the runner's versions", typeof record.producer === "string" && record.producer.trim(), "no producer");
+    return r;
+  },
+
+  async "1b"() {
+    const r = result("1b", "A failed handshake writes no launch record and claims no turn");
+    for (const [label, scenario] of [["initialize error", { initialize: "error" }], ["initialize exit", { initialize: "exit" }],
+      ["session/new error", { newSession: "error" }], ["session/new exit", { newSession: "exit" }]]) {
+      const h = await launch(`1b-${label.replace(/[^a-z]+/g, "-")}`, { scenario });
+      await waitFor(() => h.events("exit")[0] || h.log().some((e) => e.dir === "out" && e.msg.error), opts.waitMs);
+      await sleep(scale(1500));
+      const record = h.record();
+      r.check(`${label}: no conversation in a launch record`, !record?.nativeRef, `record: ${JSON.stringify(record)}`);
+      r.check(`${label}: no accepted prompt or idle report`, h.hooks("UserPromptSubmit").length === 0 && nonExitStops(h).length === 0,
+        `posts: ${h.posts().map((p) => `${p.body?.hookEvent}/${p.body?.subtype}`).join(", ")}`);
+    }
+    return r;
+  },
+
+  async "2"() {
+    const r = result("2", "Input framing, queued messages and cancel");
+    const h = await launch("2", { scenario: { turns: [{}, { holdMs: scale(15000) }, { holdMs: scale(2500), ignoreCancel: true }, {}] } });
+    if (!r.check("the runner opens its conversation", await ready(h), "no launch record")) return r;
+    const block = message(h, "first line\n/abort\nthird line");
+    h.paste(block);
+    await waitFor(() => h.prompts().length >= 1);
+    await waitFor(() => nonExitStops(h).length >= 1, scale(3000));
+    await sleep(scale(500));
+    r.check("one pasted block is exactly one session/prompt", h.prompts().length === 1, `got ${h.prompts().length} prompts`);
+    r.check("the prompt holds the whole block, byte for byte", h.prompts()[0] === block, `got ${JSON.stringify(h.prompts()[0])}`);
+
+    h.paste(message(h, "please hold"));
+    if (!r.check("the next block starts the next turn", await waitFor(() => h.prompts().length >= 2), "no second prompt")) return r;
+    // A block that arrives while the agent works, with a line reading /abort, is still one message for later.
+    const queued = message(h, "sent while the agent works\n/abort\nstill part of the message");
+    h.paste(queued);
+    await sleep(scale(1200));
+    r.check("a block sent mid-turn waits for the next turn", h.prompts().length === 2, `got ${h.prompts().length} prompts during the turn`);
+    r.check("a line reading /abort inside a block isn't control", h.received("session/cancel").length === 0, "the runner sent session/cancel");
+
+    const before = Date.now();
+    h.paste("/abort");
+    const cancel = await waitFor(() => h.received("session/cancel")[0], scale(3000));
+    r.check("a block that is exactly /abort sends session/cancel", cancel, "no session/cancel");
+    const stopped = await waitFor(() => nonExitStops(h, before)[0], scale(3000));
+    r.check("the cancelled turn reports Stop/cancelled", stopped?.body?.subtype === "cancelled", `got ${stopped ? `Stop/${stopped.body.subtype}` : "no Stop"}`);
+    await sleep(scale(300));
+    r.check("one /abort sends one session/cancel", h.received("session/cancel").length === 1, `got ${h.received("session/cancel").length}`);
+    r.check("the waiting block becomes the next prompt", await waitFor(() => h.prompts()[2] === queued, scale(4000)), `prompts: ${JSON.stringify(h.prompts().slice(2))}`);
+
+    const interrupted = Date.now();
+    await sleep(scale(200));
+    h.keys("C-c");
+    r.check("Ctrl-C outside a paste sends session/cancel", await waitFor(() => h.received("session/cancel").length >= 2, scale(3000)), "no second session/cancel");
+    const ended = await waitFor(() => nonExitStops(h, interrupted)[0], scale(6000));
+    r.check("cancelled is reported only on the agent's own stop reason", ended?.body?.subtype === "end_turn",
+      `the agent ended the turn end_turn; the runner reported ${ended ? `Stop/${ended.body.subtype}` : "no Stop"}`);
+    return r;
+  },
+
+  async "3"() {
+    const r = result("3", "Activity: turn boundaries, posted in order with the seat's identity");
+    const h = await launch("3");
+    if (!r.check("the runner opens its conversation", await ready(h), "no launch record")) return r;
+    h.paste(message(h, "a short turn"));
+    const accepted = await waitFor(() => h.hooks("UserPromptSubmit", "accepted")[0]);
+    r.check("UserPromptSubmit/accepted when the prompt is given", accepted, `posts: ${h.posts().map((p) => p.body?.hookEvent).join(", ")}`);
+    r.check("the accepted post names a turnId", accepted && accepted.body.turnId !== undefined && accepted.body.turnId !== "", "no turnId");
+    const end = await waitFor(() => h.hooks("Stop", "end_turn")[0]);
+    r.check("Stop/end_turn when the turn ends", end, "no Stop/end_turn");
+    r.check("accepted comes before the turn's end", accepted && end && accepted.at <= end.at, "out of order");
+    const all = h.posts();
+    const missing = all.flatMap((p) => ["sessionName", "nodeId", "runtime", "generation", "sequence", "occurredAt", "hookEvent", "subtype"]
+      .filter((k) => p.body?.[k] === undefined).map((k) => `${p.body?.hookEvent ?? "?"} lacks ${k}`));
+    r.check("every post carries the contract's fields", missing.length === 0, missing.join("; "));
+    r.check("every post uses the hook token", all.every((p) => p.token === h.token), "a post without the bearer token");
+    r.check("every post names this seat and preset",
+      all.every((p) => p.body?.sessionName === h.session && p.body?.nodeId === h.nodeId && p.body?.runtime === opts.name),
+      all.map((p) => `${p.body?.sessionName}/${p.body?.nodeId}/${p.body?.runtime}`).join(", "));
+    const seqs = all.map((p) => Number(p.body?.sequence));
+    r.check("sequence rises with each post", seqs.every((s, i) => Number.isFinite(s) && (i === 0 || s > seqs[i - 1])), `sequences: ${seqs.join(", ")}`);
+    const beats = all.filter((p) => p.body?.livenessSeconds !== undefined);
+    r.check("livenessSeconds, when given, is a whole number from 1 to 3600",
+      beats.every((p) => Number.isInteger(p.body.livenessSeconds) && p.body.livenessSeconds >= 1 && p.body.livenessSeconds <= 3600),
+      beats.map((p) => p.body.livenessSeconds).join(", "));
+    return r;
+  },
+
+  // Section 7's default is permit: when no one can be asked, the agent gets its own allow option at once, allow_once before
+  // allow_always. Only the seat's opt-in refusal (denied_by_seat), or a request with no allow option, is refused at once,
+  // shown, and held as needs input until the next prompt, even when the turn is cancelled.
+  async "4"() {
+    const r = result("4", "Permissions: permitted when no one can be asked; refused and held only on denied_by_seat or no allow option");
+    const option = (optionId, kind) => ({ optionId, name: optionId, kind });
+    const withAlways = [option("allow-once", "allow_once"), option("allow-always", "allow_always"), option("reject-once", "reject_once")];
+    const rejectOnly = [option("reject-once", "reject_once"), option("reject-always", "reject_always")];
+    const cells = [
+      // label, the stand-in's answered path (none: an OpenRig without it, which answers 404), the options, permitted, and
+      // whether /abort cancels the turn after the refusal
+      ["404", undefined, withAlways, true],
+      ["unreachable", { submit: { drop: true } }, withAlways, true],
+      ["denied_by_seat", { submit: { status: 409, code: "denied_by_seat" } }, withAlways, false],
+      ["denied_by_seat, then /abort", { submit: { status: 409, code: "denied_by_seat" } }, withAlways, false, true],
+      ["no allow option", undefined, rejectOnly, false],
+    ];
+    await Promise.all(cells.map(async ([label, asks, options, permitted, abort]) => {
+      const turns = [{ permission: { options }, holdMs: scale(abort ? 4000 : 1500) }, {}];
+      const h = await launch(`4-${label.replace(/[^a-z0-9]+/g, "-")}`, { scenario: { turns }, asks });
+      if (!r.check(`${label}: the runner opens its conversation`, await ready(h), "no launch record")) return;
+      h.paste(message(h, "edit the notes"));
+      const asked = await waitFor(() => h.sent("session/request_permission")[0]);
+      if (!r.check(`${label}: the agent's permission request is sent`, asked, "the turn never reached the request")) return;
+      // A runner may take a moment to find OpenRig unreachable; everything else is answered at once.
+      const answer = await waitFor(() => h.answerTo("session/request_permission"), scale(label === "unreachable" ? 10000 : 3000));
+      if (!r.check(`${label}: the request is answered, not left pending`, answer, "no answer")) return;
+      if (label !== "unreachable") r.check(`${label}: it's answered at once (within 2 s)`, answer.at - asked.at <= 2000, `${answer.at - asked.at} ms`);
+      const got = answer.msg.result?.outcome;
+      if (permitted) {
+        r.check(`${label}: the agent gets its allow_once option`, got?.outcome === "selected" && got.optionId === "allow-once",
+          `answer: ${JSON.stringify(answer.msg.result ?? answer.msg.error)}`);
+        r.check(`${label}: the turn's end is reported as usual`, await waitFor(() => h.hooks("Stop", "end_turn")[0], scale(4000)),
+          "no Stop/end_turn after the permit");
+        r.check(`${label}: nothing is held as needs input`, h.hooks("Notification", "permission_prompt").length === 0, "the runner posted permission_prompt");
+        return;
+      }
+      r.check(`${label}: the answer is a refusal, never an allow option`, got?.outcome === "cancelled",
+        `answer: ${JSON.stringify(answer.msg.result ?? answer.msg.error)}`);
+      const shown = await waitFor(() => h.hooks("Notification", "permission_prompt")[0], scale(3000));
+      r.check(`${label}: Notification/permission_prompt is posted`, shown, "no permission_prompt post");
+      if (label === "denied_by_seat") r.check(`${label}: what was asked is shown in the pane`, await waitFor(() => h.screen().includes("Edit notes.txt"), scale(3000)),
+        "the pane never showed the tool call's title, Edit notes.txt");
+      // A cancelled turn is still a turn's end: the refusal stays held until the next accepted prompt.
+      if (abort) {
+        h.paste("/abort");
+        r.check(`${label}: /abort sends session/cancel`, await waitFor(() => h.received("session/cancel")[0], scale(3000)), "no session/cancel");
+      }
+      await waitFor(() => h.log().some((e) => e.dir === "out" && e.msg.result?.stopReason), scale(abort ? 6000 : 4000));
+      await sleep(scale(800));
+      r.check(`${label}: no idle Stop or beat while holding that attention`, shown && nonExitStops(h, shown.at).length === 0,
+        `posted ${nonExitStops(h, shown?.at ?? 0).map((p) => `Stop/${p.body.subtype}`).join(", ")}`);
+      const next = Date.now();
+      h.paste(message(h, "carry on"));
+      r.check(`${label}: the next prompt is accepted as usual, which clears it`,
+        await waitFor(() => h.hooks("UserPromptSubmit", "accepted").some((p) => p.at >= next)), "no accepted post for the next prompt");
+    }));
+    return r;
+  },
+
+  // Section 7's answered path: the runner submits the live request, delivers the option that comes back (the person's
+  // answer, or OpenRig's permit) once while that request is live, and withdraws it when the turn or the agent ends.
+  async "4b"() {
+    const r = result("4b", "Answered permissions: the person's answer or OpenRig's permit reaches the live request, and nothing else does");
+    const options = ["allow-once", "reject-once"];
+    const submits = (h) => h.permissionCalls("POST").filter((p) => p.path === "/api/activity/permission-requests");
+    const withdraws = (h) => h.permissionCalls("POST", "/withdraw");
+    const answers = (h) => { const asked = h.sent("session/request_permission")[0];
+      return asked ? h.log().filter((e) => e.dir === "in" && e.msg.id === asked.msg.id && !e.msg.method) : []; };
+    const outcome = (h) => answers(h)[0]?.msg.result?.outcome;
+    const open = async (label, asks, turns = [{ permission: true }, {}]) => {
+      const h = await launch(`4b-${label}`, { scenario: { turns }, asks });
+      if (!await ready(h)) { r.check(`${label}: the runner opens its conversation`, false, "no launch record"); return undefined; }
+      h.paste(message(h, "edit the notes"));
+      await waitFor(() => h.sent("session/request_permission")[0]);
+      return h;
+    };
+
+    // An answer from the chosen person: the offered option, once, and the turn then reports as usual.
+    for (const [label, optionId] of [["allow", "allow-once"], ["deny", "reject-once"]]) {
+      const h = await open(label, { answer: { afterMs: 400, state: "answered", optionId } });
+      if (!h) continue;
+      const submitted = await waitFor(() => submits(h)[0], scale(3000));
+      if (!r.check(`${label}: the request is submitted to /api/activity/permission-requests`, submitted, "no submit")) continue;
+      const asked = h.sent("session/request_permission")[0];
+      const accepted = h.hooks("UserPromptSubmit", "accepted")[0];
+      const sent = submitted.body ?? {};
+      if (label === "allow") {
+        r.check("the submit passes OpenRig's checks", h.asks.submitted[0]?.problems.length === 0, h.asks.submitted[0]?.problems.join("; "));
+        r.check("it uses the hook token", submitted.token === h.token, "no bearer token");
+        r.check("requestId is the agent's JSON-RPC id", idText(sent.requestId) === String(asked?.msg.id), `requestId=${JSON.stringify(sent.requestId)}`);
+        r.check("turnId is the accepted prompt's", accepted && idText(sent.turnId) === String(accepted.body.turnId), `turnId=${JSON.stringify(sent.turnId)}`);
+        r.check("nativeRef and generation name this conversation and occupant", sent.nativeRef === h.record()?.nativeRef && sent.generation === h.generation,
+          `nativeRef=${sent.nativeRef} generation=${sent.generation}`);
+        r.check("the agent's options are passed on as offered", JSON.stringify((sent.options ?? []).map((o) => o.optionId)) === JSON.stringify(options),
+          JSON.stringify(sent.options));
+      }
+      const delivered = await waitFor(() => outcome(h), scale(4000));
+      r.check(`${label}: the agent gets the chosen option`, delivered?.outcome === "selected" && delivered.optionId === optionId, `got ${JSON.stringify(delivered)}`);
+      await waitFor(() => h.hooks("Stop", "end_turn")[0], scale(3000));
+      await sleep(scale(300));
+      r.check(`${label}: it's delivered once`, answers(h).length === 1, `${answers(h).length} answers`);
+      if (label === "allow") {
+        r.check("OpenRig records the prompt; the runner doesn't post permission_prompt itself", h.hooks("Notification", "permission_prompt").length === 0,
+          "the runner posted permission_prompt for a created ask");
+        const reads = h.permissionCalls("GET");
+        // Section 7 bounds the wait at 25 s and no lower; OpenRig's read takes 0 to 25 or none.
+        r.check("each read waits at most 25 s", reads.length > 0 && reads.every((p) => p.query.waitSeconds === undefined || Number(p.query.waitSeconds) <= 25),
+          reads.map((p) => p.query.waitSeconds ?? "(none)").join(", "));
+        r.check("after the answer, the turn's end is reported as usual", h.hooks("Stop", "end_turn").length > 0, "no Stop/end_turn after delivery");
+      }
+    }
+
+    // OpenRig's permit, at submit or at a read, is delivered like an answer. So is the runner's own permit when OpenRig
+    // can't give an answer: a refusal section 7's table marks "permits" (two of them 409s, like stale_generation, so the
+    // code decides, never the status; a revoked grant can still be the live occupant's, since one is reissued after a lost
+    // response), a read that gets no answer, or a status, code or read state the section doesn't list. A refusal saying
+    // the request isn't the current occupant's, and a refused, stale or withdrawn read, can't take an answer: the agent
+    // gets cancelled, and nothing is held.
+    const outcomes = [
+      ["permitted at submit", { submit: { permitted: true } }, "allow-once"],
+      ["permitted at read", { answer: { afterMs: 300, state: "permitted", optionId: "allow-once" } }, "allow-once"],
+      ["duplicate_request", { submit: { status: 409, code: "duplicate_request" } }, "allow-once"],
+      ["activity_grant_pending", { submit: { status: 409, code: "activity_grant_pending" } }, "allow-once"],
+      ["activity_grant_revoked", { submit: { status: 401, code: "activity_grant_revoked" } }, "allow-once"],
+      ["unreachable read", { read: { drop: true } }, "allow-once"],
+      ["unlisted read status (500)", { read: { status: 500, code: "internal_error" } }, "allow-once"],
+      ["unlisted read state (deferred)", { answer: { afterMs: 300, state: "deferred" } }, "allow-once"],
+      ["stale_generation", { submit: { status: 409, code: "stale_generation" } }, null],
+      ["activity_grant_stale read", { read: { status: 409, code: "activity_grant_stale" } }, null],
+      ["refused read (said_no)", { answer: { afterMs: 300, state: "refused", reason: "said_no" } }, null],
+      ["stale ask", { answer: { afterMs: 300, state: "stale" } }, null],
+      ["withdrawn ask", { answer: { afterMs: 300, state: "withdrawn" } }, null],
+    ];
+    await Promise.all(outcomes.map(async ([label, asks, optionId]) => {
+      const h = await open(label.replace(/[^a-z0-9]+/g, "-"), asks, [{ permission: true, holdMs: scale(800) }, {}]);
+      if (!h) return;
+      // Section 7 sets no time limit for what follows a read, so the check waits its ordinary per-step bound (--wait-ms),
+      // and longer when the read got no answer, which a runner may retry a few times before it permits.
+      const got = await waitFor(() => outcome(h), label === "unreachable read" ? scale(10000) : opts.waitMs);
+      if (!optionId) {
+        r.check(`${label}: the agent gets cancelled`, got?.outcome === "cancelled", `got ${JSON.stringify(got)}`);
+        r.check(`${label}: nothing is held`, await waitFor(() => h.hooks("Stop", "end_turn")[0], scale(3000))
+          && !h.hooks("Notification", "permission_prompt").length, "no Stop/end_turn after the cancel, or the runner posted permission_prompt");
+        return;
+      }
+      r.check(`${label}: the agent gets its allow_once option`, got?.outcome === "selected" && got.optionId === optionId, `got ${JSON.stringify(got)}`);
+      await waitFor(() => h.hooks("Stop", "end_turn")[0], scale(3000));
+      await sleep(scale(300));
+      r.check(`${label}: it's delivered once`, answers(h).length === 1, `${answers(h).length} answers`);
+      r.check(`${label}: the turn's end is reported as usual`, h.hooks("Stop", "end_turn").length > 0, "no Stop/end_turn after the permit");
+      if (label === "permitted at submit") r.check(`${label}: there's no ask to read`, h.permissionCalls("GET").length === 0,
+        `${h.permissionCalls("GET").length} reads`);
+      // A read that got no answer leaves OpenRig's ask open, so it must not stay in the person's Feed after the turn.
+      if (label === "unreachable read" || label.startsWith("unlisted read")) r.check(`${label}: the ask is still withdrawn at the turn's end`,
+        await waitFor(() => withdraws(h).some((w) => w.body?.reason === "turn_ended"), scale(3000)), `withdraws: ${JSON.stringify(withdraws(h).map((w) => w.body))}`);
+    }));
+
+    // No answer: the request stays pending and visible, a pasted message isn't an answer, and /abort withdraws it, after
+    // which a late answer isn't delivered.
+    {
+      const h = await open("pending", { late: { state: "answered", optionId: "allow-once" } });
+      if (h) {
+        await sleep(scale(1500));
+        r.check("pending: the runner keeps reading the ask", h.permissionCalls("GET").length >= 1, "no read of the ask");
+        r.check("pending: nothing is delivered while no one has answered", answers(h).length === 0, `got ${JSON.stringify(outcome(h))}`);
+        r.check("pending: no idle Stop while it waits", nonExitStops(h).length === 0, nonExitStops(h).map((p) => p.body.subtype).join(", "));
+        h.paste(message(h, "allow it"));
+        await sleep(scale(800));
+        r.check("pending: a pasted message isn't an answer", answers(h).length === 0, `got ${JSON.stringify(outcome(h))}`);
+        const abortedAt = Date.now();
+        h.paste("/abort");
+        const withdrawn = await waitFor(() => withdraws(h)[0], scale(3000));
+        r.check("pending: /abort withdraws the ask with turn_cancelled", withdrawn?.body?.reason === "turn_cancelled", `withdraw: ${JSON.stringify(withdrawn?.body)}`);
+        r.check("pending: the agent's request is answered cancelled", (await waitFor(() => outcome(h), scale(3000)))?.outcome === "cancelled",
+          `got ${JSON.stringify(outcome(h))}`);
+        // Once the ask is withdrawn no person is still waiting, so the cancelled turn's own end is reported as usual (the
+        // pasted message above runs as the next turn, so only that turn's Stop/cancelled counts). The contract doesn't
+        // order the Stop and the withdraw, so the Stop is any after /abort, not only one after the withdraw arrived.
+        r.check("pending: after /abort, the turn's end is reported, holding nothing",
+          await waitFor(() => h.hooks("Stop", "cancelled").some((p) => p.at >= abortedAt), scale(3000)), "no Stop/cancelled after /abort");
+        await sleep(scale(1200));
+        r.check("pending: an answer after the withdraw isn't delivered", answers(h).every((a) => a.msg.result?.outcome?.outcome !== "selected"),
+          `answers: ${JSON.stringify(answers(h).map((a) => a.msg.result))}`);
+      }
+    }
+    // The turn ending, or the agent exiting, while the ask waits also withdraws it.
+    for (const [label, permission, reason] of [["turn end", { waitMs: scale(1500) }, "turn_ended"], ["agent exit", { exitAfterMs: scale(1500) }, "exited"]]) {
+      const h = await open(label.replace(/[^a-z]+/g, "-"), {}, [{ permission }, {}]);
+      if (!h) continue;
+      const withdrawn = await waitFor(() => withdraws(h)[0], scale(4000));
+      r.check(`${label}: the ask is withdrawn with ${reason}`, withdrawn?.body?.reason === reason, `withdraw: ${JSON.stringify(withdrawn?.body)}`);
+      // The turn's own end: a Stop with the agent's own stop reason, after the agent answered that turn's session/prompt,
+      // on either side of the withdraw (the contract doesn't order them). A Stop posted before the turn ended doesn't count.
+      if (reason === "turn_ended") {
+        const turnEnd = await waitFor(() => {
+          const asked = h.sent("session/request_permission")[0];
+          const prompt = asked && h.received("session/prompt").filter((p) => p.at <= asked.at).at(-1);
+          return prompt && h.replyTo(prompt);
+        }, scale(4000));
+        const stopReason = turnEnd?.msg.result?.stopReason;
+        r.check(`${label}: nothing is held once the ask is withdrawn`,
+          stopReason && await waitFor(() => h.hooks("Stop", stopReason).some((p) => p.at >= turnEnd.at), scale(3000)),
+          `no Stop/${stopReason ?? "(no turn end)"} after the agent ended the turn`);
+      }
+    }
+    return r;
+  },
+
+  async "5"() {
+    const r = result("5", "Generation: every post carries the launch's occupant generation");
+    const h = await launch("5");
+    if (!r.check("the runner opens its conversation", await ready(h), "no launch record")) return r;
+    h.paste(message(h, "a short turn"));
+    await waitFor(() => h.hooks("Stop", "end_turn")[0]);
+    const all = h.posts();
+    r.check("the runner posted something", all.length > 0, "no posts");
+    r.check("generation is OPENRIG_OCCUPANT_GENERATION on every post", all.every((p) => p.body?.generation === h.generation),
+      all.map((p) => p.body?.generation).join(", "));
+    return r;
+  },
+
+  async "6"() {
+    const r = result("6", "Resume: the same conversation, or an honest missing or unconfirmed");
+    if (!opts.resume) return r.skip("the preset doesn't declare resume (--resume): OpenRig passes OPENRIG_RESUME_REF only to one that does");
+    const reopenCalls = (h) => [...h.received("session/load"), ...h.received("session/resume")];
+    {
+      const h = await launch("6-same", { scenario: { knownSessions: ["fake-old-1"] }, env: { OPENRIG_RESUME_REF: "fake-old-1" } });
+      const record = await ready(h);
+      const call = reopenCalls(h)[0];
+      r.check("same: the agent is asked to reopen OPENRIG_RESUME_REF", call?.msg.params?.sessionId === "fake-old-1",
+        call ? `asked for ${call.msg.params?.sessionId}` : "no session/load or session/resume");
+      r.check("same: in the launch's working folder", call && realpathSync(String(call.msg.params?.cwd ?? "")) === realpathSync(h.cwd),
+        `cwd=${call?.msg.params?.cwd}`);
+      r.check("same: the record reports that conversation", record?.nativeRef === "fake-old-1", `record: ${JSON.stringify(h.record())}`);
+      r.check("same: no new conversation starts", h.received("session/new").length === 0, "session/new was sent");
+    }
+    {
+      const h = await launch("6-missing", { scenario: { reopenError: "missing" }, env: { OPENRIG_RESUME_REF: "fake-gone" } });
+      const record = await waitFor(() => h.record());
+      r.check("missing: the agent's own not-found is reported as error: missing", record?.error === "missing" && !record.nativeRef,
+        `record: ${JSON.stringify(record)}`);
+      await sleep(scale(800));
+      r.check("missing: no new conversation starts", h.received("session/new").length === 0, "session/new was sent");
+    }
+    {
+      const h = await launch("6-other", { scenario: { reopenError: "error" }, env: { OPENRIG_RESUME_REF: "fake-elsewhere" } });
+      await waitFor(() => reopenCalls(h)[0] && h.log().some((e) => e.dir === "out" && e.msg.error));
+      await sleep(scale(2500));
+      const record = h.record();
+      r.check("other failure: not reported missing or resumed", !record || (record.error && record.error !== "missing" && !record.nativeRef),
+        `record: ${JSON.stringify(record)}`);
+      r.check("other failure: no new conversation starts", h.received("session/new").length === 0, "session/new was sent");
+    }
+    return r;
+  },
+
+  async "7"() {
+    const r = result("7", "Death mid-turn: Stop/<preset>_exited, never an idle report");
+    const h = await launch("7", { scenario: { turns: [{ exit: true }] } });
+    if (!r.check("the runner opens its conversation", await ready(h), "no launch record")) return r;
+    h.paste(message(h, "this turn kills the agent"));
+    const died = await waitFor(() => h.events("exit")[0]);
+    if (!r.check("the agent exits mid-turn", died, "the agent didn't exit")) return r;
+    const exited = await waitFor(() => h.hooks("Stop", `${opts.name}_exited`)[0], scale(4000));
+    r.check(`Stop/${opts.name}_exited is posted`, exited, `posts: ${h.posts().map((p) => `${p.body?.hookEvent}/${p.body?.subtype}`).join(", ")}`);
+    r.check("no idle Stop after the exit", nonExitStops(h, died.at).length === 0, nonExitStops(h, died.at).map((p) => p.body.subtype).join(", "));
+    return r;
+  },
+
+  async "7b"() {
+    const r = result("7b", "Errors: a failed prompt is a runtime error, not a permission request");
+    const h = await launch("7b", { scenario: { turns: [{ error: true }, {}] } });
+    if (!r.check("the runner opens its conversation", await ready(h), "no launch record")) return r;
+    h.paste(message(h, "this prompt fails"));
+    const posted = await waitFor(() => h.hooks("Notification", "runtime_error")[0]);
+    r.check("a failed session/prompt posts Notification/runtime_error", posted,
+      `posts: ${h.posts().map((p) => `${p.body?.hookEvent}/${p.body?.subtype}`).join(", ")}`);
+    r.check("it isn't reported as a permission request", h.hooks("Notification", "permission_prompt").length === 0, "permission_prompt posted");
+    return r;
+  },
+
+  // Section 9 marks this duty "(Proposed)", and a runner may list in `limits` what keeps running, so the result is reported
+  // on its own: never a pass or a fail, and never the exit status.
+  async "7c"() {
+    const r = result("7c", "Shutdown: closing the pane ends the agent (proposed in section 9; advisory)", { proposed: true });
+    const h = await launch("7c", { scenario: { stubborn: true } });
+    if (!r.check("the runner opens its conversation", await ready(h), "no launch record")) return r;
+    const pid = h.events("started")[0]?.pid;
+    h.stop();
+    const gone = await waitFor(() => { try { process.kill(pid, 0); return false; } catch { return true; } }, scale(5000));
+    r.check("closing the pane ends the agent process", gone, `agent pid ${pid} still running ${scale(5000) / 1000} s later; list it in limits if that's expected`);
+    if (!gone) try { process.kill(pid, "SIGKILL"); } catch { /* already gone */ }
+    return r;
+  },
+
+  async "8"() {
+    const r = result("8", "Unknown requests from the agent get -32601, and the runner carries on");
+    const h = await launch("8", { scenario: { unknownRequest: true } });
+    const answer = await waitFor(() => h.answerTo("fake/unknown_method"));
+    r.check("the unknown request is answered -32601", answer?.msg.error?.code === -32601, `answer: ${JSON.stringify(answer?.msg ?? null)}`);
+    if (!r.check("the runner opens its conversation", await ready(h), "no launch record")) return r;
+    h.paste(message(h, "still working?"));
+    r.check("the next prompt still goes through", await waitFor(() => h.prompts().length === 1), "no prompt after the unknown request");
+    return r;
+  },
+
+  async "9"() {
+    const r = result("9", "First input: OPENRIG_FIRST_INPUT's exact bytes are the first prompt");
+    if (!opts.firstInput) return r.skip("the preset doesn't declare first_input (--first-input): OpenRig passes no OPENRIG_FIRST_INPUT");
+    const dir = mkdtempSync(join(tmpdir(), "runner-check-first-"));
+    const file = join(dir, "first-input");
+    const text = "Read the plan first.\n\n  Then say ready.\n";
+    writeFileSync(file, text, { mode: 0o600 });
+    const h = await launch("9", { env: { OPENRIG_FIRST_INPUT: file } });
+    const first = await waitFor(() => h.prompts()[0]);
+    r.check("the first prompt is the file's bytes, newlines and spaces included", first === text, `got ${JSON.stringify(first)}`);
+    if (!opts.keep) rmSync(dir, { recursive: true, force: true });
+    return r;
+  },
+};
+
+const order = ["1", "1b", "2", "3", "4", "4b", "5", "6", "7", "7b", "7c", "8", "9"];
+const selected = opts.cases ?? order;
+const unknown = selected.filter((id) => !order.includes(id));
+if (unknown.length) { console.error(`unknown case ${unknown.join(", ")}; cases are ${order.join(", ")}`); process.exit(2); }
+let results;
+try {
+  results = await Promise.all(order.filter((id) => selected.includes(id)).map(async (id) => {
+    try { return await cases[id](); } catch (error) { return { id, title: id, status: "fail", checks: [{ name: "the case ran", ok: false, detail: error.message }] }; }
+  }));
+} finally {
+  tmux("kill-server");
+  server.close();
+  // A runner or agent can still be writing its last lines as its pane closes, so cleanup waits a moment and never fails the report.
+  await sleep(scale(300));
+  if (!opts.keep) for (const h of launches) try { rmSync(h.dir, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 }); } catch { /* left in the temp folder */ }
+}
+
+// A proposed duty's result is advisory: it says whether the runner met it, outside pass, fail and the exit status.
+for (const r of results) if (r.proposed && r.status !== "skip") { r.met = r.status === "pass"; r.status = "advisory"; }
+const summary = Object.fromEntries(["pass", "fail", "skip", "advisory"].map((s) => [s, results.filter((r) => r.status === s).length]));
+if (opts.json) {
+  console.log(JSON.stringify({ check: "openrig-runner-conformance", format: 1, runner: opts.runner, name: opts.name, results, summary }, null, 2));
+} else {
+  console.log(`OpenRig runner conformance check (docs/reference/harness-runner-contract.md)\nrunner: ${opts.runner}\npreset: ${opts.name}\n`);
+  for (const r of results) {
+    const label = r.status === "advisory" ? "NOTE" : r.status.toUpperCase();
+    const outcome = r.status === "advisory" ? (r.met ? ": met" : ": not met") : r.reason ? `: ${r.reason}` : "";
+    console.log(`${label.padEnd(5)} ${r.id.padEnd(3)} ${r.title}${outcome}`);
+    for (const c of r.checks.filter((c) => !c.ok)) console.log(`      x ${c.name}: ${c.detail}`);
+  }
+  console.log(`\n${summary.pass} pass, ${summary.fail} fail, ${summary.skip} skipped, ${summary.advisory} advisory (not counted)${opts.keep ? `; launch folders kept under ${tmpdir()}` : ""}`);
+}
+process.exit(summary.fail ? 1 : 0);

docs/reference/runner-conformance/fake-acp-agent.mjs

@@ -0,0 +1,143 @@
+#!/usr/bin/env node
+// A scriptable fake ACP agent for the runner conformance check (check-runner.mjs). It speaks just enough of ACP v1 over
+// stdio, newline-delimited JSON-RPC, to exercise a runner without a model, credentials or the network. It was written
+// after the fake agent in bpamiri/dsh-seat (Apache-2.0), which showed the approach.
+//
+//   node fake-acp-agent.mjs --scenario <file.json> --log <file.jsonl> [any arguments the runner adds are ignored]
+//
+// The scenario says how to answer; the log records every message in and out, so the check can compare them.
+//   sessionId        the id `session/new` returns (default "fake-session-1")
+//   initialize       "error" or "exit" to fail the handshake there
+//   newSession       "error" or "exit" to fail `session/new`
+//   newDelayMs       wait this long before answering `session/new`
+//   knownSessions    ids `session/load` and `session/resume` reopen
+//   reopenError      "missing" (the agent's own "no such session") or "error", for an id it doesn't know
+//   unknownRequest   send the runner a request with a method it can't know, once, after `initialize`
+//   stubborn         keep running after a hangup or after input closes; only SIGTERM or SIGKILL ends it
+//   turns            one entry per `session/prompt`, in order (the last one repeats):
+//     holdMs         keep the turn open this long; `session/cancel` ends it early as "cancelled"
+//     ignoreCancel   on `session/cancel`, keep the turn open and end it with stopReason (default end_turn) anyway
+//     permission     ask `session/request_permission` first, then carry on whatever the answer. As an object:
+//                      options      the options offered (default: allow-once, allow_once; reject-once, reject_once)
+//                      waitMs       stop waiting for the answer after this long, and end the turn
+//                      exitAfterMs  exit with status 3 this long after asking, while the request waits
+//                    A `session/cancel` while the request waits ends the turn "cancelled" once it's answered.
+//     error          answer the prompt with a JSON-RPC error
+//     exit           exit with status 3 instead of answering
+//     stopReason     the stop reason to return (default "end_turn")
+import { appendFileSync, readFileSync } from "node:fs";
+import { createInterface } from "node:readline";
+
+const arg = (name) => { const i = process.argv.indexOf(name); return i > 0 ? process.argv[i + 1] : undefined; };
+const scenarioFile = arg("--scenario") ?? process.env.RUNNER_CHECK_SCENARIO;
+const logFile = arg("--log") ?? process.env.RUNNER_CHECK_AGENT_LOG;
+const scenario = scenarioFile ? JSON.parse(readFileSync(scenarioFile, "utf8")) : {};
+
+const log = (entry) => { if (logFile) appendFileSync(logFile, `${JSON.stringify({ at: Date.now(), ...entry })}\n`); };
+const send = (msg) => { log({ dir: "out", msg }); process.stdout.write(`${JSON.stringify(msg)}\n`); };
+const reply = (id, result) => send({ jsonrpc: "2.0", id, result });
+const fail = (id, code, message) => send({ jsonrpc: "2.0", id, error: { code, message } });
+const exit = (status, why) => { log({ dir: "event", event: "exit", status, why }); process.exit(status); };
+const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
+
+log({ dir: "event", event: "started", pid: process.pid, cwd: process.cwd(), argv: process.argv.slice(2) });
+// Ctrl-C is the runner's to handle, so the agent never takes it as a signal. A stubborn agent also outlives a hangup and
+// its input closing, as some real agents do: only the runner ending it (SIGTERM or SIGKILL) stops it.
+process.on("SIGINT", () => log({ dir: "event", event: "signal", signal: "SIGINT" }));
+process.on("SIGHUP", () => scenario.stubborn ? log({ dir: "event", event: "signal", signal: "SIGHUP" }) : exit(0, "SIGHUP"));
+process.on("SIGTERM", () => exit(0, "SIGTERM"));
+
+let sessionId;
+let turn = 0;
+let nextId = 1;
+const answers = new Map();
+let cancelTurn = null;
+
+// Numbers, as JSON-RPC agents commonly use. A runner passes one on unchanged, and OpenRig reads it as its decimal string.
+function ask(method, params) {
+  const id = 1000 + nextId++;
+  send({ jsonrpc: "2.0", id, method, params });
+  return new Promise((resolve) => answers.set(id, resolve));
+}
+
+async function prompt(id, params) {
+  const spec = (scenario.turns ?? [])[Math.min(turn, (scenario.turns?.length ?? 1) - 1)] ?? {};
+  turn += 1;
+  if (spec.exit) exit(3, `turn ${turn}`);
+  if (spec.error) { fail(id, -32603, "fake agent: this turn fails"); return; }
+  send({ jsonrpc: "2.0", method: "session/update",
+    params: { sessionId: params.sessionId, update: { sessionUpdate: "agent_message_chunk", content: { type: "text", text: `turn ${turn}` } } } });
+  let stopReason = spec.stopReason ?? "end_turn";
+  if (spec.permission) {
+    const p = typeof spec.permission === "object" ? spec.permission : {};
+    const asked = ask("session/request_permission", {
+      sessionId: params.sessionId,
+      toolCall: { toolCallId: `call-${turn}`, title: "Edit notes.txt", kind: "edit" },
+      options: p.options ?? [
+        { optionId: "allow-once", name: "Allow once", kind: "allow_once" },
+        { optionId: "reject-once", name: "Reject", kind: "reject_once" },
+      ],
+    });
+    if (p.exitAfterMs) setTimeout(() => exit(3, "while a permission request waits"), p.exitAfterMs);
+    let cancelled = false;
+    cancelTurn = () => { cancelled = true; };
+    const answer = await (p.waitMs ? Promise.race([asked, sleep(p.waitMs).then(() => null)]) : asked);
+    cancelTurn = null;
+    if (!answer) log({ dir: "event", event: "permission-wait-ended" });
+    if (cancelled) { reply(id, { stopReason: "cancelled" }); return; }
+  }
+  if (spec.holdMs) {
+    const cancelled = await new Promise((resolve) => {
+      const timer = setTimeout(() => { cancelTurn = null; resolve(false); }, spec.holdMs);
+      cancelTurn = () => { if (spec.ignoreCancel) return; clearTimeout(timer); cancelTurn = null; resolve(true); };
+    });
+    if (cancelled) stopReason = "cancelled";
+  }
+  reply(id, { stopReason });
+}
+
+async function reopen(id, params) {
+  if ((scenario.knownSessions ?? []).includes(params.sessionId)) { sessionId = params.sessionId; reply(id, null); return; }
+  if (scenario.reopenError === "missing") fail(id, -32002, `Session not found: ${params.sessionId}`);
+  else fail(id, -32603, "fake agent: reopen failed");
+}
+
+async function handle(msg) {
+  if (msg.id !== undefined && msg.method === undefined) { answers.get(msg.id)?.(msg); answers.delete(msg.id); return; }
+  switch (msg.method) {
+    case "initialize":
+      if (scenario.initialize === "exit") exit(2, "initialize");
+      if (scenario.initialize === "error") { fail(msg.id, -32603, "fake agent: initialize fails"); return; }
+      reply(msg.id, { protocolVersion: 1, agentCapabilities: { loadSession: true }, authMethods: [] });
+      if (scenario.unknownRequest) await ask("fake/unknown_method", { note: "a method no runner knows" });
+      return;
+    case "session/new":
+      if (scenario.newDelayMs) await sleep(scenario.newDelayMs);
+      if (scenario.newSession === "exit") exit(2, "session/new");
+      if (scenario.newSession === "error") { fail(msg.id, -32603, "fake agent: session/new fails"); return; }
+      sessionId = scenario.sessionId ?? "fake-session-1";
+      reply(msg.id, { sessionId });
+      return;
+    case "session/load":
+    case "session/resume":
+      await reopen(msg.id, msg.params ?? {});
+      return;
+    case "session/prompt":
+      await prompt(msg.id, msg.params ?? {});
+      return;
+    case "session/cancel":
+      cancelTurn?.();
+      return;
+    default:
+      if (msg.id !== undefined) fail(msg.id, -32601, `Method not found: ${msg.method}`);
+  }
+}
+
+const lines = createInterface({ input: process.stdin });
+lines.on("line", (line) => {
+  let msg;
+  try { msg = JSON.parse(line); } catch { log({ dir: "event", event: "garbage", line }); return; }
+  log({ dir: "in", msg });
+  void handle(msg);
+});
+lines.on("close", () => { if (scenario.stubborn) { log({ dir: "event", event: "stdin-closed" }); setInterval(() => {}, 60_000); } else exit(0, "stdin closed"); });

docs/reference/sdlc-conventions.md

@@ -165,6 +165,12 @@ commit. Evidence capture (`proof add`), policy acceptance, higher outcome
 judgment and publication are distinct. This optional component does not select
 Part B or add a mandatory review gate.
 
+A judgment binds the slice's path, the item's text and the policy. Moving or
+renaming a slice (`rig scope slice move`, or by hand) reopens its judged items:
+`rig proof show` reports each as `unknown`, and each needs a new judgment on its
+current revision. `rig scope audit` checks structure only and never reads proof
+readiness, so a passing audit doesn't mean the proof still stands.
+
 The project readiness that `rig proof show --project` reports counts only
 missions whose `mission.yaml` carries `metadata.status: active` or
 `metadata.status: release-candidate`. `rig scope mission create` writes

docs/reference/seat-delivery.md

@@ -16,6 +16,12 @@ type: `openrig-urgent`, `urgent-only` and `sends-only`. Sibling seats keep their
 
 Lifecycle work is refused at `openrig-urgent` and `urgent-only`, as at `hold`.
 
+A terminal member (`runtime: terminal`) defaults to `sends-only`, which holds OpenRig's own notices. Set one
+to `off` and those notices, such as park reminders and other watchdog messages, type into whatever the pane
+is running, and a shell prompt runs each line as a command, From and To lines included: OpenRig doesn't check
+a terminal pane's foreground process. `openrig-urgent` and `draft-aware` type only into a recognized empty agent composer, so
+a shell prompt holds them. Keep the default for a seat that may fall back to a shell.
+
 Runtime permissions are separate. Recipient and interactive-prompt checks still
 apply. `--raw` and `--force` do not bypass the typing guard. A deliberate
 `--dangerously-interact --reason` remains available at every interruption level

docs/reference/slack-app-setup.md

@@ -109,6 +109,14 @@ reaction on any other message, including a person's reply in the thread, is igno
 `reaction_added` event: add both under **OAuth & Permissions** and **Event Subscriptions**, then
 reinstall the app to the workspace.
 
+To have reactions ignored instead, run `rig slack setup --reactions off`. The agent that asked
+never hears of a reaction, including one still waiting to be retried, and a person answers an ask by
+typing a reply in its thread. Slack still sends the event and the manifest keeps it, so switching
+back with `--reactions route` needs no change to the app. Like any saved change, the setting reaches
+a running connector when it next rewires, which a daemon restart does (`rig daemon stop`, then
+`rig daemon start`). See [One channel per rig or seat](#one-channel-per-rig-or-seat-optional) for
+why disabling and re-enabling the connector isn't the same.
+
 The manifest also turns on **Interactivity**, so the human can answer a decision's structured
 questions by clicking a button (`rig queue create --human-questions-file`). In Socket Mode the
 clicks arrive over the same socket, so no request URL is needed. An app created from an older

docs/reference/terminal-sessions.md

@@ -30,16 +30,44 @@ With neither a provider nor a session, the daemon's open and preview APIs try
 a live Herdr first, then a live cmux. An explicit session always selects Herdr
 and skips that fallback, including when the daemon's default Herdr endpoint is down.
 
-An explicit session resolves under the daemon user's Herdr session directory;
-it takes precedence over that daemon's default socket override. The selection
-belongs to the request. Saved views still describe membership and layout, so
-pass `--session` again when reopening the view in that session.
+Session paths use the daemon's `XDG_CONFIG_HOME` when set, or `HOME/.config`
+when unset. Relative paths (including an empty value) resolve from the daemon's
+startup directory, matching Herdr's configuration selection. Named sessions live under
+`herdr/sessions/<name>/herdr.sock`; `default` uses `herdr/herdr.sock` directly.
+These settings are captured when the daemon starts.
+
+If the requested name equals the daemon's `HERDR_SESSION`, it keeps that
+session's explicit `HERDR_SOCKET_PATH`. A different requested name uses the
+configuration-tree path, without reusing the override or inferring a session
+from an arbitrary socket path. With an explicit socket but no configured
+session name, omit `--session` to keep that exact endpoint. A missing selected
+socket never falls back to another session. Viewer and follow operations carry
+that exact socket to the CLI through `HERDR_SOCKET_PATH`, without an explicit
+CLI `--session` that would override it. They check that the CLI reports the
+selected socket before changing its panes. Daemon-side CLI calls retain the
+startup environment; desktop follow uses the socket confirmed by the daemon.
+
+The selection belongs to the request. Saved views still describe membership
+and layout, so pass `--session` again when reopening the view in that session.
 
 The existing view rules apply: saved views, rig and pod views, read-only mission
 and slice views, pagination, absent seats and partial results all use the same
 composer. Selecting a session changes where those tiles are placed. It does
 not relaunch the seats' tmux sessions or move queue ownership.
 
+## Open results
+
+For Herdr, a successful layout request alone does not count its seats as opened.
+OpenRig checks the returned tab's pane list. A missing tab identity or an
+unreadable, empty or unattributable listing puts the affected seats in
+`degraded`, with `code: "pane_unconfirmed"`, instead of `opened`. The CLI and
+TUI name those seats as unconfirmed. Inspect the terminal before retrying:
+the layout may have taken effect even though its panes could not be confirmed.
+Confirmed seats on other pages still count; with none confirmed, the command
+returns a nonzero exit status. Reusing an existing workspace keeps its separate
+reuse result. This pane check does not prove desktop visibility or create a
+verified viewer mapping.
+
 ## Replacing a seat's session
 
 New tiles keep their place when a seat's tmux session is replaced. They wait up
@@ -156,7 +184,7 @@ GET /api/terminal/preview?view=qitem:<id>&provider=herdr
 POST /api/terminal/open {"view":"qitem:<id>","provider":"herdr","expectedPlan":"<preview planId>"}
 ```
 
-Preview and open include `task: {qitemId, name, complete, issues, members}`. Each member
+Preview and open include `task: {kind:"qitem", id, qitemId, name, complete, issues, members}`. Each member
 names its structured `{hostId, nodeId}` key (null when unresolved), seat,
 contributing `obligations`, `availability` and `viewerStatus`. Stopped, missing,
 external and unresolved remote owners remain visible. Remote queues are not
@@ -180,7 +208,263 @@ its name, falling back to the qitem ID. The existing preview `view` string is
 unchanged. A following view derives membership when requested; it does not
 mean an open workspace updates itself.
 
-Preview is read-only. In this membership increment, `viewerStatus` is
-`not-provisioned`: the response does not verify a reusable provider terminal
+Preview is read-only. A member without owned viewer provenance is
+`not-provisioned`; the ensure operation below can establish a reusable provider
 mapping. An explicit open retains the provider's existing workspace behavior;
 it does not continuously reconcile the view or change an agent's lifecycle.
+
+## Following one task
+
+`rig terminal open qitem:<id> --follow` keeps one workspace following the
+task's current obligation owners. A joining member gets a verified local
+viewer tile without taking focus. A departing member loses only its viewer,
+after a grace period; the agent and its conversation keep running. The
+explanation strip names joins, departures, unavailable terminals and limits.
+The first workspace on an empty provider may become focused.
+
+This increment supports the qualified **Linux Herdr 0.9.5/protocol 22**
+endpoint. Other platforms/providers and other selector kinds retain ordinary
+one-shot open. Use `--provider herdr --json` to operate without opening a
+desktop window, and `--session <name>` to select that endpoint. Opening a tile
+can change the agent's terminal size; `ignore-size` does not guarantee isolation.
+
+```bash
+rig terminal open qitem:example --follow --provider herdr --json
+rig terminal open qitem:example --follow --provider herdr --pin reviewer@team
+rig terminal open qitem:example --follow --provider herdr --max-tiles 12
+rig terminal open qitem:example --follow --provider herdr --suppress reviewer@team --suppression-ms 60000
+rig terminal open qitem:example --no-follow
+```
+
+`--unpin` and `--unsuppress` reverse those presentation choices. The default
+cap is eight tiles; explicitly raising `--max-tiles` overrides it. Pins can
+exceed the cap rather than being evicted. A departed tile retained for typed or
+unconfirmed input does not consume a slot for joining members. Disabling removes
+the saved following selector and keeps the existing layout intact.
+
+The default departure grace is four seconds (`--grace-ms`). A focused leaver
+gets another twelve seconds for reading (`--focus-retention-ms`); before its
+tile closes, focus moves to a verified surviving member, or to the explanation
+strip when none remains. Typed or unconfirmed input keeps the tile, labeled
+as no longer participating. Explicit suppression also retains a typed draft.
+Pi, terminal and community harness input may be unreadable to the composer
+inspector: those tiles remain while input is unconfirmed, even after unpinning.
+Each input-retained departure gets one delayed recheck after thirty seconds.
+After that, queue or node/session lifecycle events, or an explicit repeat of
+the follow command, can reconsider it. Activity hooks do not wake following,
+and input is not repeatedly polled. Draft text is never saved.
+
+Following consumes Core events and one-shot retention/suppression deadlines,
+not repeated CLI membership queries. Complete empty membership drains the
+tiles and preserves the workspace. Once no retained tiles remain, following
+stops and the selector releases its active slot; new work does not automatically
+restart it. Run `--follow` again to resume the same workspace. An incomplete source, failed provider read
+or changed pane/terminal generation retains the previous view and reports
+the uncertainty. It never authorizes a blind workspace rebuild or close.
+`rig terminal views --json` and `rig terminal status --json` include the saved
+active selectors and latest following status. Active selectors and owned viewer provenance
+survive daemon restarts; task membership is always re-derived.
+
+The owner-authorized HTTP operation is `POST /api/terminal/follow` with
+`{view:"qitem:<id>",provider:"herdr",session?,enabled:true,maxTiles?,pin?,unpin?,
+suppress?,unsuppress?,suppressionMs?,graceMs?,focusRetentionMs?,expectedPlan?}`.
+Seat controls are arrays of names from the current/retained view. It uses the
+existing terminal bearer boundary, like ensure-viewer. The response includes
+the stable `workspaceId`, normal open partitions, `viewDescriptor`, and
+`following:{id,enabled,state,maxTiles,tiles,events,overflow,nextAt?}`. Presentation
+provenance grants no membership or input authority. This is a local native
+viewer operation; browser input and remote-host provisioning are separate.
+
+## Ensuring an owner viewer
+
+The owner-facing HTTP operation `POST /api/terminal/viewer/ensure` accepts
+`{view:"qitem:<id>", member:{hostId,nodeId}, provider:"herdr", session,
+expectedPlan}`. Use a current preview's member key and plan. This explicit
+operation uses the daemon's existing terminal bearer and browser boundary;
+a bridge must retain its signed-in owner, Origin and expiry controls. It does
+not grant authority based on task membership or accept a caller terminal ID.
+
+On the qualified Linux Herdr 0.9.5/protocol 22 endpoint, it creates one viewer
+or reuses an owned, verified `{workspaceId,paneId,terminalId,readOnly}` mapping.
+A repeated/concurrent request does not create another viewer. A lost create
+reply is reconciled before starting anything else; a lost run reply is checked
+against the actual foreground process, never blindly repeated. Unknown or
+changed terminal identity remains unconfirmed. The operation does not relayout,
+close departed viewers, take over control, focus existing panes or launch/stop
+agents. A first workspace on an otherwise empty Herdr server may become focused.
+
+Preview remains passive and names `not-provisioned`, `verified` or `unconfirmed`
+viewer status. Studio can ensure a new member, add its browser frame, and retain
+a pinned leaver labeled as no longer participating. The provider mapping is
+small presentation provenance, separate from queue membership. Herdr display
+labels are not that provenance. Other providers/platforms remain unsupported
+for ensure-viewer until qualified; ordinary open retains its existing behavior.
+
+Writable tile clients use `tmux attach -f ignore-size`; read-only clients keep
+`-r` (which also implies ignore-size). This can avoid contributing to sizing
+when other clients establish the agent window's dimensions. It is **not resize
+isolation**: opening a tile can change the agent's terminal size when it is the
+only sizing client. Observers of a Herdr terminal do not themselves resize it;
+a controlling browser can. No cross-version resize guarantee is implied.
+
+
+## Following a slice or mission's current work
+
+Use `rig terminal open slice:<name-or-id> --follow --provider herdr` to keep
+one workspace following the current destination owners of that slice's active
+queue rows. `pending`, `in-progress` and `blocked` rows count. A member can own
+several rows; `task.members[].obligations` preserves every row and state.
+
+A slice resolves within the daemon's configured slice index, by exact directory
+name or unique frontmatter `id`. Tags must name it exactly: `rig queue create
+--slice <name-or-id> ...` associates a row. Handoff inherits those tags unless
+replaced; merely linking an untagged child does not associate it. A mission-only
+tag does not put a row in every slice. No project is inferred from a tag or the
+caller's working directory. A bare name or ID claimed by multiple slices is
+`scope_ambiguous`; no first-match scope is selected. Selecting a unique key can
+still follow a slice whose other alias is reused: a row carrying that colliding
+alias must have exactly one `mission:` tag naming exactly one indexed claimant.
+Rows placed in another mission are excluded. Missing, contradictory or otherwise
+unplaceable mission/alias tags make the snapshot incomplete, retaining previous
+viewers. A unique alias on its own retains exact matching. Flat legacy slices
+without an indexed parent mission cannot disambiguate a colliding alias.
+
+Mission views use `mission:<directory-id>`, from the same configured index.
+A mission is known when its directory contains `slices/` (even empty) or a
+readable `mission.yaml` declaration. Its directory ID is its exact identity;
+mission frontmatter is not an alias. Membership is the union of active rows
+with that exact `mission:` tag and rows associated with any physically indexed
+child slice by its exact name or unique ID. Each row counts once even when it
+has both tags. Flat legacy slices without that parent association are not
+inferred into a mission. A child's historical status or composition activation
+flag does not remove its current obligations from the union. Duplicate mission
+IDs are ambiguous. Shared child names/IDs are omitted from bare slice matching;
+rows with one indexed claimant mission are placed through their mission tag.
+A direct mission tag takes precedence, including work tagged to both missions.
+Without that direct tag, a shared-key row with missing, contradictory or
+non-claimant mission tags is named in `issues` and makes the snapshot incomplete;
+known owners remain.
+The launcher reads this same terminal preview and submits its `expectedPlan`
+when opening, rather than using the review-agent roster.
+
+Slice and mission keys are case-sensitive and can include Unicode, plus signs and internal
+spaces. They cannot be empty or contain slash, backslash, control characters,
+or leading/trailing whitespace in the scope reader. The service trims the whole
+selector before composition, so a trailing selector space is accepted by
+preview/ensure with the same plan; ensure rejects a leading selector space.
+Prefer the exact canonical selector in every call. Pass it in JSON or a quoted
+CLI argument; encode it once as a GET query value. Do not double-decode it.
+
+Slice previews and open results use `task: {kind:"slice", id, name, complete,
+issues, members}`. `id` is the unique frontmatter id, or the directory name when
+no id is present. There is no synthetic `qitemId`. Preserve the preview's `view`
+for the ensure request. All successful previews, including qitem previews,
+include `capabilities: {ensureViewer:boolean}`. True means that the resolved
+view/provider implements ensure; the existing member, completeness, owner and
+`expectedPlan` checks still apply. A missing capability/task on an older daemon
+is unsupported scoped membership, not an empty team.
+
+Mission results use the same shape with `task.kind:"mission"`; `task.id` and
+`task.name` are the directory ID, with no synthetic `qitemId`. Ordinary mission
+preview/open uses this current-obligation membership too: past participants
+without active obligations are excluded, while unavailable current owners
+remain named. Follow, preview and ensure share the same authority.
+
+New slice and mission workspaces default to four automatic tiles; `--max-tiles` overrides
+that. Qitem views retain their eight-tile default. The full membership and
+obligation list remains available when tiles exceed capacity. Pins, drafts,
+unknown input and focus retention still protect departed viewers. Closing a
+viewer never stops its agent. Opening a tile can change the agent's dimensions.
+
+A known empty slice or mission remains enabled in the same workspace; later associated
+work can join it. An incomplete queue read (`task.complete:false` plus `issues`)
+retains previous viewers. A missing scope is `scope_not_found`, duplicate scope
+identity is `scope_ambiguous`, unreadable/capped index data is `scope_unavailable`,
+and an unavailable resolver is `scope_unsupported`. None is authoritative empty.
+Ensure of an incomplete snapshot returns `viewer_unconfirmed`; changed plans
+return `preview_changed` before provisioning.
+
+Ordinary ownership changes follow queue/session/node events. Node and directory
+changes under the primary configured index use the existing filesystem watcher,
+with a coalesced scope invalidation event. There is no repeating membership
+poll. Structural edits under additional legacy roots, or a filesystem where
+watching is unavailable, require `rig`'s existing slice refresh endpoint
+`POST /api/slices/refresh` or reopening the view. Missing or unreadable sources
+keep prior viewers; they never certify that the work disappeared.
+
+
+Scope identity reads are bounded to 4,096 directory entries, 16 KiB per selected
+node-header read, and 8 MiB of source bytes per index generation. The byte budget
+counts the bounded read, including text after short frontmatter. An index that
+exceeds these limits is too large for this read; use a smaller configured index
+or a qitem view. This is separate from unreadable or malformed sources.
+Queue membership reads are bounded to 50,000 active rows plus one lookahead,
+16 KiB of encoded tags per row and 8 MiB of aggregate encoded tags. They remain
+fresh on each render. Identities are reused within a scope-source generation
+only while the primary-root watcher is live and existing additional roots are
+covered by that root. A source event or explicit refresh invalidates them;
+errors discard the cache. Unavailable watching or an existing unwatched legacy
+root uses fresh bounded identity reads instead. Missing legacy roots are checked
+on each call so an appearing root cannot reuse an older cached identity set.
+
+Error bodies remain authoritative. `scope_unsupported` is HTTP 400,
+`scope_not_found` is 404, and `scope_ambiguous`/`scope_unavailable` are 409.
+An incomplete queue snapshot is a successful preview with `task.complete:false`;
+ensure returns `viewer_unconfirmed` without a provider effect. Unsupported
+ensure view kinds now return `view_required` (400), where they previously returned
+200 with an error body. Unsupported follow view kinds
+retain `follow_view_required` in a 200 `ok:false` body, as before. A changed
+preview remains `preview_changed` (409). During automatic refresh a missing or
+ambiguous scope retains the existing tiles and names the source problem in the
+explanation strip; refresh or reopen it under the current unambiguous key.
+
+
+## Reading or sending to an expected native conversation
+
+An application that selects an original Claude Code or Codex conversation can opt
+into an expected-identity check. The identity is the exact tuple `nodeId`, canonical
+`sessionName`, `runtime` (`claude-code` or `codex`) and native `nativeId`. A registry
+resume token by itself is not a current-process identity observation.
+
+Read `GET /api/sessions/:sessionName/generation-record-expected` with query fields
+`expectedNodeId`, `expectedRuntime`, `expectedNativeId` and optional `sinceBytes`.
+It resolves the verified current occupant and native process, opens that native
+record, checks its identity, and returns identity and bounded suffix bytes from
+the same opened file. Its existing byte cursor and `generationId` semantics match
+`generation-record`. Success adds `identityContract:1` and `checkedIdentity`, the
+exact expected tuple. Check both before displaying the suffix as that conversation.
+
+Send `POST /api/transport/send-expected` with the ordinary send body and a required
+`expectedIdentity` object containing those four fields. The check runs again after
+buffer preparation immediately before paste, and before Enter after any wait.
+A retained input stores the constraint in its outbox receipt; committed dispatch
+and reuse of its delivery ID cannot select a different original conversation or
+omit the stored constraint. Delivery IDs still do not deduplicate explicit input.
+Success adds `identityContract:1` and `checkedIdentity`. Retained is not sent;
+`checkedIdentity` is an observation at an input boundary, not model consumption.
+If replacement is observed after paste, the result reports `inputPhase:"written"`
+and `sent:true`, with Enter refused. Inspect that input before deciding what to do.
+
+Malformed expectations return 400 (`invalid_expected_identity`). A different
+observed identity returns 409 (`expected_identity_mismatch`); a missing or
+unverified identity returns 409 (`expected_identity_unavailable`). Conflicting
+reuse of a delivery ID returns 409 (`delivery_identity_conflict`). No conditional
+operation falls back to the replacement. Use these dedicated paths when detecting
+support: older daemons return 404 instead of silently ignoring an unknown field.
+Ordinary reads and sends without an expectation or an ID bound by an earlier
+conditional send keep their existing behavior. Reusing a bound ID also carries
+its stored constraint when the caller omits the expectation.
+
+While a runtime switch is preparing or confirmed, a conditional send returns
+409 `expected_identity_unavailable` with `sent:false` and `inputPhase:"not_reached"`.
+It retains neither the input nor the expectation. Retry the conditional endpoint
+with the original `expectedIdentity`; an ordinary send is not an identity-bound
+retry unless its ID already has a stored constraint. Ordinary sends during the
+switch keep their existing retained response.
+
+This is a finite process/record observation followed by a terminal write; tmux has
+no atomic native-conversation compare-and-write operation. A native conversation
+change after the final observation remains a race. The check does not turn a
+terminal attachment into a conversation pin: existing terminal viewers follow
+managed pane/session bindings and may display their replacement. Applications
+must keep that attachment limit separate from checked record reads and sends.

docs/reference/tmux-servers.md

@@ -0,0 +1,166 @@
+# Choosing the tmux server for an OpenRig instance
+
+`terminal.tmux_server` controls where the daemon creates and reads managed seats.
+It accepts `inherit` (the default) or `isolated`. The environment override is
+`OPENRIG_TMUX_SERVER`.
+
+To keep managed seats out of your personal `tmux ls`, set this before starting a
+new instance, or before a deliberate restart of an existing daemon:
+
+```sh
+rig config set terminal.tmux_server isolated
+```
+
+The setting is read once at daemon startup. Changing it does not move a running
+seat. Inspect the effective choice with:
+
+```sh
+rig terminal status --json
+```
+
+The `tmux` object reports `requestedMode` (read at the last startup), effective
+`mode`, `socketPath`, `serverArgs`, `migrationPending`, and `blockingSessions`.
+The settings browser shows the configured value; terminal status shows what the
+running daemon actually uses.
+
+## Server identity and configuration
+
+In `inherit` mode, a new instance uses the launching shell's `TMUX` socket, or the
+default socket under `TMUX_TMPDIR` (otherwise `/tmp`) when outside tmux. Native
+commands retain tmux's existing environment-based selection. Printed attachments
+and terminal metadata name a non-default server so a new terminal can reach it.
+The default mode does not require a server record or an additional instance
+identity check to start the daemon or use existing attachment paths.
+
+In `isolated` mode, the socket is named `openrig-<hash>` under tmux's per-user
+directory. The hash identifies the canonical `OPENRIG_HOME`: different homes get
+different servers, and symlink aliases of one home share a server. Choose a short
+`TMUX_TMPDIR` before the first isolated start if your normal temporary path would
+exceed the platform's Unix socket path limit.
+
+When isolation is requested or active, the daemon records the effective socket in
+`OPENRIG_HOME/tmux-server.json` using an atomic, owner-only write. Keep this file
+and the durable `tmux-server.initialized` marker with the instance state. A later shell
+with different `TMUX` or `TMUX_TMPDIR` does not redirect an established isolated
+instance. An invalid or unreadable isolation record is reported rather than overwritten.
+If the record disappears while isolation is selected, startup and `destroy --all` stop before
+touching sessions. Restore the record from an instance backup, or inspect the
+running daemon's `rig terminal status --json` to recover its effective selection.
+An ordinary inherited instance may record its socket for a later migration, but
+missing, invalid or unwritable records do not prevent its normal startup.
+The record's `pinned` flag distinguishes a fixed isolated or migration socket
+from an advisory inherited record. Returning to ordinary inheritance commits
+`pinned: false` and removes the active marker; an interrupted update must not
+leave startup using one server and offline cleanup selecting another. A valid
+unpinned record takes precedence over a leftover marker. Legacy inherited records
+and the old `1` marker do not make ordinary inheritance depend on that file.
+
+A newly started server loads tmux's normal system and user configuration,
+including `~/.tmux.conf`; OpenRig does not substitute an empty configuration.
+Existing OpenRig session options, such as `terminal.status_bar`, still apply.
+An existing server does not reread its configuration merely because the daemon
+restarts. Settings and sessions in a separate personal server are left alone.
+
+## Existing teams and deferred migration
+
+At startup, the daemon checks the previously selected server before switching.
+If registered sessions still exist there, it keeps that server and reports a
+pending migration. Explicit `OPENRIG_HOME` ownership markers distinguish another
+instance's namesakes; unmarked legacy sessions are retained conservatively. If
+the old server cannot be inspected, the daemon also keeps it and reports why.
+
+Stop the listed sessions deliberately, then restart the daemon again to apply the
+new selection. Merely editing the setting or waiting for the last seat to stop
+does not switch a running daemon. The same rule applies when returning to
+`inherit` from isolation. A normal inherited start continues to follow its launching
+environment, as before. No
+live pane is moved, restarted or killed by server selection. Old viewing windows
+can be closed separately after their conversations have stopped.
+
+Bring a stopped team back with `rig up <name> --existing`. Its normal restoration
+rules still apply: a seat without usable provider history requires an explicit
+fresh-start choice before it launches.
+
+On the first upgrade from a version without a socket record, the old-server
+candidate comes from the launching environment. If live seats use a non-default
+server, start the upgraded daemon from that original server to establish its
+record, or stop those seats before changing servers. Earlier versions did not
+persist enough information to find an arbitrary old socket from a different shell.
+
+## Attachments, discovery and cleanup
+
+`rig capture`, managed launch/stop/recovery, terminal previews and desktop viewers
+use the effective server. `rig ps --nodes --json` includes `tmuxServerArgs` and a
+matching `tmuxAttachCommand` for bound tmux seats. `rig tui --shared` uses that
+metadata. To attach a terminal directly, use the canonical session name:
+
+```sh
+rig terminal attach '<session>'
+rig terminal attach --read-only -- '<session>'
+```
+
+These commands attach on the daemon's machine. They do not create a seat or
+switch somebody else's tmux client. A local CLI pointed at an HTTP remote daemon
+cannot attach a same-named local session. For an effectively isolated server,
+terminal attachment, shared TUI and desktop windows also compare the daemon's
+instance identity with the local durable identity before using local terminal
+tools. A loopback URL through a port forward does not bypass this isolation check;
+use the matching `OPENRIG_HOME` on the daemon's machine. Inherited servers retain
+the existing local-host checks without requiring a matching durable identity.
+SSH view tiles follow the viewing instance's effective mode. In `inherit`, they
+use remote tmux through `ssh -t`, including `-r` for read-only views, and keep the
+bounded, identity-checked reattachment when a seat's session is replaced. The
+remote host needs only tmux and a POSIX shell, not OpenRig. Copyable recovery
+commands remain plain `ssh -t ... tmux attach`. In `isolated`, tiles run
+`rig terminal attach` on the remote host,
+using that host's local instance and socket; both the remote CLI and daemon must
+support this command. The remote shell must select the right `OPENRIG_HOME` when
+using a custom home. Local socket paths are never sent to it, and an isolated
+attachment failure does not fall back to a same-named default-server session.
+An inherited view does not discover an isolated remote instance's socket; use
+the remote instance's `rig terminal attach` directly for that mixed setup.
+
+Discovery and adoption see the selected server. Existing sessions on another
+server are not imported automatically. `rig attach --self` reports its current
+socket and, for an isolated instance, rejects a different server before binding
+or adding a node; an older
+CLI without socket metadata must be updated to self-attach to an isolated
+instance. A shell outside tmux can still attach as `external_cli`.
+
+`rig destroy --all` reads the recorded effective server before stopping the daemon
+or removing its state when isolation or a deferred migration pins it. Ordinary
+`inherit` cleanup uses the invoking environment's normal tmux selection, as before;
+an old advisory socket is not used to redirect it. Cleanup kills only registered
+session names on the selected server.
+
+Dedicated selection requires POSIX tmux. On Windows, psmux retains its existing
+server behavior; requesting `isolated` produces a status note and a pending choice.
+
+## Two instances on one machine
+
+Isolation selects only a tmux server. Give each daemon its own OpenRig home,
+database and port as well. For example, from a directory where you want to keep
+the two instances:
+
+```sh
+mkdir -p .openrig-a .openrig-b
+env OPENRIG_HOME="$PWD/.openrig-a" OPENRIG_DB="$PWD/.openrig-a/rig.db" \
+  OPENRIG_PORT=5311 OPENRIG_TMUX_SERVER=isolated rig daemon start
+env OPENRIG_HOME="$PWD/.openrig-b" OPENRIG_DB="$PWD/.openrig-b/rig.db" \
+  OPENRIG_PORT=5312 OPENRIG_TMUX_SERVER=isolated rig daemon start
+```
+
+Use the same instance-specific environment for every subsequent command, including
+`rig up`, `rig terminal status --json`, attachment and `rig daemon stop`. Do not
+reuse a database between these homes. Other inherited `OPENRIG_*` variables still
+apply; start from a shell without selectors for another running instance.
+
+The operating-system home, provider credentials and provider conversation history
+remain shared unless configured separately. This setting does not isolate the
+filesystem or network. Both inherited and isolated servers support the opt-in
+`OPENRIG_TMUX_CONTROL=1` execution path; its control client connects to the same
+server as the daemon's other tmux commands. On servers older than tmux 3.7c, or
+whose version cannot be recognized as a fixed release, OpenRig uses spawned
+clients instead: starting a control client while another session is created can
+crash those servers (#2304). This avoids creating OpenRig's own control client;
+it cannot protect the server from other tools' control clients.

docs/releases/README.md

@@ -1,38 +1,15 @@
 # Release Notes
 
-This directory is the lightweight release history for OpenRig.
+This directory holds one note per release for OpenRig v0.1.12 through v0.6.3, plus v0.6.12 (for example `v0.1.12.md`).
 
-It is intentionally simpler than a monolithic `CHANGELOG.md`.
-
-Each shipped release gets its own note:
-
-- `v0.1.12.md`
-- `v0.2.0.md`
-- `v0.3.0.md`
-- and so on
-
-## Why This Exists
-
-We want a practical release-management pattern that works with how OpenRig is actually being shipped today:
-
-- npm package release
-- optional Git tag
-- optional GitHub Release
-- short, human-written summary of what is included
-
-This keeps release notes:
-
-- easy to author
-- easy to link in GitHub Releases
-- easy to paste into announcements
-- durable in the repo
+Notes for later releases are in [`CHANGELOG.md`](../../CHANGELOG.md) and on the [GitHub Releases page](https://github.com/mvschwarz/openrig/releases).
 
 ## Minimal Process
 
 For each release:
 
-1. Copy `_template.md` to `vX.Y.Z.md`.
-2. Fill in the release summary, included changes, operator notes, known limitations, and verification performed.
+1. Add the release's section to [`CHANGELOG.md`](../../CHANGELOG.md).
+2. Write the release's GitHub Release notes. No new file is added to this directory.
 3. Run the **substance gate** at the exact release cut. The gate derives every
    file under the shipped pack sources, verifies a hash-bound human judgment
    and an individual disposition for each mechanical candidate, scans the
@@ -79,16 +56,16 @@ For each release:
    git push origin vX.Y.Z
    ```
 
-5. Create a GitHub Release using the same file:
+5. Create the GitHub Release with the notes from step 2:
 
    ```bash
    gh release create vX.Y.Z \
      --repo mvschwarz/openrig \
-     --title "OpenRig vX.Y.Z" \
-     --notes-file docs/releases/vX.Y.Z.md
+     --title "OpenRig X.Y.Z" \
+     --notes-file <release-notes.md>
    ```
 
-6. Publish the npm package if that is part of the release flow.
+6. Publish the npm package.
 
 ## Guidance
 
@@ -101,5 +78,3 @@ For each release:
 ## Scope
 
 This directory is a release-note archive, not a full historical changelog taxonomy.
-
-If OpenRig later wants a curated `CHANGELOG.md`, it can be generated or summarized from the release notes here.

package-lock.json

@@ -1136,7 +1136,6 @@
       "version": "1.32.2",
       "resolved": "https://registry.npmjs.org/@modelcontextprotocol/sdk/-/sdk-1.32.2.tgz",
       "integrity": "sha512-SWsVtH/0vULSQStoZ6xxH2K09Li8fXYTMC64ZeogCdpZ3XSi1eCPOByV1OpuJxuI3OcmHhosDWYyii07tpPsLw==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "@hono/node-server": "^1.19.9 || ^2.0.5",
@@ -2902,7 +2901,6 @@
       "version": "2.0.0",
       "resolved": "https://registry.npmjs.org/accepts/-/accepts-2.0.0.tgz",
       "integrity": "sha512-5cvg6CtKwfgdmVqY1WIiXKc3Q1bkRqGLi+2W/6ao+6Y7gu/RCwRuAhGEzh5B4KlszSuTLgZYuqFqo5bImjNKng==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "mime-types": "^3.0.0",
@@ -3125,7 +3123,6 @@
       "version": "2.4.0",
       "resolved": "https://registry.npmjs.org/body-parser/-/body-parser-2.4.0.tgz",
       "integrity": "sha512-Ifk1KcLrxaGt1occQagQNFNNOqRt1LFuehEnHHzB8Z6cRgIxRU1TBaiIsUiCdA+gy5PL7Y1IX31DNKkRmxtXjQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "bytes": "^3.1.2",
@@ -3150,7 +3147,6 @@
       "version": "2.1.0",
       "resolved": "https://registry.npmjs.org/content-type/-/content-type-2.1.0.tgz",
       "integrity": "sha512-mj7UPXE0jaqaOsukNZRUEfEi2AcL7C/vwmwcHV0O97eO1E1pxBZuyjlZrx5seTaNBg1U6+o35wpa35Qfcc+7ag==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=18"
@@ -3164,7 +3160,6 @@
       "version": "0.7.3",
       "resolved": "https://registry.npmjs.org/iconv-lite/-/iconv-lite-0.7.3.tgz",
       "integrity": "sha512-IKXpvIzjnC9XTAUbVBcMfGS0EPaIXtW6v+zr+RRp+hqULEpo0owZax6wyRwPOJbWbzjYspQwusTsfVr0ifh4uQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "safer-buffer": ">= 2.1.2 < 3.0.0"
@@ -3226,7 +3221,6 @@
       "version": "3.1.2",
       "resolved": "https://registry.npmjs.org/bytes/-/bytes-3.1.2.tgz",
       "integrity": "sha512-/Nf7TyzTx6S3yRJObOAV7956r8cr2+Oj8AC5dt8wSP3BQAoeX58NoHyCU8P8zGkNXStjTSi6fzO6F0pBdcYbEg==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.8"
@@ -3246,7 +3240,6 @@
       "version": "1.0.2",
       "resolved": "https://registry.npmjs.org/call-bind-apply-helpers/-/call-bind-apply-helpers-1.0.2.tgz",
       "integrity": "sha512-Sp1ablJ0ivDkSzjcaJdxEunN5/XvksFJ2sMBFfq6x0ryhQV/2b/KwFe21cMpmHtPOSij8K99/wSfoEuTObmuMQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "es-errors": "^1.3.0",
@@ -3260,7 +3253,6 @@
       "version": "1.0.4",
       "resolved": "https://registry.npmjs.org/call-bound/-/call-bound-1.0.4.tgz",
       "integrity": "sha512-+ys997U96po4Kx/ABpBCqhA9EuxJaQWDQg7295H4hBphv3IZg0boBKuwYpt4YXp6MZ5AmZQnU/tyMTlRpaSejg==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "call-bind-apply-helpers": "^1.0.2",
@@ -3405,7 +3397,6 @@
       "version": "2.0.1",
       "resolved": "https://registry.npmjs.org/content-disposition/-/content-disposition-2.0.1.tgz",
       "integrity": "sha512-e+H0ZXHSWYrENhQzw1LPuP4oF5MzVKmDU6d3hxlvaPEYLLg62MxtQNPRx4SYSuYJSBUgnQIG4HIN2tEtNv7Dog==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=18"
@@ -3419,7 +3410,6 @@
       "version": "1.0.5",
       "resolved": "https://registry.npmjs.org/content-type/-/content-type-1.0.5.tgz",
       "integrity": "sha512-nTjqfcBFEipKdXCv4YDQWCfmcLZKm81ldF0pAopTvyrFGVbcR6P/VAAd5G7N+0tTr8QqiU0tFadD6FK4NtJwOA==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.6"
@@ -3436,7 +3426,6 @@
       "version": "0.7.2",
       "resolved": "https://registry.npmjs.org/cookie/-/cookie-0.7.2.tgz",
       "integrity": "sha512-yki5XnKuf750l50uGTllt6kKILY4nQ1eNIQatoXEByZ5dWgnKqbnqmTrBE5B4N7lrMJKQ2ytWMiTO2o0v6Ew/w==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.6"
@@ -3452,7 +3441,6 @@
       "version": "1.2.2",
       "resolved": "https://registry.npmjs.org/cookie-signature/-/cookie-signature-1.2.2.tgz",
       "integrity": "sha512-D76uU73ulSXrD1UXF4KE2TMxVVwhsnCgfAyTg9k8P6KGZjlXKrOLe4dJQKI3Bxi5wjesZoFXJWElNWBjPZMbhg==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=6.6.0"
@@ -3462,7 +3450,6 @@
       "version": "2.8.6",
       "resolved": "https://registry.npmjs.org/cors/-/cors-2.8.6.tgz",
       "integrity": "sha512-tJtZBBHA6vjIAaF6EnIaq6laBBP9aq/Y3ouVJjEfoHbRBcHBAHYcMh/w8LDrk2PvIMMq8gmopa5D4V8RmbrxGw==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "object-assign": "^4",
@@ -3480,7 +3467,6 @@
       "version": "7.0.6",
       "resolved": "https://registry.npmjs.org/cross-spawn/-/cross-spawn-7.0.6.tgz",
       "integrity": "sha512-uV2QOWP2nWzsy2aMp8aRibhi9dlzF5Hgh5SHaB9OiTGEyDTiJJyx0uy51QXdyWbtAHNua4XJzUKca3OzKUd3vA==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "path-key": "^3.1.0",
@@ -3664,7 +3650,6 @@
       "version": "4.4.3",
       "resolved": "https://registry.npmjs.org/debug/-/debug-4.4.3.tgz",
       "integrity": "sha512-RGwwWnwQvkVfavKVt22FGLw+xYSdzARwm0ru6DhTVA3umU5hZc28V3kO4stgYryrTlLpuvgI9GiijltAjNbcqA==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "ms": "^2.1.3"
@@ -3699,7 +3684,6 @@
       "version": "2.0.0",
       "resolved": "https://registry.npmjs.org/depd/-/depd-2.0.0.tgz",
       "integrity": "sha512-g7nH6P6dyDioJogAAGprGpCtVImJhpPk/roCzdb3fIh61/s/nPsfR6onyMwkCAR/OlC3yBC0lESvUoQEAssIrw==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.8"
@@ -3745,7 +3729,6 @@
       "version": "1.0.1",
       "resolved": "https://registry.npmjs.org/dunder-proto/-/dunder-proto-1.0.1.tgz",
       "integrity": "sha512-KIN/nDJBQRcXw0MLVhZE9iQHmG68qAVIBg9CqmUYjmQIhgij9U5MFvrqkUL5FbtyyzZuOeOt0zdeRe4UY7ct+A==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "call-bind-apply-helpers": "^1.0.1",
@@ -3760,7 +3743,6 @@
       "version": "1.1.1",
       "resolved": "https://registry.npmjs.org/ee-first/-/ee-first-1.1.1.tgz",
       "integrity": "sha512-WMwm9LhRUo+WUaRN+vRuETqG89IgZphVSNkdFgeb6sS/E4OrDIN7t48CAewSHXc6C8lefD8KKfr5vY61brQlow==",
-      "dev": true,
       "license": "MIT"
     },
     "node_modules/electron-to-chromium": {
@@ -3773,7 +3755,6 @@
       "version": "2.0.0",
       "resolved": "https://registry.npmjs.org/encodeurl/-/encodeurl-2.0.0.tgz",
       "integrity": "sha512-Q0n9HRi4m6JuGIV1eFlmvJB7ZEVxu93IrMyiMsGC0lrMJMWzRgx6WGquyfQgZVb31vhGgXnfmPNNXmxnOkRBrg==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.8"
@@ -3796,7 +3777,6 @@
       "version": "1.0.1",
       "resolved": "https://registry.npmjs.org/es-define-property/-/es-define-property-1.0.1.tgz",
       "integrity": "sha512-e3nRfgfUZ4rNGL232gUgX06QNyyez04KdjFrF+LTRoOXmrOgFKDg4BCdsjW8EnT69eqdYGmRpJwiPVYNrCaW3g==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.4"
@@ -3806,7 +3786,6 @@
       "version": "1.3.0",
       "resolved": "https://registry.npmjs.org/es-errors/-/es-errors-1.3.0.tgz",
       "integrity": "sha512-Zf5H2Kxt2xjTvbJvP2ZWLEICxA6j+hAmMzIlypy4xcBg1vKVnx89Wy0GbS+kf5cwCVFFzdCFh2XSCFNULS6csw==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.4"
@@ -3823,7 +3802,6 @@
       "version": "1.1.2",
       "resolved": "https://registry.npmjs.org/es-object-atoms/-/es-object-atoms-1.1.2.tgz",
       "integrity": "sha512-HWcBoN6NileqtSydK2FqHbS/LoDd2pqrnQHLyJzBj4kOp/ky2MWMN694xOfkK8/SnUsW2DH7EfyVlydKCsm1Zw==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "es-errors": "^1.3.0"
@@ -3887,7 +3865,6 @@
       "version": "1.0.3",
       "resolved": "https://registry.npmjs.org/escape-html/-/escape-html-1.0.3.tgz",
       "integrity": "sha512-NiSupZ4OeuGwr68lGIeym/ksIZMJodUGOSCZ/FSnTxcrekbvqrgdUxlJOMpijaKZVjAJrWrGs/6Jy8OMuyj9ow==",
-      "dev": true,
       "license": "MIT"
     },
     "node_modules/estree-walker": {
@@ -3904,7 +3881,6 @@
       "version": "1.8.1",
       "resolved": "https://registry.npmjs.org/etag/-/etag-1.8.1.tgz",
       "integrity": "sha512-aIL5Fx7mawVa300al2BnEE4iNvo1qETxLrPI/o05L7z6go7fCw1J6EQmbK4FmJ2AS7kgVF/KEZWufBfdClMcPg==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.6"
@@ -3914,7 +3890,6 @@
       "version": "3.0.7",
       "resolved": "https://registry.npmjs.org/eventsource/-/eventsource-3.0.7.tgz",
       "integrity": "sha512-CRT1WTyuQoD771GW56XEZFQ/ZoSfWid1alKGDYMmkt2yl8UXrVR4pspqWNEcqKvVIzg6PAltWjxcSSPrboA4iA==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "eventsource-parser": "^3.0.1"
@@ -3927,7 +3902,6 @@
       "version": "3.1.1",
       "resolved": "https://registry.npmjs.org/eventsource-parser/-/eventsource-parser-3.1.1.tgz",
       "integrity": "sha512-EKN1vKAMcZ8MlYMpaNuxN6R9yakzH6uajHcHVTqWJzvu5pWw9DyhbP35HH8MVBQ+dZjAfDxk+A8NiR9KWaXiyQ==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=18.0.0"
@@ -3947,7 +3921,6 @@
       "version": "5.3.0",
       "resolved": "https://registry.npmjs.org/express/-/express-5.3.0.tgz",
       "integrity": "sha512-qCqy1xSoaugxOD4IVHb1ZWdkvCF2W3aE2okKouWj8VwJ+ztskSiXoBnCSJqfKT+cVPJhuJqe9YLbY425+Zh00g==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "accepts": "^2.0.0",
@@ -3991,7 +3964,6 @@
       "version": "8.7.1",
       "resolved": "https://registry.npmjs.org/express-rate-limit/-/express-rate-limit-8.7.1.tgz",
       "integrity": "sha512-2FXbqkygil3mZiXzEtM0IjE/UBmBNot0J1GEVrJT4JbrvwOSoEtqtO+wHbiMQELWe1eNvHoYRg2dkjOid+6JkA==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "debug": "^4.4.3",
@@ -4011,7 +3983,6 @@
       "version": "2.1.0",
       "resolved": "https://registry.npmjs.org/content-type/-/content-type-2.1.0.tgz",
       "integrity": "sha512-mj7UPXE0jaqaOsukNZRUEfEi2AcL7C/vwmwcHV0O97eO1E1pxBZuyjlZrx5seTaNBg1U6+o35wpa35Qfcc+7ag==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=18"
@@ -4113,7 +4084,6 @@
       "version": "2.1.1",
       "resolved": "https://registry.npmjs.org/finalhandler/-/finalhandler-2.1.1.tgz",
       "integrity": "sha512-S8KoZgRZN+a5rNwqTxlZZePjT/4cnm0ROV70LedRHZ0p8u9fRID0hJUZQpkKLzro8LfmC8sx23bY6tVNxv8pQA==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "debug": "^4.4.0",
@@ -4135,7 +4105,6 @@
       "version": "0.2.0",
       "resolved": "https://registry.npmjs.org/forwarded/-/forwarded-0.2.0.tgz",
       "integrity": "sha512-buRG0fpBtRHSTCOASe6hD258tEubFoRLb4ZNA6NxMVHNw2gOcwHo9wyablzMzOA5z9xA9L1KNjk/Nt6MT9aYow==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.6"
@@ -4158,7 +4127,6 @@
       "version": "2.0.0",
       "resolved": "https://registry.npmjs.org/fresh/-/fresh-2.0.0.tgz",
       "integrity": "sha512-Rx/WycZ60HOaqLKAi6cHRKKI7zxWbJ31MhntmtwMoaTeF7XFH9hhBp8vITaMidfljRQ6eYWCKkaTK+ykVJHP2A==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.8"
@@ -4213,7 +4181,6 @@
       "version": "1.3.0",
       "resolved": "https://registry.npmjs.org/get-intrinsic/-/get-intrinsic-1.3.0.tgz",
       "integrity": "sha512-9fSjSaos/fRIVIp+xSJlE6lfwhES7LNtKaCBIamHsjr2na1BiABJPo0mOjjz8GJDURarmCPGqaiVg5mfjb98CQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "call-bind-apply-helpers": "^1.0.2",
@@ -4247,7 +4214,6 @@
       "version": "1.0.1",
       "resolved": "https://registry.npmjs.org/get-proto/-/get-proto-1.0.1.tgz",
       "integrity": "sha512-sTSfBjoXBp89JvIKIefqw7U2CCebsc74kiY6awiGogKtoSGbgjYE/G/+l9sF3MWFPNc9IcoOC4ODfKHfxFmp0g==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "dunder-proto": "^1.0.1",
@@ -4286,7 +4252,6 @@
       "version": "1.2.0",
       "resolved": "https://registry.npmjs.org/gopd/-/gopd-1.2.0.tgz",
       "integrity": "sha512-ZUKRh6/kUFoAiTAtTYPZJ3hw9wNxx+BIBOijnlG9PnrJsCcSjs1wyyD6vJpaYtgnzDrKYRSqf3OO6Rfa93xsRg==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.4"
@@ -4308,7 +4273,6 @@
       "version": "1.1.0",
       "resolved": "https://registry.npmjs.org/has-symbols/-/has-symbols-1.1.0.tgz",
       "integrity": "sha512-1cDNdwJ2Jaohmb3sg4OmKaMBwuC48sYni5HUw2DvsC8LjGTLK9h+eb1X6RyuOHe4hT0ULCW68iomhjUoKUqlPQ==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.4"
@@ -4355,7 +4319,6 @@
       "version": "2.0.1",
       "resolved": "https://registry.npmjs.org/http-errors/-/http-errors-2.0.1.tgz",
       "integrity": "sha512-4FbRdAX+bSdmo4AUFuS0WNiPz8NgFt+r8ThgNWmlrjQjt1Q7ZR9+zTlce2859x4KSXrwIsaeTqDoKQmtP8pLmQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "depd": "~2.0.0",
@@ -4427,14 +4390,12 @@
       "version": "2.0.4",
       "resolved": "https://registry.npmjs.org/inherits/-/inherits-2.0.4.tgz",
       "integrity": "sha512-k/vGaX4/Yla3WzyMCvTQOXYeIHvqOKtnqBduzTHpzpQZzAskKMhZ2K+EnBiSM9zGSoIFeMpXKxa4dYeZIQqewQ==",
-      "dev": true,
       "license": "ISC"
     },
     "node_modules/ip-address": {
       "version": "10.7.5",
       "resolved": "https://registry.npmjs.org/ip-address/-/ip-address-10.7.5.tgz",
       "integrity": "sha512-t2v0rkILYv9WXmSL7TpAaU3HXgaFYOPUcH3QHgwQ++gpJuH3bqyMy4Fvt/DJCTKwzH3Cf+bAp5W/mFL39KPLzA==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 12"
@@ -4444,7 +4405,6 @@
       "version": "1.9.1",
       "resolved": "https://registry.npmjs.org/ipaddr.js/-/ipaddr.js-1.9.1.tgz",
       "integrity": "sha512-0KI/607xoxSToH7GjN1FfSbLoU0+btTicjsQSWQlh/hZykN8KpmMf7uYwPW3R+akZ6R/w18ZlXSHBYXiYUPO3g==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.10"
@@ -4518,7 +4478,6 @@
       "version": "4.0.0",
       "resolved": "https://registry.npmjs.org/is-promise/-/is-promise-4.0.0.tgz",
       "integrity": "sha512-hvpoI6korhJMnej285dSg6nu1+e6uxs7zG3BYAm5byqDsgJNWwxzM6z6iZiAgQR4TJ30JmBTOwqZUw3WlyH3AQ==",
-      "dev": true,
       "license": "MIT"
     },
     "node_modules/isbot": {
@@ -4534,7 +4493,6 @@
       "version": "2.0.0",
       "resolved": "https://registry.npmjs.org/isexe/-/isexe-2.0.0.tgz",
       "integrity": "sha512-RHxMLp9lnKHGHRng9QFhRCMbYAcVpn69smSGcq3f36xjgVVWThj4qqLbTLlq7Ssj8B+fIQ1EuCEGI2lKsyQeIw==",
-      "dev": true,
       "license": "ISC"
     },
     "node_modules/jiti": {
@@ -4550,7 +4508,6 @@
       "version": "6.2.2",
       "resolved": "https://registry.npmjs.org/jose/-/jose-6.2.2.tgz",
       "integrity": "sha512-d7kPDd34KO/YnzaDOlikGpOurfF0ByC2sEV4cANCtdqLlTfBlw2p14O/5d/zv40gJPbIQxfES3nSx1/oYNyuZQ==",
-      "dev": true,
       "license": "MIT",
       "funding": {
         "url": "https://github.com/sponsors/panva"
@@ -4626,7 +4583,6 @@
       "version": "8.0.2",
       "resolved": "https://registry.npmjs.org/json-schema-typed/-/json-schema-typed-8.0.2.tgz",
       "integrity": "sha512-fQhoXdcvc3V28x7C7BMs4P5+kNlgUURe2jmUT1T//oBRMDrqy1QPelJimwZGo7Hg9VPV3EQV5Bnq4hbFy2vetA==",
-      "dev": true,
       "license": "BSD-2-Clause"
     },
     "node_modules/json5": {
@@ -4723,7 +4679,6 @@
       "version": "1.1.0",
       "resolved": "https://registry.npmjs.org/math-intrinsics/-/math-intrinsics-1.1.0.tgz",
       "integrity": "sha512-/IXtbwEk5HTPyEwyKX6hGkYXxM9nbj64B+ilVJnC/R6B0pH5G4V3b0pVbL7DBj4tkhBAppbQUlf6F6Xl9LHu1g==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.4"
@@ -4733,7 +4688,6 @@
       "version": "1.1.1",
       "resolved": "https://registry.npmjs.org/media-typer/-/media-typer-1.1.1.tgz",
       "integrity": "sha512-yz3xRaG20c6/BOzvYoDaGtPmGscs7YivItZEEqe6GbwNfHuxu9YNmvnEkMzKldAGY4/80pRcQRZSEnhquk9XuQ==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.8"
@@ -4747,7 +4701,6 @@
       "version": "2.0.0",
       "resolved": "https://registry.npmjs.org/merge-descriptors/-/merge-descriptors-2.0.0.tgz",
       "integrity": "sha512-Snk314V5ayFLhp3fkUREub6WtjBfPdCPY1Ln8/8munuLuiYhsABgBVWsozAG+MWMbVEvcdcpbi9R7ww22l9Q3g==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=18"
@@ -4794,7 +4747,6 @@
       "version": "1.54.0",
       "resolved": "https://registry.npmjs.org/mime-db/-/mime-db-1.54.0.tgz",
       "integrity": "sha512-aU5EJuIN2WDemCcAp2vFBfp/m4EAhWJnUNSSw0ixs7/kXbd6Pg64EmwJkNdFhB8aWt1sH2CTXrLxo/iAGV3oPQ==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.6"
@@ -4804,7 +4756,6 @@
       "version": "3.0.2",
       "resolved": "https://registry.npmjs.org/mime-types/-/mime-types-3.0.2.tgz",
       "integrity": "sha512-Lbgzdk0h4juoQ9fCKXW4by0UJqj+nOOrI9MJ1sSj4nI8aI2eo1qmvQEie4VD1glsS250n15LsWsYtCugiStS5A==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "mime-db": "^1.54.0"
@@ -4852,7 +4803,6 @@
       "version": "2.1.3",
       "resolved": "https://registry.npmjs.org/ms/-/ms-2.1.3.tgz",
       "integrity": "sha512-6FlzubTLZG3J2a/NVCAleEhjzq5oxgHyaCU9yYXvcLsvoVaHJq/s5xXI6/XXP6tz7R9xAOtHnSO/tXtF3WRTlA==",
-      "dev": true,
       "license": "MIT"
     },
     "node_modules/mz": {
@@ -4888,7 +4838,6 @@
       "version": "1.1.0",
       "resolved": "https://registry.npmjs.org/negotiator/-/negotiator-1.1.0.tgz",
       "integrity": "sha512-NMPBRMJgiQHjbd8phG3Vebdx4kZ1H121rbl5IkMqeOsahptB9BKo/d7oJ3zTXqTgagn2bWlNSXkh0QUGM31RYg==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "content-type": "^2.1.0"
@@ -4905,7 +4854,6 @@
       "version": "2.1.0",
       "resolved": "https://registry.npmjs.org/content-type/-/content-type-2.1.0.tgz",
       "integrity": "sha512-mj7UPXE0jaqaOsukNZRUEfEi2AcL7C/vwmwcHV0O97eO1E1pxBZuyjlZrx5seTaNBg1U6+o35wpa35Qfcc+7ag==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=18"
@@ -4968,7 +4916,6 @@
       "version": "1.13.4",
       "resolved": "https://registry.npmjs.org/object-inspect/-/object-inspect-1.13.4.tgz",
       "integrity": "sha512-W67iLl4J2EXEGTbfeHCffrjDfitvLANg0UlX3wFUUSTx92KXRFegMHUVgSqE+wvhAbi4WqjGg9czysTV2Epbew==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.4"
@@ -4981,7 +4928,6 @@
       "version": "2.4.2",
       "resolved": "https://registry.npmjs.org/on-finished/-/on-finished-2.4.2.tgz",
       "integrity": "sha512-ATWUv387i+nLD49EiDoOft5scf32Vsdz3ktrQJG9bi7jRFmaXTssvyvj1EmeKVVgvesbqUtvQQvaS2TpTukAZQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "ee-first": "1.1.1"
@@ -4998,7 +4944,6 @@
       "version": "1.4.0",
       "resolved": "https://registry.npmjs.org/once/-/once-1.4.0.tgz",
       "integrity": "sha512-lNaJgI+2Q5URQBkccEKHTQOPaXdUxnZZElQTZY0MFUAuaEqe1E+Nyvgdz/aIyNi6Z9MzO5dv1H8n58/GELp3+w==",
-      "dev": true,
       "license": "ISC",
       "dependencies": {
         "wrappy": "1"
@@ -5021,7 +4966,6 @@
       "version": "1.3.3",
       "resolved": "https://registry.npmjs.org/parseurl/-/parseurl-1.3.3.tgz",
       "integrity": "sha512-CiyeOxFT/JZyN5m0z9PfXw4SCBJ6Sygz1Dpl0wqjlhDEGGBP1GnsUVEL0p63hoG1fcj3fHynXi9NYO4nWOL+qQ==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.8"
@@ -5031,7 +4975,6 @@
       "version": "3.1.1",
       "resolved": "https://registry.npmjs.org/path-key/-/path-key-3.1.1.tgz",
       "integrity": "sha512-ojmeN0qd+y0jszEtoY48r0Peq5dwMEkIlCOu6Q5f41lfkswXuKtYrhgoTpLnyIcHm24Uhqx+5Tqm2InSwLhE6Q==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=8"
@@ -5047,7 +4990,6 @@
       "version": "8.4.2",
       "resolved": "https://registry.npmjs.org/path-to-regexp/-/path-to-regexp-8.4.2.tgz",
       "integrity": "sha512-qRcuIdP69NPm4qbACK+aDogI5CBDMi1jKe0ry5rSQJz8JVLsC7jV8XpiJjGRLLol3N+R5ihGYcrPLTno6pAdBA==",
-      "dev": true,
       "license": "MIT",
       "funding": {
         "type": "opencollective",
@@ -5111,7 +5053,6 @@
       "version": "5.0.1",
       "resolved": "https://registry.npmjs.org/pkce-challenge/-/pkce-challenge-5.0.1.tgz",
       "integrity": "sha512-wQ0b/W4Fr01qtpHlqSqspcj3EhBvimsdh0KlHhH8HRZnMsEa0ea2fTULOXOS9ccQr3om+GcGRk4e+isrZWV8qQ==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=16.20.0"
@@ -5293,7 +5234,6 @@
       "version": "2.0.8",
       "resolved": "https://registry.npmjs.org/proxy-addr/-/proxy-addr-2.0.8.tgz",
       "integrity": "sha512-5nnx0yGyVUcY6t9RnWcARWtwT9F1D8O9rt08htPvnd49W1IgZtmLkhu9WfMzQj1cFxjHIO6connUNVW5k7AVyQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "forwarded": "0.2.0",
@@ -5321,7 +5261,6 @@
       "version": "6.16.1",
       "resolved": "https://registry.npmjs.org/qs/-/qs-6.16.1.tgz",
       "integrity": "sha512-5Px6J7rfHBAGUhNQ3es4YzcOdBBw0SDcVtThJml4RdnKw4tKs2epA0kPZ+K8X+WiHktT93RHVcQ/3ym5aTCgtg==",
-      "dev": true,
       "license": "BSD-3-Clause",
       "dependencies": {
         "es-define-property": "^1.0.1",
@@ -5358,7 +5297,6 @@
       "version": "1.3.0",
       "resolved": "https://registry.npmjs.org/range-parser/-/range-parser-1.3.0.tgz",
       "integrity": "sha512-hek2mFQpPuI4E1BBKrSto+BU3e3x4xuarsbiwr3+lf7p44juvFMV0XFWQAP3xUyqXA4RrXLIoaSUGbSt056ZMw==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.6"
@@ -5372,7 +5310,6 @@
       "version": "3.0.2",
       "resolved": "https://registry.npmjs.org/raw-body/-/raw-body-3.0.2.tgz",
       "integrity": "sha512-K5zQjDllxWkf7Z5xJdV0/B0WTNqx6vxG70zJE4N0kBs4LovmEYWJzQGxC9bS9RAKu3bgM40lrd5zoLJ12MQ5BA==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "bytes": "~3.1.2",
@@ -5388,7 +5325,6 @@
       "version": "0.7.3",
       "resolved": "https://registry.npmjs.org/iconv-lite/-/iconv-lite-0.7.3.tgz",
       "integrity": "sha512-IKXpvIzjnC9XTAUbVBcMfGS0EPaIXtW6v+zr+RRp+hqULEpo0owZax6wyRwPOJbWbzjYspQwusTsfVr0ifh4uQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "safer-buffer": ">= 2.1.2 < 3.0.0"
@@ -5654,7 +5590,6 @@
       "version": "2.2.0",
       "resolved": "https://registry.npmjs.org/router/-/router-2.2.0.tgz",
       "integrity": "sha512-nLTrUKm2UyiL7rlhapu/Zl45FwNgkZGaCpZbIHajDYgwlJCOzLSk+cIPAnsEqV955GjILJnKbdQC1nVPz+gAYQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "debug": "^4.4.0",
@@ -5701,7 +5636,6 @@
       "version": "2.1.2",
       "resolved": "https://registry.npmjs.org/safer-buffer/-/safer-buffer-2.1.2.tgz",
       "integrity": "sha512-YZo3K82SD7Riyi0E1EQPojLz7kpepnSQI9IyPbHHg1XXXevb5dJI7tpyN2ADxGcQbHG7vcyRHk0cbwqcQriUtg==",
-      "dev": true,
       "license": "MIT"
     },
     "node_modules/saxes": {
@@ -5737,7 +5671,6 @@
       "version": "1.2.1",
       "resolved": "https://registry.npmjs.org/send/-/send-1.2.1.tgz",
       "integrity": "sha512-1gnZf7DFcoIcajTjTwjwuDjzuz4PPcY2StKPlsGAQ1+YH20IRVrBaXSWmdjowTJ6u8Rc01PoYOGHXfP1mYcZNQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "debug": "^4.4.3",
@@ -5785,7 +5718,6 @@
       "version": "2.2.1",
       "resolved": "https://registry.npmjs.org/serve-static/-/serve-static-2.2.1.tgz",
       "integrity": "sha512-xRXBn0pPqQTVQiC8wyQrKs2MOlX24zQ0POGaj0kultvoOCstBQM5yvOhAVSUwOMjQtTvsPWoNCHfPGwaaQJhTw==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "encodeurl": "^2.0.0",
@@ -5805,14 +5737,12 @@
       "version": "1.2.0",
       "resolved": "https://registry.npmjs.org/setprototypeof/-/setprototypeof-1.2.0.tgz",
       "integrity": "sha512-E5LDX7Wrp85Kil5bhZv46j8jOeboKq5JMmYM3gVGdGH8xFpPWXUMsNrlODCrkoxMEeNi/XZIwuRvY4XNwYMJpw==",
-      "dev": true,
       "license": "ISC"
     },
     "node_modules/shebang-command": {
       "version": "2.0.0",
       "resolved": "https://registry.npmjs.org/shebang-command/-/shebang-command-2.0.0.tgz",
       "integrity": "sha512-kHxr2zZpYtdmrN1qDjrrX/Z1rR1kG8Dx+gkpK1G4eXmvXswmcE1hTWBWYUzlraYw1/yZp6YuDY77YtvbN0dmDA==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "shebang-regex": "^3.0.0"
@@ -5825,7 +5755,6 @@
       "version": "3.0.0",
       "resolved": "https://registry.npmjs.org/shebang-regex/-/shebang-regex-3.0.0.tgz",
       "integrity": "sha512-7++dFhtcx3353uBaq8DDR4NuxBetBzC7ZQOhmTQInHEd6bSrXdiEyzCvG07Z44UYdLShWUyXt5M/yhz8ekcb1A==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=8"
@@ -5835,7 +5764,6 @@
       "version": "1.1.1",
       "resolved": "https://registry.npmjs.org/side-channel/-/side-channel-1.1.1.tgz",
       "integrity": "sha512-6x6dK6zJdpTzF4sQeNYxwtvBzf6Eg4GtlesS94HOvTudUeyK2WXAaIfmDgsyslYrRBeFIlsi54AYsFGUuhmvrQ==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "es-errors": "^1.3.0",
@@ -5855,7 +5783,6 @@
       "version": "1.0.1",
       "resolved": "https://registry.npmjs.org/side-channel-list/-/side-channel-list-1.0.1.tgz",
       "integrity": "sha512-mjn/0bi/oUURjc5Xl7IaWi/OJJJumuoJFQJfDDyO46+hBWsfaVM65TBHq2eoZBhzl9EchxOijpkbRC8SVBQU0w==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "es-errors": "^1.3.0",
@@ -5872,7 +5799,6 @@
       "version": "1.0.1",
       "resolved": "https://registry.npmjs.org/side-channel-map/-/side-channel-map-1.0.1.tgz",
       "integrity": "sha512-VCjCNfgMsby3tTdo02nbjtM/ewra6jPHmpThenkTYh8pG9ucZ/1P8So4u4FGBek/BjpOVsDCMoLA/iuBKIFXRA==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "call-bound": "^1.0.2",
@@ -5891,7 +5817,6 @@
       "version": "1.0.2",
       "resolved": "https://registry.npmjs.org/side-channel-weakmap/-/side-channel-weakmap-1.0.2.tgz",
       "integrity": "sha512-WPS/HvHQTYnHisLo9McqBHOJk2FkHO/tlpvldyrnem4aeQp4hai3gythswg6p01oSoTl58rcpiFAjF2br2Ak2A==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "call-bound": "^1.0.2",
@@ -5946,7 +5871,6 @@
       "version": "2.0.2",
       "resolved": "https://registry.npmjs.org/statuses/-/statuses-2.0.2.tgz",
       "integrity": "sha512-DvEy55V3DB7uknRo+4iOGT5fP1slR8wQohVdknigZPMpMstaKJQWhwiYBACJE3Ul2pTnATihhBYnRhZQHGBiRw==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.8"
@@ -6246,7 +6170,6 @@
       "version": "1.0.1",
       "resolved": "https://registry.npmjs.org/toidentifier/-/toidentifier-1.0.1.tgz",
       "integrity": "sha512-o5sSPKEkg/DIQNmH43V0/uerLrpzVedkUh8tGNvaeXpfpuwjKenlSox/2O/BTlZUtEe+JG7s5YhEz608PlAHRA==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=0.6"
@@ -6314,7 +6237,6 @@
       "version": "2.1.0",
       "resolved": "https://registry.npmjs.org/type-is/-/type-is-2.1.0.tgz",
       "integrity": "sha512-faYHw0anBbc/kWF3zFTEnxSFOAGUX9GFbOBthvDdLsIlEoWOFOtS0zgCiQYwIskL9iGXZL3kAXD8OoZ4GmMATA==",
-      "dev": true,
       "license": "MIT",
       "dependencies": {
         "content-type": "^2.0.0",
@@ -6333,7 +6255,6 @@
       "version": "2.1.0",
       "resolved": "https://registry.npmjs.org/content-type/-/content-type-2.1.0.tgz",
       "integrity": "sha512-mj7UPXE0jaqaOsukNZRUEfEi2AcL7C/vwmwcHV0O97eO1E1pxBZuyjlZrx5seTaNBg1U6+o35wpa35Qfcc+7ag==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">=18"
@@ -6377,7 +6298,6 @@
       "version": "1.0.0",
       "resolved": "https://registry.npmjs.org/unpipe/-/unpipe-1.0.0.tgz",
       "integrity": "sha512-pjy2bYhSsufwWlKwPc+l3cN7+wuJlK6uz0YdJEOlQDbl6jo/YlPi4mb8agUkVC8BF7V8NuzeyPNqRksA3hztKQ==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.8"
@@ -6475,7 +6395,6 @@
       "version": "1.1.2",
       "resolved": "https://registry.npmjs.org/vary/-/vary-1.1.2.tgz",
       "integrity": "sha512-BNGbWLfd0eUPabhkXUVm0j8uuvREyTh5ovRa/dyow/BqAbZJyC+5fU+IzQOzmAKzYqYRAISoRhdQr3eIZ/PXqg==",
-      "dev": true,
       "license": "MIT",
       "engines": {
         "node": ">= 0.8"
@@ -7201,7 +7120,6 @@
       "version": "2.0.2",
       "resolved": "https://registry.npmjs.org/which/-/which-2.0.2.tgz",
       "integrity": "sha512-BLI3Tl1TW3Pvl70l3yq3Y64i+awpwXqsGBYWkkqMtnbXgrMD+yj7rhW0kuEDxzJaYXGjEW5ogapKNMEKNMjibA==",
-      "dev": true,
       "license": "ISC",
       "dependencies": {
         "isexe": "^2.0.0"
@@ -7234,7 +7152,6 @@
       "version": "1.0.2",
       "resolved": "https://registry.npmjs.org/wrappy/-/wrappy-1.0.2.tgz",
       "integrity": "sha512-l4Sp/DRseor9wL6EvV2+TuQn63dMkPjZ/sp9XkghTEbV9KlPS1xUsZ3u7/IQO4wxtcFB4bgpQPRcR3QCvezPcQ==",
-      "dev": true,
       "license": "ISC"
     },
     "node_modules/ws": {
@@ -7310,7 +7227,6 @@
       "version": "3.25.2",
       "resolved": "https://registry.npmjs.org/zod-to-json-schema/-/zod-to-json-schema-3.25.2.tgz",
       "integrity": "sha512-O/PgfnpT1xKSDeQYSCfRI5Gy3hPf91mKVDuYLUHZJMiDFptvP41MSnWofm8dnCm0256ZNfZIM7DSzuSMAFnjHA==",
-      "dev": true,
       "license": "ISC",
       "peerDependencies": {
         "zod": "^3.25.28 || ^4"
@@ -7353,6 +7269,7 @@
         "@hono/node-server": "^1.13.0",
         "@hono/node-ws": "^1.3.1",
         "@modelcontextprotocol/node": "^2.1.2",
+        "@modelcontextprotocol/sdk": "^1.32.2",
         "@modelcontextprotocol/server": "^2.4.0",
         "ajv": "^8.18.0",
         "ajv-formats": "^3.0.1",
@@ -7373,7 +7290,6 @@
       },
       "devDependencies": {
         "@modelcontextprotocol/client": "^2.4.0",
-        "@modelcontextprotocol/sdk": "^1.32.2",
         "@types/node": "^22.0.0",
         "typescript": "^5.7.0",
         "vitest": "^3.0.0"

packages/cli/package.json

@@ -48,6 +48,7 @@
     "@hono/node-server": "^1.13.0",
     "@hono/node-ws": "^1.3.1",
     "@modelcontextprotocol/node": "^2.1.2",
+    "@modelcontextprotocol/sdk": "^1.32.2",
     "@modelcontextprotocol/server": "^2.4.0",
     "ajv": "^8.18.0",
     "ajv-formats": "^3.0.1",
@@ -64,7 +65,6 @@
   },
   "devDependencies": {
     "@modelcontextprotocol/client": "^2.4.0",
-    "@modelcontextprotocol/sdk": "^1.32.2",
     "@types/node": "^22.0.0",
     "typescript": "^5.7.0",
     "vitest": "^3.0.0"

packages/cli/src/acp/connector.ts

@@ -179,7 +179,11 @@ export async function runAcpAssignment(queue: AcpQueue, options: AcpRunOptions):
       if (turnOver || cancelRequestedAt) return;
       cancelRequestedAt = new Date().toISOString();
       peer.notify("session/cancel", { sessionId });
-      await queue.update(row.qitemId, { note: `ACP run ${options.runId}: cancel requested at ${cancelRequestedAt}; upstream inference and spend are unknown unless the provider shows them` });
+      try {
+        await queue.update(row.qitemId, { note: `ACP run ${options.runId}: cancel requested at ${cancelRequestedAt}; upstream inference and spend are unknown unless the provider shows them` });
+      } catch (error) {
+        console.error(`[rig acp] could not record the cancel request: ${String(error)}`);
+      }
     });
     // A detach settles the turn here even when the agent keeps its stdout open after stdin closes.
     const detached = new Promise<never>((_, reject) => {

packages/cli/src/client.ts

@@ -3,7 +3,7 @@ import { assessLocalDaemonOwner, ForeignLocalDaemonError, type LocalDaemonOwner
 export { ForeignLocalDaemonError } from "./local-daemon-owner.js";
 import path from "node:path";
 import { ConfigStore } from "./config-store.js";
-import { getOpenRigHome, readOpenRigEnv } from "./openrig-compat.js";
+import { getOpenRigHome, readOpenRigEnv, readOpenRigUrl } from "./openrig-compat.js";
 import { fetchWithTimeout, FetchTimeoutError } from "./fetch-with-timeout.js";
 import { FETCH_HEADERS_TIMEOUT_MS, nodeHttpFetch } from "./long-request-fetch.js";
 
@@ -188,6 +188,29 @@ export class DaemonClient {
   private readonly sessionName?: string;
   private readonly principal?: DaemonClientOptions["principal"];
 
+  /** Isolated terminal tools require this home's durable identity, even on loopback. */
+  async isLocalInstance(): Promise<boolean> {
+    if (this.remoteTarget) return false;
+    const origin = (await import("./local-origin.js")).readLocalOrigin();
+    return !!origin && await this.matchesInstanceIdentity(origin);
+  }
+
+  private async matchesInstanceIdentity(origin: string): Promise<boolean> {
+    const controller = new AbortController();
+    let timer: ReturnType<typeof setTimeout> | undefined;
+    try {
+      const target = await Promise.race([
+        (async () => {
+          const res = await this.fetchImpl(`${this.baseUrl.replace(/\/+$/, "")}/healthz`, { signal: controller.signal });
+          return res.ok ? (await res.json() as { selfHostId?: unknown }).selfHostId : undefined;
+        })(),
+        new Promise<undefined>(resolve => { timer = setTimeout(() => resolve(undefined), Math.min(this.timeoutMs, 1000)); }),
+      ]);
+      return target === origin;
+    } catch { return false; }
+    finally { clearTimeout(timer); controller.abort(); }
+  }
+
   private async identityHeaders(): Promise<Record<string, string>> {
     const { headers, local } = await this.senderHeaders();
     // Only beside a seat this client was built for, and only to this host's own daemon: the endpoint itself must answer
@@ -230,7 +253,7 @@ export class DaemonClient {
     const localHeaders = senderIdentityHeaders(undefined, this.sessionName);
     const session = localHeaders[SENDER_IDENTITY_HEADER];
     if (!session || session.split("@").length >= 3) return { headers: localHeaders, local: false };
-    const directTarget = readOpenRigEnv("OPENRIG_URL", "RIGGED_URL");
+    const directTarget = readOpenRigUrl();
     if (!this.remoteTarget && !directTarget) return { headers: localHeaders, local: true };
     // Imported here, not at the top: it loads better-sqlite3 and yaml, which only a seat on a direct endpoint needs.
     const origin = this.originSelfHostId ?? (await import("./local-origin.js")).readLocalOrigin();
@@ -256,7 +279,7 @@ export class DaemonClient {
     if (baseUrl) {
       this.baseUrl = baseUrl;
     } else {
-      const envUrl = readOpenRigEnv("OPENRIG_URL", "RIGGED_URL");
+      const envUrl = readOpenRigUrl();
       if (envUrl) {
         this.baseUrl = envUrl;
       } else {

packages/cli/src/commands/attach.ts

@@ -11,26 +11,31 @@ function shellQuote(value: string): string {
   return `'${value.replace(/'/g, `'\\''`)}'`;
 }
 
-const defaultTmuxExec: TmuxExecFn = (cmd: string) => execSync(cmd, { encoding: "utf-8" }).trim();
+const defaultTmuxExec: TmuxExecFn = (cmd: string) => execSync(cmd, { encoding: "utf-8" }).replace(/\r?\n$/, "");
+
+// tmux writes a control character in its output as "_", so a newline between fields can't separate them, and a
+// session name may hold spaces or underscores. Ask for the two fixed-shape ids first; the rest of the line is the name.
+const TMUX_CONTEXT_FORMAT = "#{pane_id} #{window_id} #{session_name}";
+const TMUX_CONTEXT_LINE = /^(%\d+) (@\d+) (.+)$/;
 
 export function resolveAttachContext(tmuxExec: TmuxExecFn = defaultTmuxExec):
   | { attachmentType: "external_cli" }
-  | { attachmentType: "tmux"; tmuxSession: string; tmuxWindow?: string; tmuxPane: string } {
+  | { attachmentType: "tmux"; tmuxSession: string; tmuxWindow?: string; tmuxPane: string; tmuxSocketPath?: string } {
   const tmuxPane = process.env["TMUX_PANE"];
   if (!tmuxPane) {
     return { attachmentType: "external_cli" };
   }
 
   try {
-    const output = tmuxExec(`tmux display-message -p -t ${JSON.stringify(tmuxPane)} "#{session_name}\n#{window_id}\n#{pane_id}"`);
-    const [tmuxSession, tmuxWindow, resolvedPane] = output.split("\n").map((part) => part.trim());
-    if (tmuxSession && resolvedPane) {
-      return {
-        attachmentType: "tmux",
-        tmuxSession,
-        tmuxWindow: tmuxWindow || undefined,
-        tmuxPane: resolvedPane,
-      };
+    const output = tmuxExec(`tmux display-message -p -t ${shellQuote(tmuxPane)} ${shellQuote(TMUX_CONTEXT_FORMAT)}`);
+    const match = TMUX_CONTEXT_LINE.exec(output.trim());
+    if (match) {
+      const [, resolvedPane, tmuxWindow, tmuxSession] = match;
+      // Read the socket separately: tmux replaces control-character delimiters,
+      // and session names and socket paths may both contain literal spaces.
+      const tmuxSocketPath = tmuxExec(`tmux display-message -p -t ${shellQuote(resolvedPane!)} '#{socket_path}'`).replace(/\r?\n$/, "");
+      return { attachmentType: "tmux", tmuxSession: tmuxSession!, tmuxWindow, tmuxPane: resolvedPane!,
+        ...(tmuxSocketPath ? { tmuxSocketPath } : {}) };
     }
   } catch {
     // Fall back to external_cli if tmux metadata cannot be resolved.
@@ -119,6 +124,7 @@ export function attachCommand(depsOverride?: StatusDeps): Command {
         body["tmuxSession"] = attachContext.tmuxSession;
         if (attachContext.tmuxWindow) body["tmuxWindow"] = attachContext.tmuxWindow;
         body["tmuxPane"] = attachContext.tmuxPane;
+        if (attachContext.tmuxSocketPath) body["tmuxSocketPath"] = attachContext.tmuxSocketPath;
       }
       if (opts.displayName) body["displayName"] = opts.displayName;
 

packages/cli/src/commands/capture.ts

@@ -265,8 +265,10 @@ Supported notes:
 
       const client = deps.clientFactory(getDaemonUrl(status));
       const lines = parseInt(opts.lines ?? "20", 10);
+      // The capture default is 20: a missing or non-positive count reads as 20, never 0 or negative.
+      const requestedLines = Number.isNaN(lines) || lines < 1 ? 20 : lines;
 
-      const body: Record<string, unknown> = { lines: isNaN(lines) ? 20 : lines };
+      const body: Record<string, unknown> = { lines: requestedLines };
       if (opts.rig) body.rig = opts.rig;
       if (opts.pod) body.pod = opts.pod;
       if (session) body.session = session;
@@ -334,7 +336,13 @@ async function runCrossHostCapture(
   if (opts.rig) argv.push("--rig", opts.rig);
   if (opts.pod) argv.push("--pod", opts.pod);
   // Only an explicit --lines is forwarded: the remote's own default stays the default.
-  if (opts.linesGiven || opts.historyPlusPane) argv.push("--lines", opts.lines ?? "20");
+  // A value that parses as a positive count goes over as given (the remote decides);
+  // only a missing, zero, negative or non-numeric count sends the documented default.
+  if (opts.linesGiven || opts.historyPlusPane) {
+    const given = opts.lines ?? "20";
+    const parsed = parseInt(given, 10);
+    argv.push("--lines", Number.isNaN(parsed) || parsed < 1 ? "20" : given);
+  }
   if (opts.historyPlusPane) argv.push("--history-plus-pane");
   if (opts.readable) argv.push("--readable");
   if (opts.size !== undefined) argv.push("--size", opts.size);
@@ -375,7 +383,8 @@ async function runHttpHostCapture(
   hint?: string,
 ): Promise<void> {
   const lines = parseInt(opts.lines ?? "20", 10);
-  const body: Record<string, unknown> = { lines: isNaN(lines) ? 20 : lines };
+  const requestedLines = Number.isNaN(lines) || lines < 1 ? 20 : lines;
+  const body: Record<string, unknown> = { lines: requestedLines };
   if (opts.rig) body.rig = opts.rig;
   if (opts.pod) body.pod = opts.pod;
   if (session) body.session = session;

packages/cli/src/commands/chatroom.ts

@@ -4,7 +4,24 @@ import { readOpenRigEnv } from "../openrig-compat.js";
 import { getDaemonStatus, getDaemonUrl , daemonStatusGuard} from "../daemon-lifecycle.js";
 import { realDeps } from "./daemon.js";
 import type { StatusDeps } from "./status.js";
-import { execSync } from "node:child_process";
+import { execFileSync } from "node:child_process";
+
+/** tmux's arguments for the session that runs `rig chatroom watch`. The watcher goes to tmux as separate arguments, so
+ *  tmux runs it without a shell and the rig name arrives as one argument; `--` keeps a name that starts with `-` from
+ *  reading as an option. tmux reads the `-s` value as a format, where `##` is one `#` and a run of `#` before `[` is a
+ *  style it keeps as written, so every other `#` is doubled and the session is `chatroom@<rig>`. Some tmux versions
+ *  still change a session name (3.5a writes `_` for `:` and `.`, and escapes a backslash or a control character such
+ *  as a newline or tab), so `-P -F` prints the name tmux used. */
+export function chatroomWatchTmuxArgs(rig: string): string[] {
+  const session = `chatroom@${rig}`.replace(/#+(\[)?/g, (run, style?: string) => style ? run : run + run);
+  return ["new-session", "-d", "-P", "-F", "#{session_name}", "-s", tmuxArg(session),
+    "rig", "chatroom", "watch", "--", tmuxArg(rig)];
+}
+
+/** tmux ends a command at an argument that ends in `;`, unless a `\` comes before it, so a final `;` gets that `\`. */
+function tmuxArg(value: string): string {
+  return value.endsWith(";") ? `${value.slice(0, -1)}\\;` : value;
+}
 
 interface RigSummary {
   id: string;
@@ -175,9 +192,11 @@ export function chatroomCommand(depsOverride?: StatusDeps): Command {
       if (opts.tmux) {
         const sessionName = `chatroom@${rig}`;
         try {
-          execSync(`tmux new-session -d -s ${JSON.stringify(sessionName)} "rig chatroom watch ${JSON.stringify(rig)}"`, { stdio: "ignore" });
-          console.log(`Started watch in tmux session: ${sessionName}`);
-          console.log(`Attach with: tmux attach -t ${sessionName}`);
+          // The name tmux used, which differs from sessionName when tmux changed it (a `.` in a rig name, for example).
+          const created = execFileSync("tmux", chatroomWatchTmuxArgs(rig), { encoding: "utf8", stdio: ["ignore", "pipe", "ignore"] })
+            .trim() || sessionName;
+          console.log(`Started watch in tmux session: ${created}`);
+          console.log(`Attach with: tmux attach -t ${created}`);
         } catch {
           console.error(`Failed to create tmux session '${sessionName}'. It may already exist.`);
           process.exitCode = 1;

packages/cli/src/commands/compact.ts

@@ -52,10 +52,13 @@ from --state, asking it to skip reading if it did not compact; a retry would com
 An explicit retry still waits for the live pane to be idle; no automatic retry follows
 cancellation. --skip-map skips only the map requirement for this one
 manual request. Cancel an active preparation before retrying or using --skip-map.
-Automatic preparation waits at most 25 minutes from prep delivery, using existing
-polls. A stopped attempt does not restart automatically; rerun this command to
-start a fresh attempt. Cancellation disarms future input; it cannot undo a command
-already submitted. Raw provider /compact and normal messages are unchanged.`);
+Automatic preparation allows 25 minutes from prep delivery for the map. A completed
+map stays armed for idle through the next retry interval (10 minutes by default),
+without another preparation prompt; --state includes idleDeadlineAt. A completed
+map that never reaches idle reports map_complete_seat_never_idle; an unfinished
+map reports preparation_incomplete. Manual requests still share one 180-second budget.
+Stopped attempts may retry automatically after the cooldown; cancellation does not.
+Cancellation disarms future input; it cannot undo a command already submitted. Raw provider /compact and normal messages are unchanged.`);
 
   const getDepsF = (): CompactDeps => depsOverride ?? {
     lifecycleDeps: realDeps(),

packages/cli/src/commands/config.ts

@@ -11,7 +11,7 @@ function summarizeSettings(store: ConfigStore): Record<ValidKey, ResolvedSetting
 }
 
 // These settings remain boot-only; capture interval/line settings reload live.
-const BOOT_ONLY_KEYS = ["daemon.port", "daemon.host", "daemon.db_conflict", "db.path"];
+const BOOT_ONLY_KEYS = ["daemon.port", "daemon.host", "daemon.db_conflict", "db.path", "terminal.tmux_server"];
 const BOOT_ONLY_PREFIXES = ["transcripts.enabled", "transcripts.path"];
 
 function isBootOnlyKey(key: string): boolean {
@@ -96,8 +96,10 @@ Keys:
                          wake_retry_interval_seconds, wake_retry_cap, wake_unconfirmed_window_minutes,
                          wake_swap_grace_seconds (S01 wake-or-escalate ladder),
                          wake_human_rung (always | explicit-only — post agent-to-agent escalations to a person;
-                           under explicit-only, an ask that needs a person must be addressed to the person;
-                           without a human registry, explicit-only posts everything, as always does)
+                           explicit-only posts rows addressed to a person; other agent work stays visible,
+                           permission prompts still reach a person,
+                           choice prompts follow this setting; unavailable human delivery stays a visible row;
+                           an unreadable registry does not suppress eligibility; delivery still needs a ready route)
   retention.*            enabled, transitions_days, watchdog_days,
                          watchdog_keep_per_job, batch_size
   launch.non_interruptive  accept harness first-launch warnings at full bypass for new rigs (default off)
@@ -106,6 +108,7 @@ Keys:
                          Turn on for custom setups that intentionally rely on tmux globals. Running agents are unchanged.
                          If the scoped boundary is unavailable, launch continues with its previous environment and a diagnostic.
   terminal.status_bar    show the inner tmux status bar on launch (default off)
+  terminal.tmux_server   inherit the launching server, or isolated per OpenRig home (restart required)
   terminal.history_limit rows of tmux scrollback per pane in the ~/.tmux.conf block rig setup writes (2000-1000000; default 10000).
                          To keep more: rig config set terminal.history_limit 50000, then rig setup. A running tmux server keeps
                          its old limit until it restarts or you run tmux source-file ~/.tmux.conf; before tmux 3.7 that changes new panes only
@@ -193,6 +196,7 @@ Precedence: CLI flag > environment variable > config file > default`)
         // SWEEP-c: boot-only key + running daemon = stale-until-restart; say so loudly.
         if (isBootOnlyKey(key) && (await probeDaemonRunning())) {
           console.error(`note: '${key}' is read at daemon BOOT — the running daemon keeps its current value; this takes effect on the next daemon restart (rig daemon stop && rig daemon start).`);
+          if (key === "terminal.tmux_server") console.error("note: surviving sessions keep their previous server until deliberately stopped; rig terminal status shows the effective selection and any pending migration.");
         }
       } catch (err) {
         console.error((err as Error).message);

packages/cli/src/commands/crash-cart.ts

@@ -1,7 +1,7 @@
 import { Command } from "commander";
 import { copyFileSync, lstatSync, mkdtempSync, readFileSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
-import { getOpenRigHome } from "../openrig-compat.js";
+import { getOpenRigHome, readOpenRigUrl } from "../openrig-compat.js";
 import { join } from "node:path";
 import { ConfigStore } from "../config-store.js";
 import { formatDaemonHostForUrl } from "../client.js";
@@ -54,6 +54,14 @@ function existsOrThrow(path: string): boolean {
   catch (error) { if ((error as NodeJS.ErrnoException).code === "ENOENT") return false; throw error; }
 }
 
+/** The evidence probe's URL: OPENRIG_URL with one trailing slash trimmed, as the detector trims it, else daemon.json's
+ *  host and port. The shared reader leaves a query or fragment URL as written, so the trim stays here. */
+export function crashCartEvidenceUrl(openrigUrl: string | undefined, state: { host?: string; port: number } | undefined): string {
+  return openrigUrl
+    ? `${openrigUrl.replace(/\/$/, "")}/healthz`
+    : `http://${formatDaemonHostForUrl(state?.host ?? "127.0.0.1")}:${state?.port ?? 7433}/healthz`;
+}
+
 /**
  * The real verdict: LAZY-imports the narrow @openrig/daemon/crash-cart subpath at invocation (dep
  * rail 2 — rig startup + other verbs never load the daemon module) and composes the shipped detector
@@ -63,7 +71,7 @@ function existsOrThrow(path: string): boolean {
 async function realEmit(): Promise<CrashCartEmit> {
   const cc = await import("@openrig/daemon/crash-cart");
   const home = getOpenRigHome();
-  const openrigUrl = process.env.OPENRIG_URL;
+  const openrigUrl = readOpenRigUrl();
   const probeClassified = (url: string) =>
     cc.probeHealthz(url, { fetch: (u, init) => fetch(u, init as RequestInit), timeoutMs: 700 });
   const detectDeps = {
@@ -83,9 +91,7 @@ async function realEmit(): Promise<CrashCartEmit> {
           ? `alive (pid ${state.pid})`
           : `dead (pid ${state.pid})`
         : "no daemon.json";
-      const url = openrigUrl
-        ? `${openrigUrl.replace(/\/$/, "")}/healthz`
-        : `http://${formatDaemonHostForUrl(state?.host ?? "127.0.0.1")}:${state?.port ?? 7433}/healthz`;
+      const url = crashCartEvidenceUrl(openrigUrl, state);
       const probeResult = await probeClassified(url);
       return { pidState, probeResult, failedSignal: `healthz ${probeResult} at ${url}` };
     },

packages/cli/src/commands/daemon.ts

@@ -345,13 +345,19 @@ export function daemonCommand(depsOverride?: LifecycleDeps): Command {
           console.log("Daemon PID is absent (stale state)");
           break;
         case "unverified":
+          // T2: a live start reservation names a launch in progress (listener not up yet).
+          if (status.startPending) {
+            const since = status.startPending.since ? ` since ${status.startPending.since}` : "";
+            console.log(`Daemon starting: launch reserved by pid ${status.startPending.pid}${since}; not listening yet`);
+          }
           // 1ae863d2 — C3 semantics: we could NOT confirm up or down; never claim stopped.
           if (status.siblingHint) {
             console.log("Daemon state UNVERIFIED — the resolved OPENRIG_HOME has no daemon state, but a live daemon appears under a sibling home:");
             console.log(`  resolved: ${status.siblingHint.resolvedHome}`);
             console.log(`  sibling:  ${status.siblingHint.siblingHome}`);
             console.log("  Fix: point OPENRIG_HOME at the right home (or check your shell env) — this CLI is likely resolving the wrong home.");
-          } else {
+          } else if (!status.startPending) {
+            // A pending start's probe was refused, not inconclusive; its line above says what is known.
             console.log("Daemon state UNVERIFIED — the probe timed out or was inconclusive (this is NOT evidence the daemon is down).");
             console.log("  Re-check with: rig daemon status  ·  direct: curl the daemon /healthz");
           }

packages/cli/src/commands/destroy.ts

@@ -18,6 +18,7 @@ import {
   type DestroyScope,
 } from "../destroy-helpers.js";
 import { stopDaemon } from "../daemon-lifecycle.js";
+import { readRecordedTmuxServerArgs } from "../tmux-attachment.js";
 
 export interface DestroyCommandDeps {
   configStore: Pick<ConfigStore, "resolve">;
@@ -116,6 +117,7 @@ function realDestroyDeps(): DestroyDeps {
     removePath: (targetPath: string) => rmSync(targetPath, { recursive: true, force: true }),
     mkdirp: (targetPath: string) => mkdirSync(targetPath, { recursive: true }),
     listManagedTmuxSessions: listManagedTmuxSessionsFromDb,
+    resolveTmuxServerArgs: readRecordedTmuxServerArgs,
     killTmuxSession: killTmuxSessionWithCli,
     sleep: (ms: number) => new Promise((resolve) => setTimeout(resolve, ms)),
     now: () => new Date(),

packages/cli/src/commands/doctor.ts

@@ -18,6 +18,7 @@ import { parse as parseYaml } from "yaml";
 import { compareSpecToLive, topologyFromRigSpec, topologyFromLiveLogicalIds } from "@openrig/daemon/spec-conformance";
 import { classifyNodeVersion } from "../node-support.js";
 import { checkProviderReadiness } from "../provider-auth.js";
+import { getOpenRigHome, readOpenRigUrl } from "../openrig-compat.js";
 
 interface DoctorCheck {
   name: string;
@@ -40,7 +41,12 @@ export interface DoctorDeps {
   env?: NodeJS.ProcessEnv;
   mkdirp?: (path: string) => void;
   checkWritable?: (path: string) => void;
-  fetch?: (url: string) => Promise<{ ok: boolean; json?: () => Promise<unknown> }>;
+  fetch?: (url: string, init?: { signal?: AbortSignal }) => Promise<{ ok: boolean; json?: () => Promise<unknown> }>;
+  /** The caller's effective user id, compared with the one the port's daemon reports. Absent: this process's own;
+   *  null: this platform has none. */
+  callerUid?: number | null;
+  /** How long the port's /healthz body may take to arrive (default 2000 ms); after that its owner is unknown. */
+  ownerReportTimeoutMs?: number;
   /**
    * Build B — path to a rig spec to check against the RUNNING rig of the same name (`--spec`).
    *
@@ -57,6 +63,71 @@ export interface DoctorDeps {
 const DEFAULT_PORT = 7433;
 const CMUX_SOCKET_CONTROL_FIX = 'In cmux Settings > Automation, select "Automation mode" for Socket Control Mode, or set automation.socketControlMode to "automation" in ~/.config/cmux/cmux.json (older versions use ~/.config/cmux/settings.json; current cmux reads that legacy file only for keys absent from cmux.json). cmux reloads saved settings automatically; rerun `rig doctor` to re-probe without restarting the daemon.';
 
+/** The port's OpenRig daemon, judged by the OS user it reports. /healthz's pid and euid are the daemon's own report,
+ *  not authentication, and every line says so. Another user's daemon warns, since a share set up through config alone
+ *  looks the same and must keep working; it passes when OPENRIG_URL selects this endpoint on purpose. An owner that can't
+ *  be compared, or a body that doesn't arrive in time, warns too. Nothing is stopped or refused here. */
+async function listenerOwnerCheck(where: string, base: string, res: { json?: () => Promise<unknown> }, release: () => void,
+  deps: DoctorDeps): Promise<DoctorCheck> {
+  const env = deps.env ?? process.env;
+  const at = `Port ${where} (${base}, OpenRig home ${getOpenRigHome(env)})`;
+  const callerUid = deps.callerUid === undefined ? process.geteuid?.() : deps.callerUid ?? undefined;
+  const boundMs = deps.ownerReportTimeoutMs ?? 2000;
+  let reported: { pid?: unknown; euid?: unknown } = {};
+  let late = false;
+  let timer: ReturnType<typeof setTimeout> | undefined;
+  try {
+    const read = res.json ? res.json() : Promise.resolve(null);
+    // Releasing the request after the bound rejects the abandoned read; that's expected, not an error.
+    read.catch(() => {});
+    const body: unknown = await Promise.race([
+      read,
+      new Promise<null>(resolve => { timer = setTimeout(() => { late = true; resolve(null); }, boundMs); }),
+    ]);
+    if (body && typeof body === "object") reported = body as { pid?: unknown; euid?: unknown };
+  } catch { /* no readable body: the owner stays unknown */ }
+  finally { clearTimeout(timer); release(); }
+  const pid = Number.isSafeInteger(reported.pid) && (reported.pid as number) > 0 ? `, pid ${reported.pid as number}` : "";
+  const euid = Number.isSafeInteger(reported.euid) && (reported.euid as number) >= 0 ? reported.euid as number : undefined;
+  const ownDaemon = "To run your own, give this OpenRig home a free port and start its daemon there: rig config set daemon.port <free-port>, then rig daemon start.";
+  if (euid === undefined || callerUid === undefined) {
+    return {
+      name: "port", status: "warn",
+      message: `${at} is in use by an OpenRig daemon whose OS user is unknown.`,
+      reason: late
+        ? `Its /healthz report didn't finish within ${boundMs / 1000} s, so doctor can't tell whether it's yours or another user's on this machine.`
+        : euid === undefined
+          ? "It doesn't report an effective user id, so doctor can't tell whether it's yours or another user's on this machine."
+          : "This platform has no effective user id to compare, so doctor can't tell whether it's yours.",
+      fix: `Check that it's yours with rig daemon status. ${ownDaemon}`,
+    };
+  }
+  const reports = `the endpoint reports euid ${euid}${pid}`;
+  if (euid === callerUid) return { name: "port", status: "pass", message: `${at} is in use by your OpenRig daemon: ${reports}, the same as yours.` };
+  const users = `another OS user: ${reports}; your euid is ${callerUid}`;
+  if (sameEndpoint(readOpenRigUrl(env), base)) {
+    return { name: "port", status: "pass", message: `${at} is the endpoint OPENRIG_URL selects, served by an OpenRig daemon of ${users}.` };
+  }
+  return {
+    name: "port", status: "warn",
+    message: `${at} is in use by an OpenRig daemon of ${users}.`,
+    reason: `It isn't this OpenRig home's daemon, unless that user shares it with you: a command that reaches ${base} acts in that user's OpenRig. The pid and euid are the daemon's own report on /healthz, a self-report rather than authentication.`,
+    fix: `${ownDaemon} To use a shared daemon on purpose, set OPENRIG_URL to its address.`,
+  };
+}
+
+/** Whether OPENRIG_URL names this endpoint: the same scheme, host and port, with localhost read as 127.0.0.1. */
+function sameEndpoint(url: string | undefined, base: string): boolean {
+  if (!url) return false;
+  try {
+    const [a, b] = [new URL(url), new URL(base)];
+    const host = (u: URL) => u.hostname === "localhost" ? "127.0.0.1" : u.hostname;
+    return a.protocol === b.protocol && a.port === b.port && host(a) === host(b);
+  } catch {
+    return url === base;
+  }
+}
+
 function defaultCheckPort(port: number, host: string): Promise<boolean> {
   return new Promise((resolve) => {
     const socket = new net.Socket();
@@ -234,13 +305,13 @@ export function runDoctorChecks(deps: DoctorDeps, providerChecks: DoctorCheck[]
     if (available) {
       return { name: "port", status: "pass", message: `Port ${daemonHost}:${daemonPort} available.` };
     }
-    // Port in use — check if it's our daemon via healthz
+    // Port in use — check if it's an OpenRig daemon via healthz, and whose.
+    const owner = new AbortController();
     try {
-      const res = await fetchFn(`${daemonBase}/healthz`);
-      if (res.ok) {
-        return { name: "port", status: "pass", message: `Port ${daemonHost}:${daemonPort} in use by OpenRig daemon.` };
-      }
+      const res = await fetchFn(`${daemonBase}/healthz`, { signal: owner.signal });
+      if (res.ok) return await listenerOwnerCheck(`${daemonHost}:${daemonPort}`, daemonBase, res, () => owner.abort(), deps);
     } catch { /* not our daemon */ }
+    finally { owner.abort(); }
     return {
       name: "port",
       status: "fail",

packages/cli/src/commands/edge.ts

@@ -8,7 +8,7 @@ import type { StatusDeps } from "./status.js";
 
 interface EdgeEnd {
   logicalId: string;
-  nodeId: string;
+  nodeId: string | null;
 }
 
 interface EdgeSuccess {
@@ -39,7 +39,8 @@ function report(result: EdgeOpResult, opts: { json?: boolean }): void {
     switch (result.outcome) {
       case "would_add":
         console.log(`plan     add ${line} in rig ${result.rigId}`);
-        console.log(`         from node ${edge.from.nodeId}, to node ${edge.to.nodeId}`);
+        console.log(`         from node ${edge.from.nodeId}, to ${edge.to.nodeId === null
+          ? `${edge.to.logicalId} (resolved when used)` : `node ${edge.to.nodeId}`}`);
         console.log("         nothing written (--plan)");
         break;
       case "added":
@@ -124,10 +125,10 @@ export function edgeCommand(depsOverride?: StatusDeps): Command {
 
   cmd
     .command("add")
-    .description("Ensure one edge between two existing seats; an existing edge is unchanged")
+    .description("Ensure one typed edge; an existing edge is unchanged")
     .argument("<rig>", "Target rig name or ID")
-    .argument("[from]", "Source seat logical ID, such as orch.lead")
-    .argument("[to]", "Target seat logical ID, such as review.reviewer")
+    .argument("[from]", "Existing source seat logical ID, such as orch.lead")
+    .argument("[to]", "Target logical ID, or a same-daemon address for escalates_to")
     .option("--kind <kind>", "delegates_to, spawned_by, can_observe, collaborates_with or escalates_to")
     .option("--plan", "Validate the request and show the change without writing")
     .option("--json", "JSON output for agents")
@@ -135,8 +136,12 @@ export function edgeCommand(depsOverride?: StatusDeps): Command {
 Examples:
   rig edge add my-rig orch.lead review.reviewer --kind delegates_to
   rig edge add my-rig orch.lead review.reviewer --kind delegates_to --plan --json
+  rig edge add my-rig work.lead review-advisor@coord --kind escalates_to
 
-Endpoints are fully qualified logical IDs for both same-pod and cross-pod edges.
+Use existing, fully qualified logical IDs in <rig> for both same-pod and
+cross-pod edges. For escalates_to only, the target may instead be a canonical
+same-daemon seat address (pod-member@rig), resolved when used; that rig or seat
+need not exist yet. Cross-host and synthetic destinations are unsupported.
 Run it once per edge. A repeat reports the existing edge and its ID, and writes
 nothing. No seats or sessions are changed.`)
     .action(async (rig: string, from: string | undefined, to: string | undefined, opts: { kind?: string; plan?: boolean; json?: boolean }) => {

packages/cli/src/commands/env.ts

@@ -8,7 +8,7 @@ const LONG_RUNNING_TIMEOUT_MS = 45_000;
 
 export function envCommand(depsOverride?: StatusDeps): Command {
   const cmd = new Command("env")
-    .description("Inspect and control rig environment services for service-backed rigs and managed apps")
+    .description("Inspect and control rig services; for execution placement use rig environment")
     .addHelpText("after", `
 Examples:
   rig env status secrets-manager

packages/cli/src/commands/environment.ts

@@ -0,0 +1,170 @@
+import { parse as parseYaml, stringify as stringifyYaml } from "yaml";
+import { Command } from "commander";
+import { execFile, spawn } from "node:child_process";
+import { promisify } from "node:util";
+import { existsSync, lstatSync, mkdirSync, readFileSync, realpathSync, writeFileSync } from "node:fs";
+import { join, resolve } from "node:path";
+import { Client } from "@modelcontextprotocol/sdk/client/index.js";
+import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";
+import { DaemonClient } from "../client.js";
+import { getDaemonPath, getDaemonStatus, getDaemonUrl, daemonStatusGuard } from "../daemon-lifecycle.js";
+import { getOpenRigHome } from "../openrig-compat.js";
+import { addGrant, grantFor, mcpTokenFilePath, readGrants } from "../mcp-http.js";
+import { realDeps } from "./daemon.js";
+import type { StatusDeps } from "./status.js";
+
+export interface EnvironmentCommandDeps extends StatusDeps { home?: string; assets?: string }
+interface WorkConnection { version: 1; mcpUrl: string; seat: string; rig: string; grantName: string; artifactRoot: string; token: string }
+const managedId = (id: string): string => {
+  if (!/^[A-Za-z0-9][A-Za-z0-9._-]{0,159}$/.test(id)) throw new Error("A short binding identifier is required");
+  return id;
+};
+function preserveWrite(file: string, data: string | Buffer): void {
+  mkdirSync(resolve(file, ".."), { recursive: true, mode: 0o700 });
+  if (existsSync(file) || (() => { try { return lstatSync(file).isSymbolicLink(); } catch { return false; } })()) {
+    if (lstatSync(file).isSymbolicLink() || !lstatSync(file).isFile() || !readFileSync(file).equals(Buffer.from(data)))
+      throw new Error(`An existing file is preserved: ${file}`);
+    return;
+  }
+  writeFileSync(file, data, { flag: "wx", mode: 0o600 });
+}
+function connectionRef(home: string, bindingId: string): string { return join(home, "environment-secrets", managedId(bindingId), "work.json"); }
+function readConnection(file: string): WorkConnection {
+  const stat = lstatSync(file);
+  if (!stat.isFile() || stat.isSymbolicLink() || stat.size > 16384 || (stat.mode & 0o077)) throw new Error("The managed work connection must be a private regular file");
+  const value = JSON.parse(readFileSync(file, "utf8")) as WorkConnection;
+  if (value.version !== 1 || typeof value.token !== "string" || !value.token || typeof value.mcpUrl !== "string") throw new Error("The work connection is invalid");
+  return value;
+}
+async function prepareWork(home: string, input: { bindingId: string; seat: string; rig: string; mcpUrl: string; artifactRoot: string }): Promise<string> {
+  const url = new URL(input.mcpUrl);
+  if (!["http:", "https:"].includes(url.protocol) || url.username || url.password || url.search || url.hash || !url.pathname.endsWith("/mcp"))
+    throw new Error("The work endpoint must be an HTTP(S) MCP URL without embedded credentials");
+  if (!/^[^\s@]+@[^\s@]+$/.test(input.seat) || !input.rig) throw new Error("A canonical seat and its rig are required");
+  const file = connectionRef(home, input.bindingId), grantName = `environment-${input.bindingId}`;
+  if (existsSync(file)) {
+    const prior = readConnection(file), grants = readGrants(mcpTokenFilePath(home)) ?? [];
+    const grant = grantFor(grants, prior.token);
+    if (!grant || grant.name !== grantName || prior.seat !== input.seat || prior.rig !== input.rig || prior.mcpUrl !== url.toString()
+      || prior.artifactRoot !== resolve(input.artifactRoot) || grant.seat !== input.seat || grant.owner || grant.workspace
+      || grant.rigs.length !== 1 || grant.rigs[0] !== input.rig || grant.roots.length !== 0) throw new Error("The existing managed work connection changed or was revoked; preserve it and reconcile before setup");
+    return file;
+  }
+  const token = await addGrant(mcpTokenFilePath(home), { name: grantName, seat: input.seat, rigs: [input.rig], roots: [], workspace: false, owner: false });
+  const data: WorkConnection = { version: 1, mcpUrl: url.toString(), seat: input.seat, rig: input.rig, grantName, artifactRoot: resolve(input.artifactRoot), token };
+  preserveWrite(file, JSON.stringify(data) + "\n");
+  return file;
+}
+
+export function environmentCommand(override?: EnvironmentCommandDeps): Command {
+  const cmd = new Command("environment").description("Explicit OpenShell execution placement and verified artifact return; not rig env (services)");
+  const deps = (): EnvironmentCommandDeps => override ?? { lifecycleDeps: realDeps(), clientFactory: url => new DaemonClient(url) };
+  const client = async () => { const d = deps(), status = await getDaemonStatus(d.lifecycleDeps); if (!daemonStatusGuard(status)) throw new Error("A reachable daemon is required"); return d.clientFactory(getDaemonUrl(status)); };
+  const home = () => deps().home ?? getOpenRigHome();
+  const report = (value: unknown) => console.log(JSON.stringify(value, null, 2));
+  const action = (fn: (...args: any[]) => Promise<void>) => async (...args: any[]) => { try { await fn(...args); } catch (error) { console.error((error as Error).message); process.exitCode = 1; } };
+  const checked = <T extends Record<string, any>>(res: { status: number; data: T }): T => { if (res.status >= 400) throw new Error(res.data?.error ?? `Environment request failed (${res.status})`); return res.data; };
+
+  cmd.command("setup").description("Inspect the pinned vendor CLI and install the maintained example without overwriting another preset")
+    .requiredOption("--directory <path>", "New or matching example directory")
+    .option("--binding <id>", "Prepare this binding's normal scoped MCP work connection")
+    .option("--seat <address>").option("--rig <name>").option("--mcp-url <url>", "Existing authenticated MCP HTTP listener, as reached by the guest")
+    .option("--guest-base-url <origin>", "Guest-only activity listener origin (not the main daemon API); required with work authority")
+    .option("--artifact-root <path>", "Host returned-artifact root, bound to this environment")
+    .option("--json", "JSON receipt; never credential values")
+    .action(action(async (opts: { directory: string; binding?: string; seat?: string; rig?: string; mcpUrl?: string; guestBaseUrl?: string; artifactRoot?: string }) => {
+      const binary = await promisify(execFile)("openshell", ["--version"], { timeout: 10_000, maxBuffer: 4096 });
+      if (!/\b0\.1\.6\b/.test(binary.stdout)) throw new Error("This reference requires OpenShell 0.1.6; no compute was prepared");
+      const root = resolve(opts.directory), assets = deps().assets ?? join(getDaemonPath(), "assets", "openshell");
+      const names = ["guest.py", "Dockerfile", "policy.yaml", "rig.yaml", "agents/worker/agent.yaml", "agents/worker/role.md", "agents/coordinator/agent.yaml", "agents/coordinator/role.md", "README.md", "driver-v1.schema.json", "claude-code-provider.yaml"];
+      for (const name of names) {
+        let data = readFileSync(join(assets, name));
+        if (name === "policy.yaml" && opts.mcpUrl && opts.guestBaseUrl) {
+          const mcp = new URL(opts.mcpUrl), activity = new URL(opts.guestBaseUrl);
+          if (![mcp, activity].every(url => ["http:", "https:"].includes(url.protocol) && !url.username && !url.password && !url.search && !url.hash)
+            || activity.pathname !== "/") throw new Error("Explicit guest endpoints without embedded credentials are required");
+          const mainPort = Number(process.env.OPENRIG_PORT ?? "7433");
+          const ports = [mcp, activity].map(url => Number(url.port || (url.protocol === "https:" ? 443 : 80)));
+          if (ports.includes(mainPort) || ports[0] === ports[1]) throw new Error("Select separate MCP and guest-only activity ports, never the daemon main port");
+          const policy = parseYaml(data.toString("utf8"));
+          for (const [name, url] of [["openrig_mcp", mcp], ["openrig_activity", activity]] as const) {
+            policy.network_policies[name].endpoints[0].host = url.hostname;
+            policy.network_policies[name].endpoints[0].port = Number(url.port || (url.protocol === "https:" ? 443 : 80));
+          }
+          data = Buffer.from(stringifyYaml(policy));
+        }
+        preserveWrite(join(root, name), data);
+      }
+      preserveWrite(join(home(), "community-harnesses", "openshell-claude.yaml"), readFileSync(join(assets, "preset.yaml")));
+      let ref: string | undefined;
+      if (opts.binding || opts.seat || opts.rig || opts.mcpUrl) {
+        if (!opts.binding || !opts.seat || !opts.rig || !opts.mcpUrl || !opts.guestBaseUrl) throw new Error("Preparing work authority needs --binding, --seat, --rig, --mcp-url and --guest-base-url together");
+        ref = await prepareWork(home(), { bindingId: opts.binding, seat: opts.seat, rig: opts.rig, mcpUrl: opts.mcpUrl, artifactRoot: resolve(opts.artifactRoot ?? join(root, "artifacts")) });
+      }
+      report({ version: 1, driver: "openshell", vendorVersion: "0.1.6", directory: root,
+        ...(ref ? { workConnectionRef: ref } : {}), effect: "reference and preset installed; no environment or native process created" });
+    }));
+
+  for (const verb of ["create", "attach"] as const) {
+    const sub = cmd.command(verb).argument("<binding>").description(verb === "create" ? "Journal intent, then explicitly create one owned sandbox; retries reconcile the same operation" : "Verify and bind an exact existing resource as an attached consumer")
+      .requiredOption("--seat <address>").option("--node <id>", "Existing node ID; otherwise claimed by the exact named seat on its first launch")
+      .requiredOption("--gateway <ref>").requiredOption("--workspace <ref>").requiredOption("--name <name>")
+      .requiredOption("--image <digest>").requiredOption("--provider <ref>").requiredOption("--policy <path>")
+      .requiredOption("--guest-base-url <origin>", "Guest-reachable guest-only activity listener origin (not the main daemon API)")
+      .option("--operation <id>", "Stable effect identity (default verb plus binding ID)").option("--json");
+    if (verb === "attach") sub.requiredOption("--resource-id <id>", "Immutable live resource ID").option("--fingerprint <value>", "Expected host-key fingerprint");
+    sub.action(action(async (binding: string, opts: Record<string, string>) => {
+      managedId(binding); const ref = connectionRef(home(), binding), work = readConnection(ref);
+      if (work.seat !== opts.seat) throw new Error("The managed work connection belongs to another seat; no provider operation was requested");
+      const body = { version: 1, bindingId: binding, operationId: opts.operation ?? `${verb}-${binding}`, nodeId: opts.node, seat: opts.seat,
+        baseRuntime: "claude-code", guestRoot: "/sandbox/openrig", guestBaseUrl: opts.guestBaseUrl, coordinationRef: ref, artifactRoot: work.artifactRoot,
+        placement: { gateway: opts.gateway, workspace: opts.workspace, name: opts.name, image: opts.image, provider: opts.provider, policy: realpathSync(opts.policy!) },
+        ...(verb === "attach" ? { resourceId: opts.resourceId, fingerprint: opts.fingerprint } : {}) };
+      const res = await (await client()).post<Record<string, any>>(`/api/environments/${verb}`, body, { timeoutMs: 120_000 });
+      report({ ...checked(res), operationId: body.operationId, effect: res.status === 202 ? "unconfirmed; reconcile this same operation" : "binding-confirmed; native launch not requested" });
+    }));
+  }
+  cmd.command("inspect").description("Read stored binding receipts or inspect the exact live resource").argument("<binding>").option("--live", "Observe the exact provider resource; unavailable is not absence").option("--json")
+    .action(action(async (binding: string, opts: { live?: boolean }) => report(checked(await (await client()).get<Record<string, any>>(`/api/environments/${encodeURIComponent(binding)}${opts.live ? "?live=true" : ""}`)))));
+  cmd.command("connect").argument("<binding>").requiredOption("--generation <id>").description("Open only the bound execution's viewer; closing it does not prove native or compute stopped")
+    .action(action(async (id: string, opts: { generation: string }) => {
+      const c = await client(), b = checked(await c.get<Record<string, any>>(`/api/environments/${encodeURIComponent(id)}?live=true`));
+      checked(await c.post<Record<string, any>>(`/api/environments/${encodeURIComponent(id)}/validate`, { generation: opts.generation }));
+      const args = ["--gateway", b.placement.gateway, "--workspace", b.placement.workspace, "sandbox", "connect", b.placement.name];
+      await new Promise<void>((resolveExit, reject) => {
+        const child = spawn("openshell", args, { stdio: "inherit" });
+        child.once("error", reject); child.once("exit", code => code === 0 ? resolveExit() : reject(new Error(`The environment viewer exited ${code}; native and compute state require inspection`)));
+      });
+    }));
+  cmd.command("export").argument("<binding>").requiredOption("--request <path>", "Versioned artifact request JSON with operation/task/generation/path/hash/bytes/schema/context")
+    .option("--json").description("Transfer actual bytes through the existing authenticated MCP work route, never an activity token")
+    .action(action(async (binding: string, opts: { request: string }) => {
+      const work = readConnection(connectionRef(home(), binding)), request = JSON.parse(readFileSync(opts.request, "utf8"));
+      const transport = new StreamableHTTPClientTransport(new URL(work.mcpUrl), { requestInit: { headers: { Authorization: `Bearer ${work.token}` } } });
+      const mcp = new Client({ name: "openrig-environment-export", version: "1" });
+      try {
+        await mcp.connect(transport);
+        const result = await mcp.callTool({ name: "rig_environment_export", arguments: { ...request, bindingId: binding } });
+        if (result.isError) throw new Error((result.content as Array<{ text?: string }>).map(c => c.text ?? "").join("\n"));
+        report(result.structuredContent ?? result.content);
+      } finally { await mcp.close(); }
+    }));
+  cmd.command("checkpoint").argument("<binding>").requiredOption("--generation <id>", "Selected current execution generation")
+    .requiredOption("--operation <id>", "Stable private checkpoint identity").option("--json")
+    .description("Capture exact private native history and verified prior artifact references while the provider remains accessible")
+    .action(action(async (id: string, opts: { generation: string; operation: string }) => report(checked(await (await client()).post<Record<string, any>>(`/api/environments/${encodeURIComponent(id)}/checkpoint`, { operationId: opts.operation, generation: opts.generation }, { timeoutMs: 120_000 })))));
+  cmd.command("recreate").argument("<binding>").requiredOption("--request <path>", "JSON with operationId, predecessor generation, checkpointId, exact resumeType/resumeToken and same-host placement")
+    .option("--json").description("Explicitly create successor compute only after exact predecessor cleanup and retained state verification; native launch remains normal rig restore")
+    .action(action(async (id: string, opts: { request: string }) => report(checked(await (await client()).post<Record<string, any>>(`/api/environments/${encodeURIComponent(id)}/recreate`, JSON.parse(readFileSync(opts.request, "utf8")), { timeoutMs: 120_000 })))));
+  cmd.command("completion").argument("<binding>").requiredOption("--request <path>", "JSON selecting operationId, taskId, generation, exportOperationId and stop-native or stop-compute")
+    .option("--json").description("Select finite completion for one task, generation and required verified export; persistent defaults stay unchanged")
+    .action(action(async (id: string, opts: { request: string }) => report(checked(await (await client()).post<Record<string, any>>(`/api/environments/${encodeURIComponent(id)}/completion`, JSON.parse(readFileSync(opts.request, "utf8")))))));
+  cmd.command("effect").argument("<binding>").requiredOption("--effect <name>", "close-viewer, stop-native, stop-compute or delete-owned")
+    .option("--generation <id>", "Selected execution generation; omit only for delete-owned before any generation exists")
+    .option("--resource-id <id>", "Exact immutable compute ID; required for delete-owned without a generation").requiredOption("--operation <id>", "Stable effect identity; retries reconcile without another provider mutation")
+    .option("--engine-socket <path>", "Owned local rootless Podman socket, required to capture dependent identities before delete-owned").option("--json")
+    .description("Execute one selected effect and return its separate observed outcome; attached native/storage/compute are preserved")
+    .action(action(async (id: string, opts: { effect: string; generation?: string; resourceId?: string; operation: string; engineSocket?: string }) => report(checked(await (await client()).post<Record<string, any>>(`/api/environments/${encodeURIComponent(id)}/effect`, { effect: opts.effect, generation: opts.generation ?? null, ...(opts.resourceId ? { resourceId: opts.resourceId } : {}), operationId: opts.operation, engineSocket: opts.engineSocket }, { timeoutMs: 120_000 })))));
+  cmd.addHelpText("after", "\nStart with rig environment setup --help. Run from your managed seat, or set OPENRIG_SESSION_NAME to the acting member@rig in a plain shell for existing transport attribution (not authentication). --guest-base-url names the separate guest-only activity listener, reachable from the guest; never the main API. Create/attach are explicit effects; a 202 is unconfirmed. The first native start remains rig up with the installed openshell-claude preset. Checkpoint before stop, select effects explicitly, and recreate compute before exact normal rig restore. No unknown effect permits fresh-native fallback.\n");
+  return cmd;
+}

packages/cli/src/commands/export.ts

@@ -69,7 +69,10 @@ export function exportCommand(depsOverride?: ExportDeps): Command {
       } else {
         deps.writeFile(opts.output, res.data);
         console.log(`Exported to ${opts.output}`);
-        for (const line of String(res.data).split("\n")) if (line.startsWith("# Not carried: ")) console.log(line.slice(2));
+        for (const line of String(res.data).split("\n")) {
+          if (line.startsWith("# Not carried: ")) console.log(line.slice(2));
+          if (line.startsWith("# Member startup not carried for ")) console.warn(`Warning: ${line.slice(2)}`);
+        }
 
         // Check for referenced culture_file and startup files resolving from the output directory
         const checkExists = deps.exists ?? ((p) => fs.existsSync(p));
@@ -94,7 +97,7 @@ export function exportCommand(depsOverride?: ExportDeps): Command {
             const pods = parsed["pods"] as Array<{
               id?: string;
               startup?: ExportedStartupBlock;
-              members?: Array<{ id?: string; agent_ref?: string }>;
+              members?: Array<{ id?: string; agent_ref?: string; startup?: ExportedStartupBlock }>;
             }> | undefined;
             for (const pod of Array.isArray(pods) ? pods : []) {
               for (const member of Array.isArray(pod.members) ? pod.members : []) {
@@ -123,7 +126,7 @@ export function exportCommand(depsOverride?: ExportDeps): Command {
 
             const startupBlocks = [
               parsed["startup"] as ExportedStartupBlock | undefined,
-              ...(Array.isArray(pods) ? pods.map((pod) => pod.startup) : []),
+              ...(Array.isArray(pods) ? pods.flatMap((pod) => [pod.startup, ...(pod.members ?? []).map((member) => member.startup)]) : []),
             ];
             for (const startup of startupBlocks) {
               if (startup && Array.isArray(startup.files)) {

packages/cli/src/commands/host.ts

@@ -1,12 +1,19 @@
 // OPR.0.4.4.13 — the host-registry verbs (add / list / doctor — arch
-// R13-2a: no edit/remove/tunnel/bootstrap verbs; hand-editing hosts.yaml
-// remains the path for exotica; the bootstrap ships as script + runbook).
-// OPR.0.4.6.MH1 FR-1 extends the family with `select` — the persisted
-// host-selection pointer (kubectl current-context shape). The pointer
-// lives in the config twins as `host.selected`; this verb is a THIN
-// CLIENT of the daemon config write (one write path for CLI+UI
+// R13-2a originally: no edit/remove/tunnel/bootstrap verbs; hand-editing
+// hosts.yaml remains the path for exotica; the bootstrap ships as script +
+// runbook). OPR.0.4.6.MH1 FR-1 extends the family with `select` — the
+// persisted host-selection pointer (kubectl current-context shape). The
+// pointer lives in the config twins as `host.selected`; this verb is a
+// THIN CLIENT of the daemon config write (one write path for CLI+UI
 // convergence, arch ruling); reads resolve from the local ConfigStore
 // (env > file > default — zero new daemon lookups on read paths).
+// Issue #2128 (maintainer-approved boundary) extends the family again with
+// `edit` and `remove`: the SAME parity-pinned writer contract as add — id
+// and transport stay fixed on an edit, unknown fields fail loud, empty
+// string clears an optional field, and remove deletes only the registry
+// entry (no credential files, no learned identity, no selection change).
+// When the selected host is edited or removed, the selection outcome is
+// shown plainly — never a silent switch.
 //
 // Secret hygiene (FR-1): every render carries bearer POINTERS (env var /
 // file NAMES) only — no code path in this file resolves a bearer value
@@ -18,11 +25,12 @@ import { join } from "node:path";
 import { hostname as osHostname, userInfo } from "node:os";
 import { connect } from "node:net";
 import { getOpenRigHome } from "../openrig-compat.js";
-import { addHostEntry, defaultHostRegistryPath, loadHostRegistry, validateHostRegistry, hostDisplayTarget, resolveHost, resolveRemoteBearer, bearerAuthHeaders, type HostEntry, type HttpHostEntry } from "../host-registry.js";
+import { addHostEntry, updateHostEntry, removeHostEntry, defaultHostRegistryPath, loadHostRegistry, validateHostRegistry, hostDisplayTarget, resolveHost, resolveRemoteBearer, bearerAuthHeaders, type HostEntry, type HttpHostEntry } from "../host-registry.js";
 import { runCrossHostCommand, type CrossHostResult } from "../cross-host-executor.js";
 import { recordHostObservation, loadHostBindings, describeBindingConflict, defaultHostBindingsPath, type HostObservationOutcome } from "../host-bindings.js";
 import { DaemonClient } from "../client.js";
 import { readOwnHostName, readSelectedHost } from "../host-selection.js";
+import { parseFeedHostSubscriptionKey } from "../config-store.js";
 
 // ---------------------------------------------------------------------------
 // rig host pair — the approval wait is BOUNDED.
@@ -71,6 +79,46 @@ export interface DoctorDeps {
    *  sidecar (TOFU + loud-conflict). Optional so existing fixtures keep compiling AND unit runs never
    *  write the real sidecar — absent means learning is skipped, which is the fail-open contract. */
   learnHostBinding?: (alias: string, observedHostId: string, observedUrl: string) => HostObservationOutcome;
+  /** The registry entry a reply addressed `@<hostId>` resolves to, by the lookup `rig send` makes.
+   *  Optional like the learner: absent means doctor makes no claim about where replies go. */
+  replyRoute?: (hostId: string) => HostEntry | null;
+  /** The self-id learned for a registry alias, if any. */
+  learnedHostId?: (alias: string) => string | null;
+}
+
+/** Where `rig send` takes a target ending `@<hostId>`: registry id, then a declared hostId, then the
+ *  first hosts.yaml entry with that learned binding (cross-host-target.ts). Null when none matches. */
+export function replyRouteFromDisk(hostId: string): HostEntry | null {
+  const loaded = loadHostRegistry();
+  if (!loaded.ok) return null;
+  const resolved = resolveHost(loaded.registry, hostId, loadHostBindings().bindings);
+  return resolved.ok ? resolved.host : null;
+}
+
+const REPLY_ROUTE_ORDER = "rig send resolves it by registry id, then a declared hostId, then the first hosts.yaml entry that learned it";
+
+/** A reachable entry whose self-id another entry answers for. Unknown, not fail: doctor can't tell
+ *  from here whether that other entry still reaches the machine. */
+function replyRouteElsewhereRow(host: HostEntry, hostId: string, winner: HostEntry): CheckRow {
+  return {
+    step: "reply-route",
+    status: "unknown",
+    detail: `replies addressed @${hostId} go to '${winner.id}', not '${host.id}' (${REPLY_ROUTE_ORDER})`,
+    fix: `run rig host doctor ${winner.id}; if it no longer reaches this machine, delete its entry from ${defaultHostRegistryPath()} (and its line in ${defaultHostBindingsPath()}, if any)`,
+  };
+}
+
+/** An unreachable entry that replies still resolve to: the failure a stale entry causes, named on
+ *  the entry itself. Null when the entry has no self-id or another entry answers for it. */
+function replyRouteThroughDeadEntryRow(host: HostEntry, deps: DoctorDeps): CheckRow | null {
+  const hostId = host.hostId ?? deps.learnedHostId?.(host.id) ?? null;
+  if (!hostId || deps.replyRoute?.(hostId)?.id !== host.id) return null;
+  return {
+    step: "reply-route",
+    status: "fail",
+    detail: `replies addressed @${hostId} resolve to '${host.id}', so they fail while it is unreachable`,
+    fix: `if another entry reaches this machine, delete '${host.id}' from ${defaultHostRegistryPath()} (and its line in ${defaultHostBindingsPath()}, if any), then run rig host doctor on that entry; otherwise restore this entry's connectivity`,
+  };
 }
 
 function nestedBoolean(obj: unknown, path: readonly string[]): boolean | undefined {
@@ -114,6 +162,8 @@ function defaultDoctorDeps(): DoctorDeps {
         sock.on("error", () => { sock.destroy(); resolve("closed"); });
       }),
     learnHostBinding: (alias, observedHostId, observedUrl) => recordHostObservation({ alias, observedHostId, observedUrl }),
+    replyRoute: replyRouteFromDisk,
+    learnedHostId: (alias) => loadHostBindings().bindings[alias]?.hostId ?? null,
   };
 }
 
@@ -163,6 +213,8 @@ export async function doctorLegs(host: HostEntry, deps: DoctorDeps): Promise<Che
           ? "SSH auth/permission gate — check the key, agent, and remote authorized_keys"
           : "verify the ssh target/alias, tailnet connectivity, and that sshd is running",
       });
+      const routeRow = reach.failedStep === "ssh-unreachable" ? replyRouteThroughDeadEntryRow(host, deps) : null;
+      if (routeRow) rows.push(routeRow);
       return rows; // later legs are meaningless without the shell
     }
     rows.push({ step: "transport-reachability", status: "pass", detail: `ssh to ${host.target} ok` });
@@ -255,20 +307,28 @@ export async function doctorLegs(host: HostEntry, deps: DoctorDeps): Promise<Che
         const parsed = JSON.parse(health.body) as { selfHostId?: unknown };
         if (typeof parsed.selfHostId === "string" && parsed.selfHostId.length > 0) {
           const obs = deps.learnHostBinding(host.id, parsed.selfHostId, (host as HttpHostEntry).url);
-          rows.push(obs.outcome === "conflict"
-            ? {
-                step: "host-identity-binding",
-                status: "fail",
-                detail: describeBindingConflict(host.id, obs.binding),
-                fix: `if the re-key is legitimate, delete the '${host.id}' entry in ${defaultHostBindingsPath()} and re-run doctor to re-learn`,
-              }
-            : {
-                step: "host-identity-binding",
-                status: "pass",
-                detail: obs.outcome === "bound"
-                  ? `learned host identity '${obs.binding.hostId}' (first contact) — reply hints ending @${obs.binding.hostId} now resolve to '${host.id}'`
-                  : `host identity '${obs.binding.hostId}' confirmed`,
-              });
+          if (obs.outcome === "conflict") {
+            rows.push({
+              step: "host-identity-binding",
+              status: "fail",
+              detail: describeBindingConflict(host.id, obs.binding),
+              fix: `if the re-key is legitimate, delete the '${host.id}' entry in ${defaultHostBindingsPath()} and re-run doctor to re-learn`,
+            });
+          } else {
+            // Ask the route rig send takes rather than assume it: an entry listed earlier in
+            // hosts.yaml with the same identity keeps the replies, even when it no longer answers.
+            const id = obs.binding.hostId;
+            const winner = deps.replyRoute?.(id) ?? null;
+            const routedHere = winner?.id === host.id;
+            rows.push({
+              step: "host-identity-binding",
+              status: "pass",
+              detail: obs.outcome === "bound"
+                ? `learned host identity '${id}' (first contact)${routedHere ? ` — reply hints ending @${id} now resolve to '${host.id}'` : ""}`
+                : `host identity '${id}' confirmed`,
+            });
+            if (winner && !routedHere) rows.push(replyRouteElsewhereRow(host, id, winner));
+          }
         }
       } catch {
         // body not JSON — an older or non-daemon endpoint; learning is skipped, nothing breaks.
@@ -276,6 +336,8 @@ export async function doctorLegs(host: HostEntry, deps: DoctorDeps): Promise<Che
     }
   } catch (err) {
     rows.push({ step: "transport-reachability", status: "fail", detail: `daemon unreachable at ${(host as HttpHostEntry).url}: ${(err as Error).message}`, fix: "check tailnet connectivity and that the remote daemon is running (rig daemon start on the host)" });
+    const routeRow = replyRouteThroughDeadEntryRow(host, deps);
+    if (routeRow) rows.push(routeRow);
     return rows;
   }
   rows.push({
@@ -481,6 +543,21 @@ function authPointer(h: HostEntry): string {
   return h.bearer_env ? `env:${h.bearer_env}` : h.bearer_file ? `file:${h.bearer_file}` : "—";
 }
 
+function printHostNextSteps(host: HostEntry): void {
+  console.log(`Verify it: rig host doctor ${host.id}`);
+  if (host.transport !== "http") {
+    console.log("Remote FEED requires an HTTP host entry. Pair its daemon URL with: rig host pair <url>");
+    return;
+  }
+  const key = `feed.subscriptions.${host.id}.enabled`;
+  if (!parseFeedHostSubscriptionKey(key)) {
+    console.log(`For remote FEED, register this host with a different id: '${host.id}' is reserved by feed settings.`);
+    return;
+  }
+  console.log("Remote FEED is off by default. To include this host's attention items, run locally:");
+  console.log(`  rig config set ${key} true`);
+}
+
 export function hostCommand(doctorDepsOverride?: DoctorDeps): Command {
   const cmd = new Command("host").description("Manage the multi-host registry (~/.openrig/hosts.yaml)");
 
@@ -516,7 +593,7 @@ export function hostCommand(doctorDepsOverride?: DoctorDeps): Command {
         return;
       }
       console.log(`Added host '${res.entry.id}' (${res.entry.transport}) to ${res.path}`);
-      console.log(`Verify it: rig host doctor ${res.entry.id}`);
+      printHostNextSteps(res.entry);
     });
 
   cmd
@@ -617,6 +694,66 @@ export function hostCommand(doctorDepsOverride?: DoctorDeps): Command {
       console.log(`This host is now named: ${trimmed}`);
     });
 
+  cmd
+    .command("edit")
+    .description("Edit a host entry (id and transport stay fixed; an empty string clears an optional field)")
+    .argument("<id>", "A registered host id")
+    .option("--target <target>", "SSH target — ssh transport")
+    .option("--user <user>", "SSH user — ssh transport (empty string clears the value)")
+    .option("--url <url>", "Remote daemon base URL — http transport")
+    .option("--notes <text>", "Free-form operator note (empty string clears the value)")
+    .option("--json", "JSON output")
+    .action((id: string, opts: { target?: string; user?: string; url?: string; notes?: string; json?: boolean }) => {
+      const changes: Record<string, unknown> = {};
+      if (opts.target !== undefined) changes["target"] = opts.target;
+      if (opts.user !== undefined) changes["user"] = opts.user;
+      if (opts.url !== undefined) changes["url"] = opts.url;
+      if (opts.notes !== undefined) changes["notes"] = opts.notes;
+
+      const res = updateHostEntry(id.trim(), changes);
+      if (!res.ok) {
+        console.error(res.error);
+        process.exitCode = 1;
+        return;
+      }
+      const wasSelected = readSelectedHost() === res.entry.id;
+      if (opts.json) {
+        console.log(JSON.stringify({ ok: true, path: res.path, entry: res.entry, wasSelected }));
+        return;
+      }
+      console.log(`Updated host '${res.entry.id}' (${res.entry.transport}) at ${res.path}`);
+      for (const key of Object.keys(changes) as string[]) {
+        const rec = (e: HostEntry) => e as unknown as Record<string, unknown>;
+        const before = rec(res.previous)[key];
+        const after = rec(res.entry)[key];
+        if (before !== after) {
+          console.log(`  ${key}: ${before === undefined ? "(unset)" : String(before)} → ${after === undefined ? "(cleared)" : String(after)}`);
+        }
+      }
+      if (wasSelected) console.log(`note: '${res.entry.id}' is the selected host — it was edited in place and the selection still points at it.`);
+    });
+
+  cmd
+    .command("remove")
+    .description("Remove a host entry (deletes only the registry entry — credential files, the learned identity and the selection stay as they are)")
+    .argument("<id>", "A registered host id")
+    .option("--json", "JSON output")
+    .action((id: string, opts: { json?: boolean }) => {
+      const res = removeHostEntry(id.trim());
+      if (!res.ok) {
+        console.error(res.error);
+        process.exitCode = 1;
+        return;
+      }
+      const wasSelected = readSelectedHost() === res.entry.id;
+      if (opts.json) {
+        console.log(JSON.stringify({ ok: true, path: res.path, entry: res.entry, wasSelected }));
+        return;
+      }
+      console.log(`Removed host '${res.entry.id}' (${res.entry.transport}) from ${res.path}`);
+      if (wasSelected) console.log(`note: '${res.entry.id}' was the selected host — the selection still points at '${res.entry.id}' and was NOT changed. Pick another: rig host select <id>`);
+    });
+
   cmd
     .command("pair")
     .description("Pair with a remote host from one pasted address — one approval on the target, done (FR-6)")
@@ -786,7 +923,7 @@ export function hostCommand(doctorDepsOverride?: DoctorDeps): Command {
         return;
       }
       console.log(`Paired. Host '${id}' registered (${targetBase}, bearer_file ${tokenPath}).`);
-      console.log(`Verify it: rig host doctor ${id}`);
+      printHostNextSteps(added.entry);
     });
 
   cmd

packages/cli/src/commands/mcp.ts

@@ -55,11 +55,14 @@ stays the endpoint; a proxy in front forwards the metadata path unchanged.`)
       let daemonUrl: string;
       if (opts.port) {
         const daemonPort = parseInt(opts.port, 10);
-        if (isNaN(daemonPort)) {
+        if (Number.isNaN(daemonPort)) {
           console.error("Invalid port number");
           process.exitCode = 1;
           return;
         }
+        if (daemonPort < 1 || daemonPort > 65535) {
+          console.error(`Warning: daemon port ${daemonPort} is outside [1, 65535]; MCP will start, but daemon tool calls may fail. Check --port.`);
+        }
         daemonUrl = `http://127.0.0.1:${daemonPort}`;
       } else {
         const status = await getDaemonStatus(deps.lifecycleDeps);

packages/cli/src/commands/provider.ts

@@ -1,3 +1,4 @@
+import { accountHomePlanMatches } from "../lib/account-home-switch.js";
 import fs from "node:fs";
 import { Command } from "commander";
 import { DaemonClient, DaemonTimeoutError } from "../client.js";
@@ -22,8 +23,8 @@ export const USAGE_ACROSS_HOSTS_REFRESH_TIMEOUT_MS = 75_000;
 
 type Peer = { logicalId: string; sessionName: string | null; home?: "same" | "different" | "unknown"; effect: string; step?: string };
 type SeatView = { home?: string; activity?: string; peers: Peer[] };
-type SwitchReport = { home?: string; account: string; tokenFile?: string; conversation?: string | null; loginChanged: boolean; undo: string | null; peers: Peer[];
-  restart?: { recoveryCommand?: string; remainingStep?: string }; next?: string; warnings?: string[]; tokenApplied?: boolean };
+type SwitchReport = { credentialKind?: "api-key"; home?: string; account: string; tokenFile?: string; conversation?: string | null; loginChanged: boolean; undo: string | null; peers: Peer[];
+  restart?: { recoveryCommand?: string; remainingStep?: string }; next?: string; warnings?: string[]; tokenApplied?: boolean; effort?: string };
 
 // A switch waits as long as `rig seat restart` does: the seat's in-place restart runs inside it.
 const SWITCH_TIMEOUT_MS = 120_000;
@@ -38,23 +39,25 @@ function printSeat(seat: SeatView): void {
   }
 }
 
-type AccountOpts = { toAccount?: string; tokenFile?: string; storedLogin?: boolean };
+type AccountOpts = { toAccount?: string; tokenFile?: string; apiKeyFile?: string; storedLogin?: boolean };
 
 /** The account a switch names: a Codex login's saved name, a Claude token's label with its file, or a Claude seat's stored
  *  login. Null when neither --to-account nor --stored-login was given. */
-function targetAccount(opts: AccountOpts): { toAccount: string; tokenFile?: string | null } | null {
+function targetAccount(opts: AccountOpts): { toAccount: string; tokenFile?: string | null; apiKeyFile?: string } | null {
+  if (opts.apiKeyFile !== undefined && (opts.tokenFile !== undefined || opts.storedLogin)) return null;
   if (opts.storedLogin) return { toAccount: "stored-login", tokenFile: null };
   if (!opts.toAccount) return null;
-  return { toAccount: opts.toAccount, ...(opts.tokenFile !== undefined ? { tokenFile: opts.tokenFile } : {}) };
+  return { toAccount: opts.toAccount, ...(opts.tokenFile !== undefined ? { tokenFile: opts.tokenFile } : {}), ...(opts.apiKeyFile !== undefined ? { apiKeyFile: opts.apiKeyFile } : {}) };
 }
 
-function printClaude(claude: { label: string; tokenFile: string | null; checked: string[]; notChecked: string[]; instead?: string }): void {
-  console.log(`Claude token: ${claude.tokenFile ? `"${claude.label}", read from ${claude.tokenFile} when Claude starts` : "none (the config dir's stored login)"}`);
+function printClaude(claude: { credentialKind?: "api-key"; label: string; tokenFile: string | null; checked: string[]; notChecked: string[]; instead?: string }): void {
+  console.log(`Claude ${claude.credentialKind === "api-key" ? "API key" : "token"}: ${claude.tokenFile ? `"${claude.label}", read from ${claude.tokenFile} when Claude starts` : "none (the config dir's stored login)"}`);
   if (claude.instead) console.log(`instead of the stored login, the seat would use ${claude.instead}`);
   if (claude.checked.length) console.log(`checked for anything that outranks it: ${claude.checked.join("; ")}`);
   if (claude.notChecked.length) console.log(`not checked (the seat's /status shows what it uses): ${claude.notChecked.join("; ")}`);
 }
 
+/** Build provider views and account switches; an isolated-home switch first requires the daemon to acknowledge its plan. */
 export function providerCommand(depsOverride?: StatusDeps): Command {
   const cmd = new Command("provider").description(
     "Provider accounts and usage signals, and moving one Codex seat to another login saved in its Codex home",
@@ -510,6 +513,7 @@ export function providerCommand(depsOverride?: StatusDeps): Command {
     .requiredOption("--seat <s>", "The seat session")
     .option("--to-account <a>", "Codex: the login's saved name in the seat's Codex home (rig auth list --codex-home <home>). Claude: your label for the token")
     .option("--token-file <path>", "Claude: the absolute path of a 0600 file holding the seat's own subscription token; it's read only as Claude starts")
+    .option("--api-key-file <path>", "Claude: a file containing its API key; resume the same conversation and settings")
     .option("--stored-login", "Claude: move the seat back to its config dir's stored login")
     .option("--json", "Output as parseable JSON")
     .action(async (opts: AccountOpts & { seat: string; json?: boolean }) => {
@@ -525,7 +529,8 @@ export function providerCommand(depsOverride?: StatusDeps): Command {
         return;
       }
       const qs = new URLSearchParams({ seat: opts.seat, toAccount: target.toAccount,
-        ...(target.tokenFile === null ? { storedLogin: "1" } : target.tokenFile !== undefined ? { tokenFile: target.tokenFile } : {}) });
+        ...(target.tokenFile === null ? { storedLogin: "1" } : target.tokenFile !== undefined ? { tokenFile: target.tokenFile } : {}),
+        ...(target.apiKeyFile !== undefined ? { apiKeyFile: target.apiKeyFile } : {}) });
       const res = await client.get<{ safe: boolean; reasons?: string[]; message?: string; seat?: SeatView;
         claude?: Parameters<typeof printClaude>[0] }>(`/api/provider/precheck?${qs.toString()}`);
       if (handleHttpError(res, "precheck")) return;
@@ -541,12 +546,19 @@ export function providerCommand(depsOverride?: StatusDeps): Command {
 
   cmd
     .command("switch")
-    .description("Move one seat to another subscription login, keeping its conversation: a Codex seat to a login saved in its Codex home (Linux only), a Claude seat to its own token file")
+    .description("Move one seat to subscription or API credentials, keeping its conversation: a Codex seat to a login saved in its Codex home (Linux only), a Claude seat to its own token file")
     .requiredOption("--seat <s>", "The seat session")
     .option("--to-account <a>", "Codex: the login's saved name in the seat's Codex home (rig auth list --codex-home <home>). Claude: your label for the token")
     .option("--token-file <path>", "Claude: the absolute path of a 0600 file holding the seat's own subscription token; it's read only as Claude starts")
+    .option("--api-key-file <path>", "Claude: a file containing its API key; resume the same conversation and settings")
     .option("--stored-login", "Claude: move the seat back to its config dir's stored login")
     .option("--force-unsafe", "Kept for scripts; a seat's checks can't be overridden, since each means nothing safe to restart or no such login")
+    .option("--clear-account-home", "Return to member/rig/daemon/native login selection through a fresh successor")
+    .option("--account-home <path>", "Move to an isolated native account home with a fresh successor and handover; no shared login is changed")
+    .option("--reason <text>", "Required with --account-home or --clear-account-home")
+    .option("--model <id>", "With --account-home or --clear-account-home, the successor's model")
+    .option("--codex-config-profile <name>", "With --account-home or --clear-account-home, the successor's Codex profile")
+    .option("--dry-run", "With --account-home or --clear-account-home, show the plan without changing anything")
     .option("--json", "Output as parseable JSON")
     .addHelpText("after", `
 The daemon plans the seat's in-place restart first (rig seat restart --plan), so a refusal changes nothing. It then saves the
@@ -578,9 +590,28 @@ save it. That environment is Claude's, so the tools, hooks and MCP servers it st
 environment writes it into the transcript. A token-authenticated seat can make model requests but can't use claude.ai
 connectors or Remote Control; its local MCP servers still work. --stored-login moves it back to the config dir's stored login;
 a CLAUDE_CODE_OAUTH_TOKEN the seat still receives outranks that login, and the report says which token it will use. Claude's
-/status shows token auth with no account email, so the label is yours, not an observed account.`)
-    .action(async (opts: AccountOpts & { seat: string; forceUnsafe?: boolean; json?: boolean }) => {
-      const target = targetAccount(opts);
+/status shows token auth with no account email, so the label is yours, not an observed account.
+For Claude API billing use --to-account <label> --api-key-file <absolute path to a 0600 file>. The key reaches only the
+resumed child's environment, never command arguments. API credentials fail closed if their file is unavailable. Existing
+model, permissions, config home and conversation are retained; in-flight work is interrupted. Claude may require native
+API-key approval before the first turn. OpenRig does not enforce API budgets or choose when to fall back.
+Codex can select a saved API-key auth profile with --to-account just like a subscription profile. Its shared-home effects
+and native credential-store requirements are unchanged. No fresh account home or conversation is created.
+Do not combine --api-key-file with --account-home or --clear-account-home; select a home and switch credentials separately.
+--clear-account-home returns to the configured member/rig/daemon/native home through a fresh successor with a handover.
+Use it without --to-account, --stored-login, --account-home or --token-file. It removes an account-home token reference
+only after confirmation; it does not restore historical login bytes. Anthropic active/default profiles live outside
+CLAUDE_CONFIG_DIR and are not isolated by an account home; check them before relying on a stored login.`)
+    .action(async (opts: AccountOpts & { seat: string; forceUnsafe?: boolean; json?: boolean; accountHome?: string; clearAccountHome?: boolean; reason?: string; model?: string; codexConfigProfile?: string; dryRun?: boolean }) => {
+      if (opts.apiKeyFile !== undefined && (opts.accountHome !== undefined || opts.clearAccountHome)) {
+        console.error("Use --api-key-file for an in-place credential switch; select or clear an account home in a separate switch.");
+        process.exitCode = 1; return;
+      }
+      if (opts.clearAccountHome && (opts.accountHome !== undefined || opts.tokenFile !== undefined || opts.toAccount !== undefined || opts.storedLogin)) {
+        console.error("Use --clear-account-home alone, without --account-home, --token-file, --to-account or --stored-login.");
+        process.exitCode = 1; return;
+      }
+      const target = opts.clearAccountHome ? { toAccount: "default" } : targetAccount(opts);
       if (!target) {
         console.error("Name the account: --to-account <name> (with --token-file <path> for a Claude seat), or --stored-login for a Claude seat.");
         process.exitCode = 1;
@@ -591,6 +622,85 @@ a CLAUDE_CODE_OAUTH_TOKEN the seat still receives outranks that login, and the r
         process.exitCode = 1;
         return;
       }
+      if (opts.accountHome !== undefined || opts.clearAccountHome) {
+        if (!opts.reason?.trim() || opts.storedLogin) {
+          console.error(opts.clearAccountHome ? "Pass --reason with --clear-account-home."
+            : "With --account-home pass --reason and --to-account, without --stored-login. A Claude subscription may also name --token-file.");
+          process.exitCode = 1;
+          return;
+        }
+        let preview: { status: number; data: Record<string, unknown> };
+        try {
+          const status = await client.get<{ runtime?: string }>(`/api/seat/status/${encodeURIComponent(opts.seat)}`);
+          const runtime = status.data.runtime;
+          if (status.status >= 400 || (runtime !== "codex" && runtime !== "claude-code")) {
+            const refusal = { ok: false, code: "account_home_preflight_failed", message: "The seat runtime could not be read; no switch was requested." };
+            if (opts.json) console.log(JSON.stringify(refusal)); else console.error(refusal.message);
+            process.exitCode = 1; return;
+          }
+          preview = await client.post<Record<string, unknown>>(`/api/seat/set-runtime/${encodeURIComponent(opts.seat)}`, {
+            runtime, ...(opts.clearAccountHome ? { clearAccountHome: true } : { accountHome: opts.accountHome }), reason: opts.reason, dryRun: true,
+            ...(opts.model !== undefined ? { model: opts.model } : {}),
+            ...(opts.codexConfigProfile !== undefined ? { codexConfigProfile: opts.codexConfigProfile } : {}),
+            ...(opts.tokenFile !== undefined ? { tokenFile: opts.tokenFile, accountLabel: opts.toAccount } : {}),
+          }, { timeoutMs: 30_000 });
+        } catch {
+          const refusal = { ok: false, code: "account_home_preflight_failed", message: "The read-only account-home preflight failed; no switch was requested." };
+          if (opts.json) console.log(JSON.stringify(refusal)); else console.error(refusal.message);
+          process.exitCode = 1; return;
+        }
+        const token = opts.tokenFile ? { label: opts.toAccount!, tokenFile: opts.tokenFile } : undefined;
+        if (!accountHomePlanMatches(preview.data, opts.clearAccountHome ? null : opts.accountHome!, token)) {
+          const refusal = { ok: false, code: "account_home_switch_unavailable", message: preview.data["message"]
+            ?? "The daemon did not acknowledge the selected account home and token. No switch was requested." };
+          if (opts.json) console.log(JSON.stringify(refusal)); else console.error(refusal.message);
+          process.exitCode = 1; return;
+        }
+        if (opts.dryRun) {
+          if (opts.json) console.log(JSON.stringify(preview.data));
+          else {
+            console.log(`Account-home plan for ${opts.seat}: ${opts.accountHome ?? "configured defaults"}. Nothing changed.`);
+            const settings = (preview.data["settings"] ?? {}) as Record<string, { value: string | null; source: string }>;
+            console.log(`Settings on ${String(preview.data["to"])}:`);
+            for (const [name, setting] of Object.entries(settings)) console.log(`  ${name}: ${setting.value ?? "none"} (${setting.source})`);
+            const permission = preview.data["permission"] as { mode: string; source: string } | undefined;
+            if (permission) console.log(`  permissions: ${permission.mode} (${permission.source})`);
+            const dropped = (preview.data["dropped"] as Array<{ setting: string; value: string }> | undefined) ?? [];
+            if (dropped.length) console.log(`Cleared from ${String(preview.data["from"])}: ${dropped.map(choice => `${choice.setting} ${choice.value}`).join("; ")}.`);
+            console.log("Costs:");
+            for (const cost of (preview.data["costs"] as string[] | undefined) ?? []) console.log(`  - ${cost}`);
+          }
+          return;
+        }
+        let result;
+        try {
+          result = await client.post<Record<string, unknown>>("/api/provider/switch", {
+            seat: opts.seat, ...(opts.clearAccountHome ? { clearAccountHome: true } : { ...target, accountHome: opts.accountHome }), reason: opts.reason, dryRun: false,
+            ...(opts.model !== undefined ? { model: opts.model } : {}),
+            ...(opts.codexConfigProfile !== undefined ? { codexConfigProfile: opts.codexConfigProfile } : {}),
+          }, { timeoutMs: 300_000 });
+        } catch (error) {
+          if (!(error instanceof DaemonTimeoutError)) throw error;
+          result = { status: 202, data: { ok: false, code: "runtime_switch_outcome_unknown",
+            message: `The daemon may still be switching. Inspect rig seat status ${opts.seat}; do not retry the switch.` } };
+        }
+        if (opts.json) console.log(JSON.stringify(result.data));
+        else if (result.data["recoveryPending"] === true) {
+          console.error(`Recovery is still pending for ${opts.seat}; the account-home switch is not complete.`);
+          if (typeof result.data["message"] === "string") console.error(result.data["message"]);
+          if (typeof result.data["guidance"] === "string") console.error(result.data["guidance"]);
+          for (const warning of (result.data["warnings"] as string[] | undefined) ?? []) console.error(`Warning: ${warning}`);
+        }
+        else if (result.data["ok"] === true) console.log(`Account-home switch for ${opts.seat}: ${opts.accountHome ?? "configured defaults"}. A fresh conversation carries the handover; the account is not verified until a real tool turn.`);
+        else console.error(result.data["message"] ?? "The account-home switch did not complete.");
+        process.exitCode = result.status >= 400 || result.data["ok"] !== true || result.data["recoveryPending"] === true ? 1 : 0;
+        return;
+      }
+      if (opts.dryRun || opts.reason || opts.model || opts.codexConfigProfile) {
+        console.error("--reason, --model, --codex-config-profile and --dry-run require --account-home or --clear-account-home.");
+        process.exitCode = 1;
+        return;
+      }
       type SwitchResponse = { status: number; data: { outcome?: string; reasons?: string[]; message?: string; seat?: SwitchReport } };
       let res: SwitchResponse;
       const startedAt = new Date().toISOString();
@@ -622,9 +732,10 @@ a CLAUDE_CODE_OAUTH_TOKEN the seat still receives outranks that login, and the r
             console.log("undo restores saved credential bytes only; sign-in is unverified. Saved credentials may expire, be revoked, or be superseded by refresh elsewhere. Restart the affected seat and use the existing account verification; fresh sign-in may be required.");
           }
         } else if (seat) {
-          console.log(`Claude token: ${seat.tokenFile ? `"${seat.account}", read from ${seat.tokenFile}` : "the config dir's stored login"}${seat.loginChanged
+          console.log(`Claude ${seat.credentialKind === "api-key" ? "API key" : "token"}: ${seat.tokenFile ? `"${seat.account}", read from ${seat.tokenFile}` : "the config dir's stored login"}${seat.loginChanged
             ? seat.tokenApplied === false ? " (reference saved; the token didn't take effect)" : " (set)" : ""}`);
           for (const warning of seat.warnings ?? []) console.log(`warning: ${warning}`);
+          if (seat.effort) console.log(`effort: ${seat.effort}`);
           if (seat.conversation) console.log(`conversation: ${seat.conversation}`);
           if (seat.undo) console.log(`to switch back: ${seat.undo}`);
         }

packages/cli/src/commands/queue.ts

@@ -356,10 +356,18 @@ function printResult(json: boolean, body: unknown, status: number): void {
   if (status >= 400) process.exitCode = status >= 500 ? 2 : 1;
 }
 
+/** #2276: the daemon's checks of a new row's human-facing fields (missing summary or evidence for a person,
+ *  notification, questions, reply-to). The queue repository raises them before the row's first write, inside the
+ *  create's transaction. A remote host's refusal arrives as remote_queue_write_failed instead, so it isn't one of these. */
+const PRE_WRITE_CREATE_REFUSALS = new Set([
+  "human_route_fields_required", "invalid_human_notification", "invalid_human_questions", "reply_to_requires_update", "reply_to_not_found",
+]);
+
 /** #1642: why a create provably wrote nothing, for the refusals that come before any write: the
  *  selected daemon refused the forward before contacting another host (its registry was unreadable,
- *  the host unknown or SSH-declared), or refused a local create to a rig it doesn't have. Any other
- *  failure, including a timeout or a remote's error, may follow a commit, so it returns null. */
+ *  the host unknown or SSH-declared), refused a local create to a rig it doesn't have, or refused the
+ *  new row's fields while validating it. Any other failure, including a timeout or a remote's error,
+ *  may follow a commit, so it returns null. */
 function provenCreateRefusal(status: number, body: unknown): string | null {
   if (!body || typeof body !== "object") return null;
   const { error, failureClass, hostId } = body as { error?: unknown; failureClass?: unknown; hostId?: unknown };
@@ -370,6 +378,9 @@ function provenCreateRefusal(status: number, body: unknown): string | null {
   if (status === 400 && error === "unknown_destination_rig") {
     return "the daemon refused it because the destination's rig doesn't exist on that daemon.";
   }
+  if (status === 400 && typeof error === "string" && PRE_WRITE_CREATE_REFUSALS.has(error)) {
+    return `the daemon refused it before writing anything (${error}).`;
+  }
   return null;
 }
 
@@ -515,13 +526,10 @@ function notThisCreate(row: Record<string, unknown>, sent: SentCreate, storedSou
   return null;
 }
 
-/** A create whose answer didn't arrive, or arrived unreadable, may still have committed. Read the row back once by the ID it
- *  was sent with, from the daemon that holds it: this one for a local row, the destination's registered daemon for a
- *  cross-host one. Only a row that is this create counts as committed; anything else leaves the outcome unknown. It never
- *  sends the create again. */
-async function readBackCreate(deps: QueueDeps, client: DaemonClient, hostId: string | undefined, qitemId: string, sent: SentCreate): Promise<{ row: Record<string, unknown> | null; note: string }> {
-  const remote = remoteHostOf(hostId);
-  const where = remote ? ` on host '${remote}'` : "";
+/** Both reconciliation and --verify read the daemon that holds the row, with its own bearer. */
+async function queueReadTarget(deps: QueueDeps, client: DaemonClient, hostId: string | undefined) {
+  let remote = remoteHostOf(hostId);
+  if (remote && await daemonSelfHostId(client) === remote) remote = undefined;
   let reader = client;
   let options: { timeoutMs: number; headers?: Record<string, string> } = { timeoutMs: CREATE_READ_BACK_TIMEOUT_MS };
   if (remote) {
@@ -529,15 +537,45 @@ async function readBackCreate(deps: QueueDeps, client: DaemonClient, hostId: str
     const match = registry.ok ? resolveHost(registry.registry, remote, loadHostBindings().bindings) : undefined;
     if (!match?.ok || match.host.transport !== "http") {
       const why = match?.ok ? "its entry in this CLI's host registry is SSH, not a daemon URL" : "it isn't in this CLI's host registry";
-      return { row: null, note: `This CLI couldn't read the row back from host '${remote}' (${why}).` };
+      return { error: `from host '${remote}' (${why})` };
     }
     const bearer = resolveRemoteBearer(match.host);
-    if (!bearer.ok) return { row: null, note: `This CLI couldn't read the row back from host '${remote}' (${bearer.error}).` };
+    if (!bearer.ok) return { error: `from host '${remote}' (${bearer.error})` };
     reader = deps.clientFactory(match.host.url);
     if (bearer.token) options = { ...options, headers: { Authorization: `Bearer ${bearer.token}` } };
   }
+  return { reader, options };
+}
+
+/** A successful forward isn't a delivery receipt. Never accept a different remote row as this create's receipt. */
+function createReceiptReader(deps: QueueDeps, client: DaemonClient, hostId: string | undefined, qitemId: string, sent: SentCreate): Pick<DaemonClient, "get"> {
+  if (!remoteHostOf(hostId)) return client;
+  let target: ReturnType<typeof queueReadTarget> | undefined;
+  let source: ReturnType<typeof storedSourceOf> | undefined;
+  return {
+    get: async <T>(path: string) => {
+      const selected = await (target ??= queueReadTarget(deps, client, hostId));
+      if ("error" in selected) throw new Error(`receipt could not be read ${selected.error}`);
+      const response = await selected.reader.get<T>(path, selected.options);
+      const row = response.data as Record<string, unknown> | undefined;
+      if (response.status === 200 && (!row || row.qitemId !== qitemId
+        || notThisCreate(row, sent, await (source ??= storedSourceOf(client, sent))) !== null)) {
+        throw new Error("receipt row does not match the created qitem");
+      }
+      return response;
+    },
+  };
+}
+
+/** A create whose answer didn't arrive, or arrived unreadable, may still have committed. Read the row back once by the ID it
+ *  was sent with, from the daemon that holds it. Only this create's row counts; it never sends the create again. */
+async function readBackCreate(deps: QueueDeps, client: DaemonClient, hostId: string | undefined, qitemId: string, sent: SentCreate): Promise<{ row: Record<string, unknown> | null; note: string }> {
+  const remote = remoteHostOf(hostId);
+  const where = remote ? ` on host '${remote}'` : "";
+  const target = await queueReadTarget(deps, client, hostId);
+  if ("error" in target) return { row: null, note: `This CLI couldn't read the row back ${target.error}.` };
   try {
-    const res = await reader.get<Record<string, unknown>>(`/api/queue/${encodeURIComponent(qitemId)}`, options);
+    const res = await target.reader.get<Record<string, unknown>>(`/api/queue/${encodeURIComponent(qitemId)}`, target.options);
     if (res.status === 200 && res.data?.qitemId === qitemId) {
       const differs = notThisCreate(res.data, sent, await storedSourceOf(client, sent));
       if (differs === "unread") {
@@ -1079,7 +1117,7 @@ export function queueCommand(depsOverride?: QueueDeps): Command {
           if (opts.verify && status < 400) {
             const createdId = typeof created.qitemId === "string" ? created.qitemId : null;
             const delivery = createdId
-              ? await waitForDeliveryOutcome(client, createdId, deps.deliveryVerify, { destinationSession: hostResolved.destination, nudge: opts.nudge })
+              ? await waitForDeliveryOutcome(createReceiptReader(deps, client, hostResolved.hostId, qitemId, sent), createdId, deps.deliveryVerify, { destinationSession: hostResolved.destination, nudge: opts.nudge })
               : unreadableDeliveryResult(null, hostResolved.destination, {}, "create response did not include a qitem id; delivery cannot be correlated");
             await printWriteResult(client, opts.json ?? false, { ...created, qitemId: createdId, persisted: true, delivery }, status, receipt);
             return;

packages/cli/src/commands/scope.ts

@@ -1068,7 +1068,7 @@ function buildResolveNotesCommand(): Command {
 
 function buildAuditCommand(): Command {
   return new Command("audit")
-    .description("Read-only scope audit: flag scope findings and show the advisory dependency graph")
+    .description("Read-only scope audit: flag scope structure findings and show the advisory dependency graph. It doesn't read proof readiness: use rig proof show")
     .requiredOption("--mission <name>", "Mission to audit")
     .option("--json", "Machine-readable JSON output")
     .action(async (opts, command) => {
@@ -1076,7 +1076,7 @@ function buildAuditCommand(): Command {
       const json = Boolean(opts.json);
       try {
         const missionsRoot = resolveMissionsRoot({ override: getOpts(command).workspace });
-        const { classifyScopeItem } = await import("../lib/scope/scope-audit.js");
+        const { classifyScopeItem, SLICE_DIRECTORY_NAME } = await import("../lib/scope/scope-audit.js");
         const missionName = opts.mission as string;
 
         const missionDir = path.join(missionsRoot, missionName);
@@ -1215,7 +1215,7 @@ function buildAuditCommand(): Command {
                 : null,
             });
 
-            if (!/^\d{2,}-/.test(entry)) {
+            if (!SLICE_DIRECTORY_NAME.test(entry)) {
               sliceResult.findings.push({
                 kind: "id_convention_violation",
                 severity: "high",

packages/cli/src/commands/seat.ts

@@ -1,3 +1,4 @@
+import { accountHomePlanMatches } from "../lib/account-home-switch.js";
 import { readFileSync } from "node:fs";
 import { Command } from "commander";
 import { DaemonClient, DaemonTimeoutError, terminalAuthHeaders } from "../client.js";
@@ -62,13 +63,15 @@ interface SeatStatusResponse {
   };
   session_status: string | null;
   launch_cwd?: string | null;
+  permissions_to?: string | null;
+  runner?: boolean;
   runtime_home?: {
     variable: string;
     home: string | null;
-    source: "member" | "rig" | "daemon" | "native_default";
+    source: "member" | "rig" | "daemon" | "native_default" | "account-switch";
     ignored?: { source: string; value: string };
   };
-  claude_token?: { label: string; token_file: string; file: "readable" | "unusable"; problem?: string; notice?: string };
+  claude_token?: { kind?: "api-key"; label: string; token_file: string; file: "readable" | "unusable"; problem?: string; notice?: string };
   session_reason?: string | null;
   startup_status: string | null;
   occupant_lifecycle: string;
@@ -193,22 +196,28 @@ const LEVEL_WORDS: Record<string, string> = {
   "sends-only": "Deliberate sends from people and agents type; OpenRig's own notices are held.",
 };
 
+/** Show configured seat state, including selected account references, without claiming the running login is verified. */
 function printHuman(status: SeatStatusResponse): void {
   console.log(`Seat ${status.seat_ref}`);
   console.log(`Rig: ${status.rig_name}`);
   console.log(`Logical ID: ${status.logical_id}`);
   console.log(`Current occupant: ${display(status.current_occupant)}`);
   if (status.launch_cwd !== undefined) console.log(`Working directory for future launches: ${display(status.launch_cwd)}`);
+  // Only a runner submits permission requests; a plain seat's own prompts are its harness's.
+  if (status.permissions_to || status.runner) {
+    console.log(`Runner permission requests answered by: ${status.permissions_to === "deny" ? "no one: refused (--deny)" : status.permissions_to ?? "the default person"}`);
+  }
   if (status.runtime_home) {
     const h = status.runtime_home;
     const from = h.source === "member" || h.source === "rig" ? `the seat's own, from its ${h.source} env`
+      : h.source === "account-switch" ? "the seat's selected account home"
       : h.source === "daemon" ? "the daemon's" : "the native default";
     console.log(`Runtime home (${h.variable}) for launches: ${display(h.home)} (${from}; configured, not proof of the login a running session uses)`);
     if (h.ignored) console.log(`  Ignored: its ${h.ignored.source} env sets ${h.variable}=${h.ignored.value}, which isn't an absolute path`);
   }
   if (status.claude_token) {
     const t = status.claude_token;
-    console.log(`Claude subscription token: ${t.label}, read from ${t.token_file} at each launch (${t.file === "readable" ? "readable now" : "won't apply at the next launch"})`);
+    console.log(`Claude ${t.kind === "api-key" ? "API key" : "subscription token"}: ${t.label}, read from ${t.token_file} at each launch (${t.file === "readable" ? "readable now" : "won't apply at the next launch"})`);
     if (t.notice) console.log(`  ${t.notice}`);
   }
   if (status.typingGuard) {
@@ -378,6 +387,7 @@ function printSeatError(error: SeatStatusError, fallback: string): void {
   }
 }
 
+/** Build seat lifecycle commands, requiring acknowledged plans before isolated account-home changes. */
 export function seatCommand(depsOverride?: SeatDeps & { readStdin?: () => Promise<string>; host?: RunOnHostDeps }): Command {
   const cmd = new Command("seat")
     .description("Inspect OpenRig seat observability state");
@@ -529,7 +539,7 @@ Examples:
   cmd
     .command("handover")
     .argument("<seat>", "Canonical session name or logical seat ref")
-    .option("--source <source>", "Source: fresh (default; launches a new agent), discovered:<id> (operator-prepared), fork:<id> (native fork of the source conversation), or rebuild (fresh agent primed from the seat's durable artifacts).")
+    .option("--source <source>", "Source: fresh (default; launches a new agent), discovered:<id> (operator-prepared), fork:<seat> (native fork of that seat's conversation, by its session name), or rebuild (fresh agent primed from the seat's durable artifacts).")
     .option("--reason <reason>", "Why the handover is happening")
     .option("--operator <address>", "Operator initiating the handover")
     .option("--dry-run", "Plan the handover without changing topology")
@@ -669,7 +679,7 @@ identity remains unverified; a version number alone cannot clear it.
   // Thin CLI over the daemon's SeatLifecycleService; refusals print message +
   // guidance + match list exactly as the daemon named them.
   const runLifecycleVerb = async (
-    path: "set-terminal-policy" | "set-model" | "set-cwd" | "set-effort" | "set-permissions" | "launch" | "launch-plan" | "continue" | "stop" | "clean" | "restart" | "restart-plan" | "restart-verify",
+    path: "set-terminal-policy" | "set-model" | "set-cwd" | "set-effort" | "set-permissions" | "set-permissions-to" | "launch" | "launch-plan" | "continue" | "stop" | "clean" | "restart" | "restart-plan" | "restart-verify",
     seat: string,
     body: Record<string, unknown>,
     opts: { json?: boolean },
@@ -852,6 +862,49 @@ Examples:
       });
     });
 
+  cmd
+    .command("set-permissions-to")
+    .argument("<seat>", "Canonical session name or logical seat ref")
+    .option("--to <person>", "The registered person who answers this seat's runner permission requests instead of the default person, such as mike@external")
+    .option("--clear", "Remove the override: the default person answers again")
+    .option("--deny", "Opt in to refusing every runner permission request on this seat")
+    .requiredOption("--reason <text>", "Audit reason recorded on the node.permissions_to_changed event")
+    .option("--operator <address>", "Operator recorded on the audit event (required outside a managed seat)")
+    .option("--json", "JSON output for agents")
+    .description("Choose who answers a runner seat's permission requests instead of the default person, or opt in to refusing them (audited)")
+    .addHelpText("after", `
+A runner seat's agent can ask permission for a tool call. The runner sends the
+request to OpenRig, and one person gets an ordinary ask with every option the agent
+offered; the answer goes back to that one request. That person is the default person
+(rig gateway default), unless this override names someone else. When no one can be
+asked, OpenRig's default is to permit the request.
+
+Only a registered person can be chosen (rig gateway human list). A seat address is
+refused. --clear removes the override. --deny is the opt-in for a seat whose requests
+should all be refused. See docs/reference/harness-runner-contract.md, section 7.
+Examples:
+  rig seat set-permissions-to dev-runner@my-rig --to mike@external --reason "Mike approves tool calls"
+  rig seat set-permissions-to dev-runner@my-rig --deny --reason "high-stakes repo: refuse every request"
+  rig seat set-permissions-to dev-runner@my-rig --clear --reason "back to the default person" --json`)
+    .action(async (seat: string, opts: { to?: string; clear?: boolean; deny?: boolean; reason: string; operator?: string; json?: boolean }) => {
+      if ([Boolean(opts.clear), Boolean(opts.deny), opts.to !== undefined].filter(Boolean).length !== 1) {
+        console.error("Pass exactly one of --to <person>, --clear or --deny.");
+        process.exitCode = 1;
+        return;
+      }
+      const body = opts.clear ? { clear: true, reason: opts.reason, operator: opts.operator }
+        : opts.deny ? { deny: true, reason: opts.reason, operator: opts.operator }
+        : { to: opts.to, reason: opts.reason, operator: opts.operator };
+      await runLifecycleVerb("set-permissions-to", seat, body, opts, (data) => {
+        const s = data["seat"] as { logicalId?: string; rigName?: string } | undefined;
+        const name = `${s?.logicalId}@${s?.rigName}`;
+        const owner = (value: unknown) => value === "deny" ? "refused (--deny)" : typeof value === "string" ? value : "the default person";
+        console.log(data["changed"] === false
+          ? `${name}'s permission requests already go to ${owner(data["to"])}. No change recorded.`
+          : `${name}'s permission requests: ${owner(data["from"])} -> ${owner(data["to"])} (audited).`);
+      });
+    });
+
   cmd
     .command("effort")
     .argument("<seat>", "Canonical session name or logical seat ref")
@@ -919,6 +972,10 @@ resume/successor launch reads it at call time. Examples:
     .option("--model <id>", "Target model; otherwise the seat's agent spec for the target, else the target's default")
     .option("--effort <level>", "Target effort, the target runtime's own level (not translated)")
     .option("--codex-config-profile <name>", "Codex configuration profile for a Codex successor")
+    .option("--clear-account-home", "Return to member/rig/daemon/native login selection through a fresh successor")
+    .option("--account-home <path>", "Isolated target CODEX_HOME or CLAUDE_CONFIG_DIR; also permits a fresh successor on the same runtime")
+    .option("--token-file <path>", "With a Claude account home, its required subscription token reference")
+    .option("--account-label <label>", "With --token-file, the selected account's label")
     .option("--permission-mode <mode>", "Target native permission choice; default: inherit the seat's member and rig policy")
     .option("--confirm-timeout <seconds>", "How long the successor has to confirm the handover (default 180)")
     .option("--dry-run", "Show the plan (settings, what is cleared, costs) without changing anything")
@@ -941,6 +998,9 @@ effort, Codex profile and native permission choice are cleared, not mapped; two
 processes run until cutover; messages sent to the seat during the switch are
 held, not typed, and need sending again afterwards.
 Codex and Claude Code, single-pane seats only. --dry-run lists the details.
+--clear-account-home returns to configured member/rig/daemon/native selection; it does not restore historical credentials.
+It uses a fresh successor and clears an account-home token reference only after confirmation. Do not combine it with
+--account-home or a new token reference. Anthropic's global active/default profile is not isolated by CLAUDE_CONFIG_DIR.
 
 A seat can switch itself (its CLI may end when it retires; the daemon finishes).
   rig seat set-runtime dev-impl@my-rig --runtime claude-code --reason "Codex limit" --dry-run
@@ -948,7 +1008,11 @@ A seat can switch itself (its CLI may end when it retires; the daemon finishes).
   rig seat status dev-impl@my-rig            # an open or interrupted switch shows here
   rig seat set-runtime dev-impl@my-rig --recover`)
     .action(async (seat: string, opts: { runtime?: string; reason?: string; model?: string; effort?: string; codexConfigProfile?: string;
-      permissionMode?: string; confirmTimeout?: string; dryRun?: boolean; confirm?: string; note?: string; recover?: boolean; json?: boolean }) => {
+      permissionMode?: string; accountHome?: string; clearAccountHome?: boolean; tokenFile?: string; accountLabel?: string; confirmTimeout?: string; dryRun?: boolean; confirm?: string; note?: string; recover?: boolean; json?: boolean }) => {
+      if (opts.clearAccountHome && (opts.accountHome !== undefined || opts.tokenFile !== undefined || opts.accountLabel !== undefined || opts.runtime === undefined)) {
+        console.error("Use --clear-account-home with --runtime and without --account-home, --token-file or --account-label.");
+        process.exitCode = 1; return;
+      }
       const modes = [opts.confirm !== undefined, opts.recover === true, opts.runtime !== undefined].filter(Boolean).length;
       if (modes !== 1) {
         console.error("Pass exactly one of --runtime <runtime> (with --reason), --confirm <switch-id> (with --note), or --recover.");
@@ -980,19 +1044,42 @@ A seat can switch itself (its CLI may end when it retires; the daemon finishes).
       const client = deps.clientFactory(getDaemonUrl(daemon));
       const base = `/api/seat/set-runtime/${encodeURIComponent(seat)}`;
       let res;
+      let switchRequested = false;
       try {
         res = opts.confirm !== undefined
           ? await client.post<Record<string, unknown>>(`${base}/confirm`, { switchId: opts.confirm, note: opts.note })
           : opts.recover
             ? await client.post<Record<string, unknown>>(`${base}/recover`, {}, { timeoutMs: 60_000 })
-            : await client.post<Record<string, unknown>>(base, {
-              runtime: opts.runtime, reason: opts.reason, dryRun: opts.dryRun === true,
-              ...(opts.model !== undefined ? { model: opts.model } : {}), ...(opts.effort !== undefined ? { effort: opts.effort } : {}),
-              ...(opts.codexConfigProfile !== undefined ? { codexConfigProfile: opts.codexConfigProfile } : {}),
-              ...(opts.permissionMode !== undefined ? { permissionMode: opts.permissionMode } : {}),
-              ...(confirmTimeoutSeconds !== undefined ? { confirmTimeoutSeconds } : {}),
-            }, { timeoutMs: opts.dryRun ? 30_000 : ((confirmTimeoutSeconds ?? 180) + 120) * 1000 });
+            : await (async () => {
+              const body = {
+                runtime: opts.runtime, reason: opts.reason, dryRun: opts.dryRun === true,
+                ...(opts.model !== undefined ? { model: opts.model } : {}), ...(opts.effort !== undefined ? { effort: opts.effort } : {}),
+                ...(opts.codexConfigProfile !== undefined ? { codexConfigProfile: opts.codexConfigProfile } : {}),
+                ...(opts.permissionMode !== undefined ? { permissionMode: opts.permissionMode } : {}),
+                ...(opts.accountHome !== undefined ? { accountHome: opts.accountHome } : {}),
+                ...(opts.clearAccountHome ? { clearAccountHome: true } : {}),
+                ...(opts.tokenFile !== undefined ? { tokenFile: opts.tokenFile, accountLabel: opts.accountLabel ?? "selected-account" } : {}),
+                ...(confirmTimeoutSeconds !== undefined ? { confirmTimeoutSeconds } : {}),
+              };
+              if (opts.accountHome !== undefined || opts.clearAccountHome) {
+                const preview = await client.post<Record<string, unknown>>(base, { ...body, dryRun: true }, { timeoutMs: 30_000 });
+                const token = opts.tokenFile ? { label: opts.accountLabel ?? "selected-account", tokenFile: opts.tokenFile } : undefined;
+                if (!accountHomePlanMatches(preview.data, opts.clearAccountHome ? null : opts.accountHome!, token)) {
+                  return { status: 409, data: { ok: false, code: "account_home_switch_unavailable",
+                    message: preview.data["message"] ?? "The daemon did not acknowledge the selected account home and token. No switch was requested." } };
+                }
+                if (opts.dryRun) return preview;
+              }
+              switchRequested = true;
+              return client.post<Record<string, unknown>>(base, body,
+                { timeoutMs: opts.dryRun ? 30_000 : ((confirmTimeoutSeconds ?? 180) + 120) * 1000 });
+            })();
       } catch (err) {
+        if ((opts.accountHome !== undefined || opts.clearAccountHome) && !switchRequested) {
+          const refusal = { ok: false, code: "account_home_preflight_failed", message: "The read-only account-home preflight failed; no switch was requested." };
+          if (opts.json) console.log(JSON.stringify(refusal)); else console.error(refusal.message);
+          process.exitCode = 1; return;
+        }
         if (!(err instanceof DaemonTimeoutError) || opts.dryRun) throw err;
         const unknown = { ok: false as const, code: "runtime_switch_outcome_unknown", status: "unknown",
           message: "The CLI stopped waiting; the daemon may still be switching the seat.",
@@ -1023,10 +1110,26 @@ A seat can switch itself (its CLI may end when it retires; the daemon finishes).
         console.log(`${name} already runs ${String(data["to"])}. Nothing changed.`);
         return;
       }
+      if (data["firstStart"] === true) {
+        const from = String(data["from"]), to = String(data["to"]);
+        console.log(data["dryRun"]
+          ? `${name} has never started: its saved runtime would change from ${from} to ${to}, and its first start would use it. Nothing changes until you run it without --dry-run.`
+          : `${name} has never started, so nothing was switched (no switch id): its saved runtime changed from ${from} to ${to}, and its first start uses it.`);
+        const permission = data["permission"] as { mode: string; source: string } | undefined;
+        if (permission) console.log(`  permissions: ${permission.source === "explicit" ? `${permission.mode} (as given)` : "the seat's member and rig policy"}`);
+        const dropped = (data["dropped"] as Array<{ setting: string; value: string }> | undefined) ?? [];
+        if (dropped.length) console.log(`Cleared from ${from}: ${dropped.map((d) => `${d.setting} ${d.value}`).join("; ")}. Its first start takes the rest from its agent spec for ${to}.`);
+        return;
+      }
       const label = (source: string) => source === "explicit" ? "as given" : source === "agent_spec" ? "from the seat's agent spec" : "the runtime's default";
       const settings = (data["settings"] ?? {}) as Record<string, { value: string | null; source: string }>;
       if (data["dryRun"]) console.log(`Runtime switch plan for ${name}: ${String(data["from"])} -> ${String(data["to"])}. Nothing changes until you run it without --dry-run.`);
+      else if (data["recoveryPending"] === true) console.log(`Recovery is still pending for ${name} (switch ${String(data["switchId"])}); the runtime switch is not complete.`);
       else console.log(`Switched ${name} from ${String(data["from"])} to ${String(data["to"])} (switch ${String(data["switchId"])}).`);
+      if (data["accountHome"] === null) console.log("Selected account home: configured defaults (member/rig, then daemon, then native)");
+      else if (data["accountHome"]) console.log(`Selected account home: ${String(data["accountHome"])}`);
+      const accountToken = data["accountToken"] as { label: string; tokenFile: string } | undefined;
+      if (accountToken) console.log(`Claude token reference: ${accountToken.label}, read from ${accountToken.tokenFile} at launch`);
       console.log(`Settings on ${String(data["to"])}:`);
       for (const [key, title] of [["model", "model"], ["effort", "effort"], ["advisor", "advisor"], ["codexConfigProfile", "Codex profile"]] as const) {
         const setting = settings[key];
@@ -1280,6 +1383,8 @@ session is refused (use rig unclaim). Stopping ends that agent's session and any
 work in progress: check it first (rig capture <seat>) and stop only a seat you
 own or were asked to stop. rig seat launch <seat> --fresh --reason <text> starts
 a fresh occupant there; it does not resume the stopped conversation.
+A community harness stop closes its managed pane; its native processes or compute
+may outlive it. Inspect their own lifecycle receipts.
 An intentionally stopped seat does not degrade an otherwise running rig.
 Running counts still exclude the stopped seat; stopping every seat leaves the
 rig stopped. Examples:
@@ -1290,6 +1395,7 @@ rig stopped. Examples:
         const s = data["seat"] as { logicalId?: string; rigName?: string } | undefined;
         console.log(`Stopped ${String(data["sessionName"])} (seat ${s?.logicalId}@${s?.rigName}).`);
         console.log("Session marked exited, binding cleared; siblings untouched. Relaunch via the normal launch surface.");
+        if (data["paneOnly"]) console.log("Only the community seat viewer was closed. Native execution, storage and compute may remain; inspect the selected environment or harness lifecycle.");
       });
     });
 
@@ -1464,7 +1570,7 @@ export function handoverCommand(depsOverride?: SeatDeps): Command {
   };
   return new Command("handover")
     .argument("<seat>", "Canonical session name or logical seat ref")
-    .option("--source <source>", "Successor source: fresh (default; launches a new agent), discovered:<id> (operator-prepared), fork:<id> (native fork of the source conversation), or rebuild (fresh agent primed from the seat's durable artifacts).")
+    .option("--source <source>", "Successor source: fresh (default; launches a new agent), discovered:<id> (operator-prepared), fork:<seat> (native fork of that seat's conversation, by its session name), or rebuild (fresh agent primed from the seat's durable artifacts).")
     .option("--reason <reason>", "Why the handover is happening")
     .option("--operator <address>", "Operator initiating the handover")
     .option("--dry-run", "Plan the handover without changing topology")
@@ -1472,8 +1578,8 @@ export function handoverCommand(depsOverride?: SeatDeps): Command {
     .description("Hand a seat to a successor: create -> deliver context -> verify continuity -> rebind")
     .addHelpText("after", `
 All four sources execute (OPR.0.5.5.5): fresh delivers a captured restore
-packet; discovered adopts an operator-prepared live session; fork:<id> launches
-a NATIVE FORK of the source conversation (the successor carries the incumbent
+packet; discovered adopts an operator-prepared live session; fork:<seat> launches
+a NATIVE FORK of that seat's conversation (the successor carries the incumbent
 context from its first byte); rebuild launches fresh and primes from the seat's
 durable artifact chain, recording exactly what it found. A source that cannot
 proceed (e.g. fork with no discoverable native id) refuses honestly before any

packages/cli/src/commands/send.ts

@@ -9,17 +9,17 @@ import { runCrossHostCommand, type RunCrossHostCommandOpts } from "../cross-host
 import { emitCrossHostError, emitCrossHostFailure, emitRemoteHttpFailure } from "../cross-host-cli-helpers.js";
 import { resolveCrossHostTarget } from "../cross-host-target.js";
 import { requestMayHaveArrived, runRemoteHttpOp } from "../remote-host-ops.js";
-import { readOpenRigEnv } from "../openrig-compat.js";
+import { readOpenRigUrl } from "../openrig-compat.js";
 import { resolveSenderSession, SENDER_FALLBACK } from "../sender-identity.js";
 import { COMPOSER_PROMPT_CLASS } from "@openrig/daemon/composer-prompts";
 import { resolveContextRef, walkSizedWarning } from "../context-resolve.js";
 
-const WAIT_FOR_IDLE_REQUEST_OVERHEAD_MS = 5_000;
-// A cross-host send waits for the remote to deliver, and the remote has no fixed deadline of its own (its tmux
-// calls run without a timeout), so no client budget is exact. This one sits well above a normal send, and above
-// fetch's 10 s connect timeout, so a host that never accepts the connection still reads as unreachable. A send
-// that outlasts it reads as unconfirmed, never as failed.
-const CROSS_HOST_SEND_TIMEOUT_MS = 30_000;
+// The reply follows the guarded paste/Enter (and optional verification), not acceptance of the request.
+// Successful sends can exceed the ordinary client's 5 s read budget. Keep a finite response wait on both
+// local and remote paths; a longer send or lost response remains unconfirmed and is never retried here.
+const SEND_RESPONSE_TIMEOUT_MS = 60_000;
+const UNCONFIRMED_SEND_GUIDANCE = "A busy agent may hold delivered input off-screen, so absence from a pane capture does not prove non-delivery. " +
+  "Wait for recipient acknowledgment or reconcile with the recipient before resending; another send or queue item risks duplicate work.";
 
 /** The shared composer glyph set (Claude `❯`, Codex `›`, Codex 0.153 `»`) from the daemon's
  * composer-prompts surface, so the CLI staged check can never disagree with the daemon guard. */
@@ -140,8 +140,7 @@ function printTransportFailure(err: DaemonConnectionError, opts?: { json?: boole
     ? "Delivery UNCONFIRMED — the daemon may have received and delivered the message."
     : "The message was not sent.";
   const action = outcomeUnknown
-    ? "Reconcile by EFFECT before any resend: check the pane/target for the message (rig capture <session>). " +
-      "A resend without checking risks a duplicate. Then 'rig daemon status' for the probe picture."
+    ? "Reconcile by EFFECT before any resend: check the target (rig capture <session>). " + UNCONFIRMED_SEND_GUIDANCE
     : "Inspect the configured target with 'rig status'; a failed health probe does not prove the daemon is stopped. " +
       "If the target is wrong, check OPENRIG_URL / RIGGED_URL or daemon.host + daemon.port. " +
       "If the daemon is confirmed stopped, run 'rig daemon start'.";
@@ -187,7 +186,7 @@ export function refuseEmptyMessage(body: string, verb: "send" | "broadcast", jso
  *  keeps the single-seat and fan-out callers from having to sequence it
  *  identically in two places. */
 async function resolveLocalDaemonUrl(deps: SendDeps): Promise<string> {
-  const envUrl = readOpenRigEnv("OPENRIG_URL", "RIGGED_URL");
+  const envUrl = readOpenRigUrl();
   if (envUrl) return envUrl;
   const status = await getDaemonStatus(deps.lifecycleDeps);
   if (status.state === "running" && status.port !== undefined) return getDaemonUrl(status);
@@ -283,7 +282,11 @@ determined (unknown / missing / stale telemetry) the send PROCEEDS with an
 advisory note — telemetry is advisory, not authority over whether agents can
 communicate. Use --wait-for-idle to send only after explicit idle evidence. Use
 --verify to confirm the message appeared in the pane only; it is not agent
-acknowledgement. With transport.verify_input_box on, a Claude Code or Codex send
+acknowledgement. Local and HTTP sends wait up to 60 s for the response, plus any
+--wait-for-idle duration. A timeout or missing pane text does not prove the
+message was not delivered: a busy agent can hold input off-screen. Reconcile
+with the recipient before resending or filing the same work as a queue item.
+With transport.verify_input_box on, a Claude Code or Codex send
 reads verified only when the seat was proven idle before the paste (its screen
 matched an idle layout recorded from Claude Code 2.1.296 or Codex 0.160.0; a
 screen that names another version reads unconfirmed, and a matching layout that
@@ -880,7 +883,7 @@ async function runCrossHostSend(
  * sender verbatim — honest provenance; unknown on the remote it degrades to
  * the shipped non-blocking advisory, never a refusal. `--verify` prints the
  * REMOTE route's verified/outcome verbatim (remote-authoritative, never
- * locally synthesized). Deadline: CROSS_HOST_SEND_TIMEOUT_MS, plus waitForIdleMs
+ * locally synthesized). Deadline: SEND_RESPONSE_TIMEOUT_MS, plus waitForIdleMs
  * when --wait-for-idle. A send that may have arrived reads as unconfirmed.
  *
  * Auth posture (named, v0): runRemoteHttpOp presents the REGISTRY bearer
@@ -918,7 +921,7 @@ async function runHttpHostSend(
     session, text: outboundText, exactText: raw ? undefined : text, verify: opts.verify, force: opts.force, waitForIdleMs,
     dangerouslyInteract: opts.dangerouslyInteract, typeWithoutChecking: opts.typeWithoutChecking, reason: opts.reason, actorSession: senderSession ?? null,
     ...(opts.urgent ? { urgent: true } : {}),
-  }, deps, { timeoutMs: (waitForIdleMs ?? 0) + CROSS_HOST_SEND_TIMEOUT_MS });
+  }, deps, { timeoutMs: (waitForIdleMs ?? 0) + SEND_RESPONSE_TIMEOUT_MS });
 
   // A send that may have reached the remote is unconfirmed, not unreachable: it may have been delivered, and a
   // resend without checking risks a duplicate.
@@ -926,7 +929,7 @@ async function runHttpHostSend(
   const shown = unconfirmed ? { ...result, failedStep: "remote-outcome-unknown" as const } : result;
   const unconfirmedGuidance = {
     consequence: "Delivery UNCONFIRMED — the remote daemon may have received and delivered the message.",
-    action: `Check the target before any resend: rig capture ${session} --host ${host.id}. A resend without checking risks a duplicate.`,
+    action: `Check the target before any resend: rig capture ${session} --host ${host.id}. ${UNCONFIRMED_SEND_GUIDANCE}`,
   };
 
   if (opts.json) {
@@ -1162,18 +1165,11 @@ function parseWaitForIdleMs(value: string | undefined): number | undefined | nul
   return Math.ceil(seconds * 1000);
 }
 
-function waitForIdleRequestOptions(waitForIdleMs: number | undefined): { timeoutMs: number } | undefined {
-  if (waitForIdleMs === undefined) return undefined;
-  return { timeoutMs: waitForIdleMs + WAIT_FOR_IDLE_REQUEST_OVERHEAD_MS };
-}
-
-function transportRequestOptions(waitForIdleMs?: number): { timeoutMs?: number; headers?: Record<string, string> } | undefined {
-  const waitOptions = waitForIdleRequestOptions(waitForIdleMs);
+function transportRequestOptions(waitForIdleMs?: number): { timeoutMs: number; headers?: Record<string, string> } {
   const headers = terminalAuthHeaders();
   const hasHeaders = Object.keys(headers).length > 0;
-  if (!waitOptions && !hasHeaders) return undefined;
   return {
-    ...(waitOptions ?? {}),
+    timeoutMs: (waitForIdleMs ?? 0) + SEND_RESPONSE_TIMEOUT_MS,
     ...(hasHeaders ? { headers } : {}),
   };
 }

packages/cli/src/commands/setup.ts

@@ -928,8 +928,24 @@ tmux scrollback:
       // OPR.0.3.3.04.2 (AC-1): the canonical ordered golden path. `rig setup` is
       // the primary surface for the new-operator sequence (status/doctor only
       // HINT back to it; the durable reference is docs/reference/getting-started.md).
+      // A failed doctor check also leaves setup not ready, so it's named below. So is the port check's warning, though setup
+      // stays ready: the configured port's daemon isn't confirmed as this user's, such as another user's daemon. Provider
+      // checks are the setup steps themselves, reused, so they aren't listed twice.
+      const stepIds = new Set(result.steps.map(step => step.id));
+      const doctorAttention = result.verification?.checks.filter(check => !stepIds.has(check.name)
+        && (check.status === "fail" || (check.name === "port" && check.status === "warn"))) ?? [];
+      const listChecks = () => {
+        for (const check of doctorAttention) {
+          console.log(`  ${check.name}: ${check.message}`);
+          if (check.fix) console.log(`    Fix: ${check.fix}`);
+        }
+      };
       if (result.ready) {
         console.log("\nSetup complete.\n");
+        if (doctorAttention.length) {
+          console.log("Some steps need attention:");
+          listChecks();
+        }
       } else {
         console.log("\nSome steps need attention:");
         if (opts.dryRun) console.log("Plan only: no setup changes were made; readiness was not checked.");
@@ -937,6 +953,7 @@ tmux scrollback:
           console.log(`  ${step.id}: ${step.message}`);
           if (step.fixHint) console.log(`    Fix: ${step.fixHint}`);
         }
+        listChecks();
         console.log("Only the harnesses selected for your project need a login; an unused harness does not (rerun with --providers to leave it out).");
         console.log("Run `rig doctor` to recheck the system and provider authentication.");
       }

packages/cli/src/commands/slack.ts

@@ -186,6 +186,7 @@ export function slackCommand(deps: SlackDeps = {}): Command {
     .option("--source-label <label>", "label shown in the posted message footer (where the queue lives)")
     .option("--secrets-env-file <path>", "path to the 0600 env file with SLACK_BOT_TOKEN / SLACK_APP_TOKEN")
     .option("--required-scopes <csv>", "comma-separated bot scopes to require at verify time")
+    .option("--reactions <mode>", "what a person's emoji reaction on an OpenRig post does: route (default) makes a row for the seat that asked; off ignores reactions, so an acknowledging emoji costs no agent turn")
     .option("--reason <reason>", "Reason recorded with the configuration change", "configure human delivery")
     .option("--actor <actor>", "Named operator when outside a managed seat")
     .action(async (opts) => {
@@ -200,6 +201,7 @@ export function slackCommand(deps: SlackDeps = {}): Command {
         sourceLabel: opts.sourceLabel ?? cur.sourceLabel,
         secretsEnvFile: opts.secretsEnvFile ?? cur.secretsEnvFile,
         requiredScopes: opts.requiredScopes ? String(opts.requiredScopes).split(",").map((s: string) => s.trim()).filter(Boolean) : cur.requiredScopes,
+        ...(opts.reactions !== undefined ? { reactions: opts.reactions } : {}),
       };
       const { runChannelOperation, channelStateDigest } = await import("@openrig/daemon/gateway-slack");
       const result = await runChannelOperation({

packages/cli/src/commands/specs.ts

@@ -20,6 +20,14 @@ import { daemonMayAnswer } from "../daemon-reachability.js";
 import { realDeps } from "./daemon.js";
 import type { StatusDeps } from "./status.js";
 
+const KNOWN_SPEC_KINDS = ["rig", "agent", "workflow"] as const;
+
+function warnIfUnknownKind(kind?: string): void {
+  if (kind !== undefined && !KNOWN_SPEC_KINDS.includes(kind as (typeof KNOWN_SPEC_KINDS)[number])) {
+    console.error(`Warning: unrecognized --kind '${kind}'. Supported kinds: ${KNOWN_SPEC_KINDS.join(", ")}`);
+  }
+}
+
 interface LibraryEntry {
   id: string;
   kind: "rig" | "agent" | "workflow";
@@ -147,10 +155,16 @@ function installSpecSource(source: AddSpecSource, userRoot: string): string {
 export async function resolveLibrarySpec(
   client: DaemonClient,
   nameOrId: string,
-  opts?: { kind?: LibraryEntry["kind"] },
+  opts?: { kind?: string },
 ): Promise<LibraryEntry> {
-  const res = await client.get<LibraryEntry[]>(opts?.kind ? `/api/specs/library?kind=${opts.kind}` : "/api/specs/library");
-  const entries = res.data ?? [];
+  const res = await client.get<LibraryEntry[]>(
+    opts?.kind !== undefined ? `/api/specs/library?kind=${encodeURIComponent(opts.kind)}` : "/api/specs/library",
+  );
+  const rawEntries = res.data ?? [];
+  const entries =
+    opts?.kind !== undefined && !KNOWN_SPEC_KINDS.includes(opts.kind as (typeof KNOWN_SPEC_KINDS)[number])
+      ? rawEntries.filter((e) => e.kind === opts.kind)
+      : rawEntries;
 
   // Try exact ID match first
   const byId = entries.find((e) => e.id === nameOrId);
@@ -219,11 +233,16 @@ Examples:
     .option("--json", "JSON output")
     .action(async (opts: { kind?: string; json?: boolean }) => {
       try {
+        warnIfUnknownKind(opts.kind);
         const client = await getClient();
-        const url = opts.kind ? `/api/specs/library?kind=${opts.kind}` : "/api/specs/library";
+        const url = opts.kind !== undefined ? `/api/specs/library?kind=${encodeURIComponent(opts.kind)}` : "/api/specs/library";
         const res = await client.get<LibraryEntry[]>(url);
         if (res.status !== 200) throw new Error(`Daemon returned HTTP ${res.status}`);
-        const entries = res.data ?? [];
+        const rawEntries = res.data ?? [];
+        const entries =
+          opts.kind !== undefined && !KNOWN_SPEC_KINDS.includes(opts.kind as (typeof KNOWN_SPEC_KINDS)[number])
+            ? rawEntries.filter((e) => e.kind === opts.kind)
+            : rawEntries;
 
         if (opts.json) {
           console.log(JSON.stringify(entries, null, 2));
@@ -231,7 +250,11 @@ Examples:
         }
 
         if (entries.length === 0) {
-          console.log("No specs in library. Add specs with: rig specs add <path>");
+          if (opts.kind !== undefined) {
+            console.log(opts.kind ? `No ${opts.kind} specs found` : "No specs found");
+          } else {
+            console.log("No specs in library. Add specs with: rig specs add <path>");
+          }
           return;
         }
 
@@ -250,10 +273,11 @@ Examples:
     .description("Show spec metadata and path")
     .option("--kind <kind>", "Disambiguate by kind (rig, agent, or workflow)")
     .option("--json", "JSON output")
-    .action(async (nameOrId: string, opts: { kind?: LibraryEntry["kind"]; json?: boolean }) => {
+    .action(async (nameOrId: string, opts: { kind?: string; json?: boolean }) => {
       try {
+        warnIfUnknownKind(opts.kind);
         const client = await getClient();
-        const entry = await resolveLibrarySpec(client, nameOrId, opts.kind ? { kind: opts.kind } : undefined);
+        const entry = await resolveLibrarySpec(client, nameOrId, opts.kind !== undefined ? { kind: opts.kind } : undefined);
 
         if (opts.json) {
           console.log(JSON.stringify(entry, null, 2));
@@ -278,10 +302,11 @@ Examples:
     .description("Show structured spec review, including managed app details")
     .option("--kind <kind>", "Disambiguate by kind (rig, agent, or workflow)")
     .option("--json", "JSON output")
-    .action(async (nameOrId: string, opts: { kind?: LibraryEntry["kind"]; json?: boolean }) => {
+    .action(async (nameOrId: string, opts: { kind?: string; json?: boolean }) => {
       try {
+        warnIfUnknownKind(opts.kind);
         const client = await getClient();
-        const entry = await resolveLibrarySpec(client, nameOrId, opts.kind ? { kind: opts.kind } : undefined);
+        const entry = await resolveLibrarySpec(client, nameOrId, opts.kind !== undefined ? { kind: opts.kind } : undefined);
         const res = await client.get<Record<string, unknown>>(`/api/specs/library/${encodeURIComponent(entry.id)}/review`);
         const review = res.data;
         const teamChoice = entry.sourceType === "builtin" && entry.kind === "rig" && ["starter", "factory"].includes(entry.name)

packages/cli/src/commands/start.ts

@@ -191,6 +191,13 @@ Examples:
 When the daemon is restoring a kernel a reboot left down, rig start goes on to
 your other rigs. Exit 0 then does not mean the kernel is ready (check rig
 status); a restore that returned without a ready kernel exits 1.
+
+--last restores a rig unless every seat it launched was stopped on purpose. A
+rig taken down with rig down before OpenRig 0.6.12, or whose seats exited or
+were lost without rig down, counts as last running: run rig down <rig> once
+(its id when names are shared) and --last keeps it down. --last never selects
+the kernel; under --no-kernel its automatic restore is skipped too, so bring it
+back with rig up kernel --existing.
 `);
   const getDepsF = (): StartDeps =>
     depsOverride ?? { lifecycleDeps: realDeps(), clientFactory: (url: string) => new DaemonClient(url) };

packages/cli/src/commands/startup-proof.ts

@@ -3,6 +3,7 @@ import os from "node:os";
 import path from "node:path";
 import { Command } from "commander";
 import { formatDaemonHostForUrl } from "../client.js";
+import { readOpenRigUrl } from "../openrig-compat.js";
 
 export interface StartupProofDeps {
   env?: NodeJS.ProcessEnv;
@@ -33,7 +34,9 @@ export function resolveStartupProofEndpoint(deps: StartupProofDeps = {}): Endpoi
   const env = deps.env ?? process.env;
   const readFile = deps.readFile ?? ((p: string) => fs.readFileSync(p, "utf8"));
   const home = first(env["OPENRIG_HOME"], env["RIGGED_HOME"]) ?? path.join(os.homedir(), ".openrig");
-  let baseUrl = first(env["OPENRIG_URL"], env["RIGGED_URL"]);
+  // Startup proof has always trimmed these and skipped blank ones; keep that, then pick and
+  // normalize through the shared reader like every other OPENRIG_URL reader.
+  let baseUrl = readOpenRigUrl({ OPENRIG_URL: env["OPENRIG_URL"]?.trim(), RIGGED_URL: env["RIGGED_URL"]?.trim() }) ?? null;
   let token = first(env["OPENRIG_ACTIVITY_HOOK_TOKEN"], env["RIGGED_ACTIVITY_HOOK_TOKEN"]);
 
   if (!baseUrl || !token) {

packages/cli/src/commands/status.ts

@@ -20,7 +20,9 @@ export interface StatusDeps {
 function formatSnapshotAge(snapshotAt: string | null): string {
   if (!snapshotAt) return "none";
   const now = Date.now();
-  const then = new Date(snapshotAt).getTime();
+  // Snapshot times are SQLite `datetime('now')` stamps: UTC written without a zone, which Date reads as local time.
+  const then = Date.parse(/^\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}(?:\.\d+)?$/.test(snapshotAt)
+    ? `${snapshotAt.replace(" ", "T")}Z` : snapshotAt);
   const diffMs = now - then;
   const diffMin = Math.floor(diffMs / 60_000);
   if (diffMin < 1) return "just now";

packages/cli/src/commands/terminal.ts

@@ -1,10 +1,11 @@
 // OPR.0.4.6.02 C3 — the `rig terminal` command family (the TUI-1 opaque verb +
-// the operator's terminal-provider-ride entry). Three subcommands, all hitting
+// the operator's terminal-provider-ride entry). The subcommands read
 // the ONE canonical daemon composer (`/api/terminal/...`):
 //
 //   rig terminal open <view> [--provider herdr|cmux|rex] [--json]
 //   rig terminal views [--json]
 //   rig terminal status [--provider herdr|cmux|rex] [--json]
+//   rig terminal attach <session> [--read-only]
 //
 // `<view>` resolves daemon-side: a rig name (per-rig derived) | `mission:<id>` |
 // `slice:<id>` (derived) | a saved-view id. The result is the ONE shared
@@ -17,22 +18,34 @@
 // exit 0 unless the daemon is unreachable.
 
 import { Command } from "commander";
-import { DaemonClient } from "../client.js";
+import { DaemonClient, terminalAuthHeaders } from "../client.js";
 import { getDaemonStatus, getDaemonUrl , daemonStatusGuard} from "../daemon-lifecycle.js";
 import { realDeps } from "./daemon.js";
 import type { StatusDeps } from "./status.js";
 import { openTerminalWindow, type WindowDeps } from "../terminal-window.js";
+import { assertLocalTmux, attachTmux, isLocalTmuxHost, readTmuxServerArgs, type TmuxAttachment } from "../tmux-attachment.js";
 
-export interface TerminalDeps extends StatusDeps { windowDeps?: WindowDeps }
+export interface TerminalDeps extends StatusDeps {
+  windowDeps?: WindowDeps;
+  attachTmux?: (target: TmuxAttachment) => Promise<number>;
+}
+export interface FollowControls {
+  enabled: boolean; maxTiles?: number; graceMs?: number; focusRetentionMs?: number;
+  pin?: string[]; unpin?: string[]; suppress?: string[]; unsuppress?: string[]; suppressionMs?: number;
+}
 
 /** The one shared open-result shape (mirrors the daemon `OpenViewResult`). */
 export interface OpenViewResult {
+  workspaceId?: string;
+  following?: { id: string; enabled: boolean; state: "following" | "empty" | "retained" | "disabled";
+    maxTiles: number; tiles: Array<{ key: string; seat: string; departedAt?: number; reason?: string }>;
+    events: string[]; overflow: Array<{ key: string; seat: string; reason: string }>; nextAt?: number };
   provider: string;
   session?: string;
   ok: boolean;
   opened: string[];
   absent: { seat: string; host: string | null; reason: string }[];
-  degraded: { seat: string; host: string; reason: string }[];
+  degraded: { seat: string; host: string; reason: string; code?: "pane_unconfirmed" }[];
   pages: number;
   error?: string;
   code?: string;
@@ -76,14 +89,19 @@ function humanOpen(r: OpenViewResult): string {
         ? `Terminal window requested. Reused ${target} workspace ${r.reusedWorkspace.id} for view ${JSON.stringify(r.reusedWorkspace.view)}.`
         : tiled > 0
           ? `${r.window ? "Terminal window requested. " : ""}Prepared ${tiled} tile(s) in ${target}${r.pages > 1 ? ` across ${r.pages} page(s)` : ""}.`
-          : `No tiles opened in ${target}.`,
+          : `No tiles ${r.degraded.some(d => d.code === "pane_unconfirmed") ? "confirmed" : "opened"} in ${target}.`,
   );
   if (r.window) lines.push(`  Terminal: ${r.window.app} (${r.window.surface}).`);
   if (r.space) lines.push(`  ${r.provider} session: ${JSON.stringify(r.space.label)} (${r.space.id}).`);
   if (r.error && r.code !== "terminal_window_failed") lines.push(`  provider: ${r.error}${r.code ? ` (${r.code})` : ""}`);
+  if (r.following) {
+    lines.push(`  Following: ${r.following.state}; tile limit ${r.following.maxTiles}.`);
+    for (const tile of r.following.tiles) if (tile.reason) lines.push(`  ${tile.seat}: ${tile.reason}`);
+    for (const hidden of r.following.overflow) lines.push(`  ${hidden.seat}: ${hidden.reason}`);
+  }
   for (const seat of r.opened) lines.push(`  ● ${seat}`);
   for (const a of r.absent) lines.push(`  ○ ${a.seat} — absent: ${a.reason}`);
-  for (const d of r.degraded) lines.push(`  ▲ ${d.seat} — skipped (${d.host}): ${d.reason}`);
+  for (const d of r.degraded) lines.push(`  ▲ ${d.seat} — ${d.code === "pane_unconfirmed" ? "unconfirmed" : "skipped"} (${d.host}): ${d.reason}`);
   for (const n of r.notes ?? []) lines.push(`  note: ${n}`);
   return lines.join("\n");
 }
@@ -95,7 +113,7 @@ function printOpen(json: boolean, r: OpenViewResult, status: number): void {
   } else {
     console.log(humanOpen(r));
   }
-  if (status >= 400 || (r.opened.length === 0 && !(r.ok && r.reusedWorkspace))) {
+  if (status >= 400 || (r.following ? !r.ok : r.opened.length === 0 && !(r.ok && r.reusedWorkspace))) {
     process.exitCode = status >= 500 ? 2 : 1;
   }
 }
@@ -113,27 +131,43 @@ export function terminalCommand(depsOverride?: TerminalDeps): Command {
 
   cmd
     .command("open")
-    .argument("<view>", "a rig name, mission:<id>, slice:<id>, qitem:<id> (current local obligation owners), or a saved-view id")
+    .argument("<view>", "a rig name, pod:<rig>/<pod>, seat:<rig>/<logical-id>, mission:<id>, slice:<id>, qitem:<id> (current local obligation owners), or a saved-view id")
     .description("Open a desktop terminal showing the view's live agents as interactive tiles")
     .option("--provider <name>", "herdr, cmux or rex: open in the running provider without a desktop window; each open adds a new workspace (inside Herdr, --window reuses a matching one); tmux requires --window. Inside Rex, the default opens in Rex")
     .option("--session <name>", "Herdr session on the daemon's host; --session default selects Herdr's default session")
     .option("--window", "Open a new desktop terminal tab/window (the default when --provider is omitted)")
     .option("--expected-plan <id>", "Open only if the view still matches this preview")
+    .option("--follow", "Keep one qitem:<id>, slice:<id-or-name> or mission:<id> workspace following current obligation owners (Linux Herdr 0.9.5/protocol 22)")
+    .option("--no-follow", "Stop following and keep the existing workspace and tiles as a fixed layout")
+    .option("--max-tiles <count>", "Following tile limit, 1–256; explicitly raise it to show more members", Number)
+    .option("--pin <seats...>", "Retain these following tiles after they leave the task")
+    .option("--unpin <seats...>", "Remove pins from these following tiles")
+    .option("--suppress <seats...>", "Temporarily hide these following tiles; agents keep running")
+    .option("--unsuppress <seats...>", "Show these following members again")
+    .option("--suppression-ms <ms>", "Suppression duration, default 60000, at most 86400000", Number)
+    .option("--grace-ms <ms>", "Departure grace, default 4000, at most 60000", Number)
+    .option("--focus-retention-ms <ms>", "Extra focused-reading retention, default 12000, at most 300000", Number)
     .option("--json", "JSON output for agents")
-    .action(async (view: string, opts: { provider?: string; session?: string; json?: boolean; window?: boolean; expectedPlan?: string }) => {
+    .action(async (view: string, opts: { provider?: string; session?: string; json?: boolean; window?: boolean; expectedPlan?: string;
+      follow?: boolean; maxTiles?: number; pin?: string[]; unpin?: string[]; suppress?: string[]; unsuppress?: string[];
+      suppressionMs?: number; graceMs?: number; focusRetentionMs?: number }) => {
       const deps = getDeps();
       const session = opts.session?.trim();
+      const { maxTiles, pin, unpin, suppress, unsuppress, suppressionMs, graceMs, focusRetentionMs } = opts;
+      const hasControls = [maxTiles, pin, unpin, suppress, unsuppress, suppressionMs, graceMs, focusRetentionMs].some(v => v !== undefined);
+      const following: FollowControls | undefined = opts.follow !== undefined || hasControls
+        ? { enabled: opts.follow !== false, maxTiles, pin, unpin, suppress, unsuppress, suppressionMs, graceMs, focusRetentionMs } : undefined;
       await withClient(deps, async (client) => {
         // Herdr's control socket can answer with no desktop client attached.
         // The default requests a desktop; an explicit provider reuses its workspace.
-        if (opts.window || !opts.provider) {
-          const result = await openTerminalWindow(client, view, opts.provider, deps.windowDeps, opts.expectedPlan, session);
+        if (following?.enabled !== false && (opts.window || !opts.provider)) {
+          const result = await openTerminalWindow(client, view, opts.provider, deps.windowDeps, opts.expectedPlan, session, following);
           printOpen(opts.json ?? false, result, 200);
           return;
         }
         let expectedPlan = opts.expectedPlan;
         if (session !== undefined) {
-          const preview = await client.get<{ session?: string; planId?: string; error?: string }>(`/api/terminal/preview?view=${encodeURIComponent(view)}&provider=${encodeURIComponent(opts.provider)}&session=${encodeURIComponent(session)}`);
+          const preview = await client.get<{ session?: string; planId?: string; error?: string }>(`/api/terminal/preview?view=${encodeURIComponent(view)}&provider=${encodeURIComponent(opts.provider ?? "herdr")}&session=${encodeURIComponent(session)}`);
           if (preview.status >= 400 || preview.data.session !== session || typeof preview.data.planId !== "string") {
             printResult(opts.json ?? false, { error: preview.data.error ?? "The daemon did not confirm the requested Herdr session. Update the daemon before opening a named session.", code: "terminal_session_unconfirmed" }, preview.status >= 400 ? preview.status : 409); return;
           }
@@ -143,7 +177,8 @@ export function terminalCommand(depsOverride?: TerminalDeps): Command {
           expectedPlan = preview.data.planId;
         }
         const body = { view, ...(opts.provider ? { provider: opts.provider } : {}), ...(session !== undefined ? { session } : {}), ...(expectedPlan !== undefined ? { expectedPlan } : {}) };
-        const res = await client.post<OpenViewResult>("/api/terminal/open", body, { timeoutMs: TERMINAL_OPEN_TIMEOUT_MS });
+        const res = await client.post<OpenViewResult>(following ? "/api/terminal/follow" : "/api/terminal/open",
+          { ...body, ...following }, { timeoutMs: TERMINAL_OPEN_TIMEOUT_MS, ...(following ? { headers: terminalAuthHeaders() } : {}) });
         if (session !== undefined && res.status < 400 && res.data?.session !== session) {
           printResult(opts.json ?? false, { ...res.data, ok: false, code: "terminal_session_unconfirmed",
             error: "The open response did not confirm the requested Herdr session. Its outcome is unknown; inspect that session before retrying." }, 409); return;
@@ -190,5 +225,31 @@ export function terminalCommand(depsOverride?: TerminalDeps): Command {
       });
     });
 
+  cmd.command("attach <session>")
+    .description("Attach this terminal to a session on the selected daemon's tmux server")
+    .option("--read-only", "View the session without sending input")
+    .action(async (session: string, opts: { readOnly?: boolean }) => {
+      const deps = getDeps();
+      await withClient(deps, async client => {
+        try {
+          if (client.baseUrl && !isLocalTmuxHost(client.baseUrl)) {
+            throw new Error("Terminal attachment uses local tmux; run this command on the selected daemon's machine.");
+          }
+          const response = await client.get<{ tmux?: { mode?: "inherit" | "isolated"; serverArgs?: unknown } }>("/api/terminal/status");
+          if (response.status >= 400) throw new Error(`Cannot read the tmux server (HTTP ${response.status}).`);
+          if (response.data.tmux?.serverArgs === undefined) {
+            throw new Error("This daemon cannot report its tmux server. Update and restart it before using terminal attach.");
+          }
+          await assertLocalTmux(client, response.data.tmux.mode);
+          process.exitCode = await (deps.attachTmux ?? attachTmux)({
+            session, serverArgs: readTmuxServerArgs(response.data.tmux.serverArgs), readOnly: opts.readOnly ?? false,
+          });
+        } catch (error) {
+          console.error(error instanceof Error ? error.message : String(error));
+          process.exitCode = 1;
+        }
+      });
+    });
+
   return cmd;
 }

packages/cli/src/commands/transcript.ts

@@ -76,7 +76,7 @@ OPENRIG_CLAUDE_DISABLE_ALTERNATE_SCREEN=0 — relaunch it to restore scrollback.
       const requestedLines = Number.isNaN(tailLines) || tailLines < 1 ? 50 : tailLines;
       const apiPath = useGrep
         ? `/api/transcripts/${encodeURIComponent(session)}/grep?pattern=${encodeURIComponent(opts.grep!)}`
-        : `/api/transcripts/${encodeURIComponent(session)}/tail?lines=${isNaN(tailLines) ? 50 : tailLines}`;
+        : `/api/transcripts/${encodeURIComponent(session)}/tail?lines=${requestedLines}`;
 
       // --- Cross-host path (CLI-direct GET to the remote daemon's shipped
       // transcript routes — the SAME paths the local path builds, origin

packages/cli/src/commands/tui.ts

@@ -6,6 +6,7 @@ import { pathToFileURL } from "node:url";
 import { existsSync } from "node:fs";
 import { openMissionControl, USAGE_LINES, type FrontDoorIo } from "../front-door.js";
 import { sharedTuiTarget, attachSharedTui } from "../shared-tui.js";
+import type { TmuxAttachment } from "../tmux-attachment.js";
 
 /** The registry-entry shape `rig tui commands` serializes (REGISTRY I2, ruling 64f1dbdf).
  *  Structural mirror of the TUI's CommandEntry data fields (functions are not serialized). */
@@ -38,8 +39,8 @@ async function loadRegistryFromDist(baseDir: string): Promise<TuiCommandEntry[]>
 
 export function tuiCommand(io: FrontDoorIo & {
   loadRegistry?: () => Promise<TuiCommandEntry[]>;
-  sharedTarget?: () => Promise<string>;
-  attachShared?: (target: string) => Promise<number>;
+  sharedTarget?: () => Promise<TmuxAttachment>;
+  attachShared?: (target: TmuxAttachment) => Promise<number>;
 } = {}): Command {
   const cmd = new Command("tui")
     .description("open the status dashboard, not agent chat (--shared is the OpenRig view fallback)")

packages/cli/src/commands/workspace.ts

@@ -87,7 +87,10 @@ export function workspaceCommand(depsOverride?: WorkspaceDeps): Command {
     .description(
       "Walk a workspace root, parse each .md file's YAML frontmatter, and emit a structured gap report. Advisory only — never modifies files. Default root: cwd.",
     )
-    .option("--kind <kind>", "Workspace kind to validate against: user | project | knowledge | lab | delivery")
+    .option(
+      "--kind <kind>",
+      "Workspace kind to validate against: user | project | knowledge | lab | delivery",
+    )
     .option("--no-recursive", "Do not descend into subdirectories")
     .option("--require-frontmatter", "Report a gap for every .md file without a frontmatter delimiter")
     .option("--max-files <n>", "Hard cap on .md files walked", "10000")
@@ -125,7 +128,13 @@ export function workspaceCommand(depsOverride?: WorkspaceDeps): Command {
             maxFiles,
           });
           if (res.status >= 400) {
-            console.error(JSON.stringify(res.data, null, 2));
+            if (opts.json) {
+              console.error(JSON.stringify(res.data, null, 2));
+            } else {
+              const data = res.data as { message?: string; error?: string } | null | undefined;
+              const msg = data?.message ?? data?.error ?? `Validate failed (HTTP ${res.status})`;
+              console.error(msg);
+            }
             process.exitCode = 1;
             return;
           }

packages/cli/src/config-store.ts

@@ -165,6 +165,8 @@ export interface RiggedConfig {
   // `rig config get/set terminal.status_bar` surface + twin parity.
   terminal: {
     statusBar: boolean;
+    /** Applied at daemon startup; a live previous server is retained until drained. */
+    tmuxServer: "inherit" | "isolated";
     /** Rows of tmux scrollback per pane; `rig setup` writes it into the managed `~/.tmux.conf` block. */
     historyLimit: number;
     inheritTmuxEnvironment: boolean;
@@ -355,6 +357,7 @@ const DEFAULTS = {
   // OPR.0.4.6.02 S1 — inner-tmux status bar OFF at launch by default.
   terminal: {
     statusBar: false,
+    tmuxServer: "inherit",
     historyLimit: 10000,
     inheritTmuxEnvironment: false,
   },
@@ -504,6 +507,7 @@ export const VALID_KEYS = [
   // boolean (default off), lockstep with the daemon settings-store twin
   // (the parity test pins both). Flip is future-launches-only (BR-1).
   "terminal.status_bar",
+  "terminal.tmux_server",
   // The managed ~/.tmux.conf block's history-limit, written by `rig setup`; lockstep with the daemon twin.
   "terminal.history_limit",
   "terminal.inherit_tmux_environment",
@@ -626,6 +630,7 @@ export const ENV_MAP: Record<ValidKey, { primary: string; legacy?: string }> = {
   "snapshots.periodic.retention_keep": { primary: "OPENRIG_SNAPSHOTS_PERIODIC_RETENTION_KEEP" },
   // OPR.0.4.6.02 S1 — net-new key; OPENRIG_* primary only (no RIGGED_* legacy).
   "terminal.status_bar": { primary: "OPENRIG_TERMINAL_STATUS_BAR" },
+  "terminal.tmux_server": { primary: "OPENRIG_TMUX_SERVER" },
   "terminal.history_limit": { primary: "OPENRIG_TERMINAL_HISTORY_LIMIT" },
   "terminal.inherit_tmux_environment": { primary: "OPENRIG_TERMINAL_INHERIT_TMUX_ENVIRONMENT" },
   "output.compact": { primary: "OPENRIG_OUTPUT_COMPACT" },
@@ -724,6 +729,7 @@ const KEY_TO_PATH: Record<ValidKey, string[]> = {
   "snapshots.periodic.interval_seconds": ["snapshots", "periodic", "intervalSeconds"],
   "snapshots.periodic.retention_keep": ["snapshots", "periodic", "retentionKeep"],
   "terminal.status_bar": ["terminal", "statusBar"],
+  "terminal.tmux_server": ["terminal", "tmuxServer"],
   "terminal.history_limit": ["terminal", "historyLimit"],
   "terminal.inherit_tmux_environment": ["terminal", "inheritTmuxEnvironment"],
   "output.compact": ["output", "compact"],
@@ -950,6 +956,11 @@ function percentageConstraint(key: string) {
 }
 
 const KEY_CONSTRAINTS: Partial<Record<ValidKey, (raw: string, coerced: string | number | boolean) => void>> = {
+  "terminal.tmux_server": (raw) => {
+    if (raw !== "inherit" && raw !== "isolated") {
+      throw new Error(`Invalid value for terminal.tmux_server: must be "inherit" or "isolated", got "${raw}"`);
+    }
+  },
   "daemon.db_conflict": (raw) => {
     if (raw !== "warn" && raw !== "stop") throw new Error("daemon.db_conflict must be warn or stop");
   },
@@ -1329,6 +1340,7 @@ export class ConfigStore {
       },
       terminal: {
         statusBar: v("terminal.status_bar") as boolean,
+        tmuxServer: v("terminal.tmux_server") as "inherit" | "isolated",
         historyLimit: v("terminal.history_limit") as number,
         inheritTmuxEnvironment: v("terminal.inherit_tmux_environment") as boolean,
       },

packages/cli/src/cross-host-target.ts

@@ -72,7 +72,9 @@ export function resolveCrossHostTarget(
   const suffix = rawTarget.slice(lastAt + 1);
 
   const unregisteredHint = suffix.length > 0
-    ? `no registered host '${suffix}' — if '${suffix}' was meant as a host, check \`rig host ls\``
+    ? `no registered host '${suffix}' — if '${suffix}' is another machine's self-id (the address a cross-host message shows), `
+      + `run \`rig host doctor <alias>\` for that machine's entry if it uses http (it learns the id), or set \`hostId: ${suffix}\` on that entry in hosts.yaml (any transport); `
+      + `\`rig host ls\` lists the registered hosts`
     : undefined;
 
   if (suffix.length === 0 || base.length === 0) {

packages/cli/src/daemon-lifecycle.ts

@@ -7,7 +7,7 @@ export { daemonMayAnswer } from "./daemon-reachability.js";
 import type { DaemonStartLock } from "./daemon-start-lock.js";
 import { DAEMON_STOP_WAIT_MS, DAEMON_SHUTDOWN_RECEIPT, type DaemonShutdownReceipt } from "@openrig/daemon/daemon-shutdown";
 import { ConfigStore } from "./config-store.js";
-import { OPENRIG_HOME, LEGACY_RIGGED_HOME, readOpenRigEnv } from "./openrig-compat.js";
+import { OPENRIG_HOME, LEGACY_RIGGED_HOME, readOpenRigEnv, readOpenRigUrl } from "./openrig-compat.js";
 // Types only: the module itself (with the daemon's yaml-reading system world) is imported where an
 // instance is initialized, so a command that only talks to the daemon never loads it.
 import type {
@@ -62,6 +62,9 @@ export interface DaemonStatus {
   eventLoop?: DaemonEventLoopEvidence;
   /** #275 — the OS code behind a failed health probe (e.g. EPERM when a sandbox blocked it). */
   probeErrorCode?: string;
+  /** T2 — a refused listener under a live start reservation (daemon-start.lock): a launch is in
+   *  progress and not listening yet. `pid` is the live launcher (else child) PID; `since` its start. */
+  startPending?: { pid: number; since?: string };
 }
 
 /** "unknown": the process may exist but this shell can neither signal nor inspect it. */
@@ -72,6 +75,18 @@ export interface GetDaemonStatusOptions {
   cleanupStaleState?: boolean;
 }
 
+/** True for `http://host:port` with an explicit, non-default port and no path, query or
+ *  fragment: the only OPENRIG_URL shape getDaemonUrl's host-and-port rebuild reaches unchanged
+ *  (without a port, or with :80, which URL reports as no port, status falls back to 7433). */
+function isPlainHttpOrigin(url: string): boolean {
+  try {
+    const u = new URL(url);
+    return u.protocol === "http:" && u.port !== "" && u.pathname === "/" && !u.search && !u.hash && !url.endsWith("/");
+  } catch {
+    return false;
+  }
+}
+
 /** Build the daemon HTTP URL from status. Uses persisted host or defaults to 127.0.0.1. */
 export function getDaemonUrl(status: DaemonStatus): string {
   return `http://${formatDaemonHostForUrl(status.host ?? DEFAULT_HOST)}:${status.port}`;
@@ -125,6 +140,16 @@ export function statusGuardMessage(status: DaemonStatus): DaemonNotRunningError
       ...causeCode,
     };
   }
+  // A refused probe under a live start reservation: a launch is in progress and nothing is listening yet.
+  if (status.startPending) {
+    const since = status.startPending.since ? ` since ${status.startPending.since}` : "";
+    return {
+      fact: `Daemon is starting (launch reserved by pid ${status.startPending.pid}${since}); it isn't listening yet.`,
+      consequence: "This command needs a responsive daemon.",
+      action: "Wait, then re-check with 'rig daemon status'.",
+      ...causeCode,
+    };
+  }
   // unverified, or running-but-unhealthy: we do NOT know it is down.
   return {
     fact: "Daemon did not respond — it may be busy or stopped (state not confirmed).",
@@ -830,6 +855,13 @@ async function startOwnedDaemon(opts: StartOptions, deps: LifecycleDeps, lock: D
       if (unverified) {
         throw new StartupChildPendingError(`Daemon child ${pid} responds at ${healthzUrl}, but listener verification is incomplete: ${lastProbe} Child left running; startup reservation retained. Inspect 'rig daemon status', daemon-start.lock and daemon.log before recovery.`);
       }
+      // Expiring the readiness budget is not evidence that this owned child failed.
+      // Keep launch exclusion until an operator can inspect a slow pre-bind startup.
+      assertChild();
+      if (deps.isProcessAlive(pid!)) {
+        assertChild();
+        throw new StartupChildPendingError(`Daemon startup is pending after the readiness deadline (child PID ${pid}). Readiness is not confirmed; child left running and startup reservation retained. Last probe: ${lastProbe}. Launch endpoint: ${healthzUrl}. Inspect that URL and ${LOG_FILE}; this check does not publish readiness, adopt the child or release the reservation. To restart manually, stop child PID ${pid} and wait for exit; inspect ${path.join(OPENRIG_DIR, "daemon-start.lock")}, and only after both recorded launcher and child PIDs are absent, archive it and run 'rig daemon start'. 'rig daemon stop' cannot target this unpublished child, and another start remains blocked by the reservation.`);
+      }
       throw new Error(`${summarizeDaemonStartFailure(healthzUrl, deps.readFile(resolveLifecycleFile(deps, "daemon.log")))}. Last probe: ${lastProbe}`);
     }
     assertChild();
@@ -905,7 +937,7 @@ function removeMatchingState(deps: LifecycleDeps, stateFile: string, state: Daem
 export async function stopDaemon(deps: LifecycleDeps): Promise<"stopped" | "no-target"> {
   const state = readState(deps);
   const configured = resolveConfiguredDaemonTarget();
-  const explicitUrl = readOpenRigEnv("OPENRIG_URL", "RIGGED_URL");
+  const explicitUrl = readOpenRigUrl();
   const stateUrl = state ? `http://${formatDaemonHostForUrl(state.host ?? DEFAULT_HOST)}:${state.port}` : undefined;
   const target = stateUrl ?? explicitUrl?.replace(/\/+$/, "") ?? `http://${formatDaemonHostForUrl(configured.host)}:${configured.port}`;
   if (stateUrl && explicitUrl && new URL(explicitUrl).origin !== new URL(stateUrl).origin) {
@@ -1024,12 +1056,52 @@ function findSiblingHome(deps: LifecycleDeps): DaemonStatus["siblingHint"] {
   return undefined;
 }
 
+/** T2 — read-only: the live launcher (else child) PID named by a start reservation, or undefined
+ *  when there is no reservation, it is unreadable, or every PID it names is dead. */
+function liveStartReservation(deps: LifecycleDeps): DaemonStatus["startPending"] {
+  try {
+    const file = path.join(deps.homeDir ?? OPENRIG_DIR, "daemon-start.lock");
+    const raw = deps.exists(file) ? deps.readFile(file) : null;
+    if (!raw) return undefined;
+    let launcherPid: number | undefined;
+    let childPid: number | undefined;
+    let since: string | undefined;
+    for (const line of raw.split("\n")) {
+      if (!line.trim()) continue;
+      try {
+        const rec = JSON.parse(line) as { launcherPid?: unknown; childPid?: unknown; startedAt?: unknown };
+        if (typeof rec.launcherPid === "number") launcherPid = rec.launcherPid;
+        if (typeof rec.childPid === "number") childPid = rec.childPid;
+        if (typeof rec.startedAt === "string") since = rec.startedAt;
+      } catch {
+        // partial line: keep whatever parsed
+      }
+    }
+    for (const pid of [launcherPid, childPid]) {
+      if (pid === undefined) continue;
+      const liveness: ProcessLiveness = deps.processLiveness
+        ? deps.processLiveness(pid)
+        : deps.isProcessAlive(pid) ? "alive" : "dead";
+      if (liveness !== "dead") return { pid, ...(since ? { since } : {}) };
+    }
+  } catch {
+    // best-effort only: a status read must never fail on the reservation
+  }
+  return undefined;
+}
+
 export async function getDaemonStatus(
   deps: LifecycleDeps,
   options: GetDaemonStatusOptions = {},
 ): Promise<DaemonStatus> {
-  // If OPENRIG_URL is set, bypass daemon.json and probe that URL directly
-  const openrigUrl = readOpenRigEnv("OPENRIG_URL", "RIGGED_URL");
+  // If OPENRIG_URL is set, bypass daemon.json and probe that URL directly. Commands reach
+  // it through getDaemonUrl's http://host:port rebuild, so only a plain http origin uses
+  // the normalized form; any other URL is probed as written, exactly as before.
+  const rawOpenrigUrl = readOpenRigEnv("OPENRIG_URL", "RIGGED_URL");
+  const normalizedOpenrigUrl = readOpenRigUrl();
+  const openrigUrl = normalizedOpenrigUrl !== undefined && isPlainHttpOrigin(normalizedOpenrigUrl)
+    ? normalizedOpenrigUrl
+    : rawOpenrigUrl;
   if (openrigUrl) {
     // An unverified daemon keeps the endpoint it was probed at, so a command can still make its request.
     const endpoint = (() => {
@@ -1070,9 +1142,13 @@ export async function getDaemonStatus(
       // 1ae863d2: the resolved home has NO daemon state — before asserting anything,
       // look for a live sibling home / HOME-MOVED marker (the wrong-home class).
       const siblingHint = findSiblingHome(deps);
-      if (siblingHint) return { state: "unverified", siblingHint, ...probeErrorCodeOf(err) };
-      return isRefusedError(err)
-        ? { state: "stopped" }
+      // T2: a refused listener under a live start reservation is a launch in progress, not down.
+      const refused = isRefusedError(err);
+      const startPending = refused ? liveStartReservation(deps) : undefined;
+      if (siblingHint) return { state: "unverified", siblingHint, ...probeErrorCodeOf(err), ...(startPending ? { startPending } : {}) };
+      if (refused && !startPending) return { state: "stopped" };
+      return refused
+        ? { state: "unverified", port: configured.port, host: configured.host, startPending, ...probeErrorCodeOf(err) }
         : { state: "unverified", port: configured.port, host: configured.host, ...probeErrorCodeOf(err) };
     }
   }
@@ -1100,7 +1176,12 @@ export async function getDaemonStatus(
       return { state: "running", port: state.port, host, pid: state.pid, healthy: ev.healthy, reason: ev.reason, eventLoop: ev.eventLoop, ...(ev.databaseWarnings ? { databaseWarnings: ev.databaseWarnings } : {}) };
     } catch (err) {
       // 1ae863d2: refusal = positive down; anything else (timeout/wedged/blocked) = unverified.
-      return isRefusedError(err) ? { state: "stopped" } : { state: "unverified", pid: state.pid, port: state.port, host, ...probeErrorCodeOf(err) };
+      // T2: unless a live start reservation says a launch is in progress.
+      if (isRefusedError(err)) {
+        const startPending = liveStartReservation(deps);
+        return startPending ? { state: "unverified", pid: state.pid, port: state.port, host, startPending, ...probeErrorCodeOf(err) } : { state: "stopped" };
+      }
+      return { state: "unverified", pid: state.pid, port: state.port, host, ...probeErrorCodeOf(err) };
     }
   }
 

packages/cli/src/daemon-reachability.ts

@@ -58,13 +58,14 @@ export function getOfflineHelpPath(): string | undefined {
 }
 
 /** Whether a command should make its real request: false only when the daemon is confirmed down
- *  (stopped, a stale record, or a refused health probe, meaning nothing is listening), when this
+ *  (stopped, a stale record, or a refused health probe, meaning nothing is listening, including one under a
+ *  pending start, whose `startPending` is set only on a refused probe), when this
  *  machine blocked the health probe, or when a live sibling home suggests the request would reach
  *  another home's daemon. A daemon that is alive but busy (its health check slow, failing or
  *  reporting a starved loop) gets the request, under its normal timeout. */
 export function daemonMayAnswer(status: DaemonStatus): boolean {
   if (status.state === "stopped" || status.state === "stale") return false;
-  if (status.probeErrorCode === "ECONNREFUSED") return false;
+  if (status.probeErrorCode === "ECONNREFUSED" || status.startPending) return false;
   if (status.siblingHint || isBlockedConnectionCode(status.probeErrorCode)) return false;
   return true;
 }

packages/cli/src/destroy-helpers.ts

@@ -59,7 +59,8 @@ export interface DestroyDeps {
   removePath: (path: string) => void;
   mkdirp: (path: string) => void;
   listManagedTmuxSessions: (dbPath: string) => string[];
-  killTmuxSession: (sessionName: string) => boolean;
+  resolveTmuxServerArgs?: (stateRoot: string) => string[];
+  killTmuxSession: (sessionName: string, serverArgs?: readonly string[]) => boolean;
   sleep: (ms: number) => Promise<void>;
   now: () => Date;
 }
@@ -179,6 +180,9 @@ function isLocalHost(host: string): boolean {
 }
 
 export async function executeDestroy(plan: DestroyPlan, deps: DestroyDeps): Promise<DestroyResult> {
+  // Resolve before stopping the daemon or moving its state. An unreadable record
+  // must not turn a managed-session cleanup into a kill on the caller's server.
+  const serverArgs = plan.scope === "all" ? deps.resolveTmuxServerArgs?.(plan.stateRoot) : undefined;
   const warnings = [...plan.warnings];
   const backupPaths: string[] = [];
   let stateRecreated = false;
@@ -239,7 +243,7 @@ export async function executeDestroy(plan: DestroyPlan, deps: DestroyDeps): Prom
   let tmuxMissing = 0;
   if (plan.scope === "all") {
     for (const sessionName of plan.managedTmuxSessions) {
-      if (deps.killTmuxSession(sessionName)) tmuxKilled += 1;
+      if (deps.killTmuxSession(sessionName, serverArgs)) tmuxKilled += 1;
       else tmuxMissing += 1;
     }
   }
@@ -284,9 +288,9 @@ export function findListeningPidWithLsof(port: number): number | null {
   }
 }
 
-export function killTmuxSessionWithCli(sessionName: string): boolean {
+export function killTmuxSessionWithCli(sessionName: string, serverArgs: readonly string[] = []): boolean {
   try {
-    execFileSync("tmux", ["kill-session", "-t", sessionName], { stdio: "ignore" });
+    execFileSync("tmux", [...serverArgs, "kill-session", "-t", `=${sessionName}`], { stdio: "ignore" });
     return true;
   } catch {
     return false;

packages/cli/src/host-bindings.ts

@@ -161,7 +161,7 @@ export function recordHostObservation(args: {
 
 function recordUnderLock(alias: string, observedHostId: string, observedUrl: string | undefined, at: string, path: string): HostObservationOutcome {
   const file = loadHostBindings(path);
-  const existing = file.bindings[alias];
+  const existing = Object.hasOwn(file.bindings, alias) ? file.bindings[alias] : undefined;
 
   let result: HostObservationOutcome;
   if (!existing) {

packages/cli/src/host-registry.ts

@@ -372,3 +372,128 @@ export function addHostEntry(rawEntry: Record<string, unknown>, path: string = d
   }
   return { ok: true, path, entry };
 }
+
+// ---------------------------------------------------------------------------
+// OPR.0.4.6.MH1 FR-1 extension (issue #2128, maintainer boundary) — the
+// edit/remove write paths, through the SAME ONE write contract as add:
+// load-validate (a present-but-invalid registry is NEVER overwritten),
+// candidate through validateHostRegistry (add-time errors are load-time
+// errors, verbatim), atomic tmp+rename at 0600.
+//
+// Boundary pinned here: an edit keeps the host's id and transport FIXED;
+// unknown fields fail loud with the allowed set (a typo'd flag must never
+// silently no-op an edit); an empty string clears an OPTIONAL field (user,
+// notes) — clearing is how a wrong value is dropped, while required fields
+// (target, url) keep the validator's non-empty rule; remove deletes ONLY the
+// registry entry (no credential files, no learned identity, no selection).
+//
+// The daemon twin (packages/daemon/src/domain/hosts/hosts-registry-writer.ts)
+// mirrors updateHostEntry/removeHostEntry verbatim; byte parity is pinned by
+// packages/daemon/test/hosts-registry-parity.test.ts.
+// ---------------------------------------------------------------------------
+
+/** Fields an edit may touch, per transport. id/transport never appear here. */
+const EDITABLE_FIELDS: Record<"ssh" | "http", readonly string[]> = {
+  ssh: ["target", "user", "notes"],
+  http: ["url", "notes"],
+};
+
+/** Optional editable fields an empty string CLEARS (drops the key). */
+const CLEARABLE_FIELDS = new Set(["user", "notes"]);
+
+export type UpdateHostResult =
+  | { ok: true; path: string; entry: HostEntry; previous: HostEntry }
+  | { ok: false; error: string };
+
+export type RemoveHostResult =
+  | { ok: true; path: string; entry: HostEntry }
+  | { ok: false; error: string };
+
+/** The write-path load: MISSING is an error here (nothing to edit/remove) with
+ *  the loader's canonical message; INVALID is a loud refusal, never clobbered. */
+function loadExistingForWrite(path: string): { ok: true; existing: HostEntry[] } | { ok: false; error: string } {
+  const loaded = loadHostRegistry(path);
+  if (loaded.ok) return { ok: true, existing: loaded.registry.hosts };
+  const prefix = existsSync(path) ? "refusing to modify an invalid registry: " : "";
+  return { ok: false, error: prefix + loaded.error };
+}
+
+/** Unknown-id error for a WRITE: names the id, lists known ids (resolveHost's shape). */
+function unknownHostIdError(existing: HostEntry[], id: string): string {
+  const knownIds = existing.map(h => h.id).slice(0, 10);
+  const hint = knownIds.length > 0
+    ? ` Known host ids: ${knownIds.join(", ")}${existing.length > knownIds.length ? ` (+${existing.length - knownIds.length} more)` : ""}.`
+    : " (registry is empty)";
+  return `unknown host id '${id}'.${hint}`;
+}
+
+/** The atomic registry write the write contract shares (tmp+rename, 0600). */
+function writeRegistryFile(path: string, hosts: HostEntry[]): string | null {
+  try {
+    mkdirSync(dirname(path), { recursive: true });
+    const tmp = join(dirname(path), `.hosts.yaml.tmp-${process.pid}`);
+    writeFileSync(tmp, yaml().stringify({ hosts }), { mode: 0o600 });
+    renameSync(tmp, path);
+    return null;
+  } catch (err) {
+    return `failed to write host registry at ${path}: ${(err as Error).message}`;
+  }
+}
+
+/** Apply an edit to ONE entry: id/transport fixed, unknown fields refused,
+ *  empty string clears an optional field, then the WHOLE registry validates. */
+export function updateHostEntry(id: string, changes: Record<string, unknown>, path: string = defaultHostRegistryPath()): UpdateHostResult {
+  const loaded = loadExistingForWrite(path);
+  if (!loaded.ok) return loaded;
+  const idx = loaded.existing.findIndex(h => h.id === id);
+  if (idx === -1) return { ok: false, error: unknownHostIdError(loaded.existing, id) };
+  const previous = loaded.existing[idx]!;
+
+  if (changes["id"] !== undefined && changes["id"] !== id) {
+    return { ok: false, error: `an edit keeps the host's id fixed — '${id}' cannot become '${String(changes["id"])}'` };
+  }
+  if (changes["transport"] !== undefined) {
+    return { ok: false, error: `an edit keeps the host's transport fixed — '${id}' stays '${previous.transport}'` };
+  }
+  const allowed = EDITABLE_FIELDS[previous.transport];
+  const effective: Record<string, unknown> = {};
+  for (const [key, value] of Object.entries(changes)) {
+    if (key === "id") continue; // an id echo naming the same host is a no-op
+    if (!allowed.includes(key)) {
+      return { ok: false, error: `field '${key}' is not editable on an ${previous.transport} host (editable fields: ${allowed.join(", ")})` };
+    }
+    effective[key] = value;
+  }
+  if (Object.keys(effective).length === 0) {
+    return { ok: false, error: "no changes given" };
+  }
+
+  const updated = { ...previous } as Record<string, unknown>;
+  for (const [key, value] of Object.entries(effective)) {
+    if (CLEARABLE_FIELDS.has(key) && value === "") delete updated[key];
+    else updated[key] = value;
+  }
+  const candidate = loaded.existing.map((h, i) => (i === idx ? updated as unknown as HostEntry : h));
+  const validated = validateHostRegistry({ hosts: candidate }, path);
+  if (!validated.ok) return { ok: false, error: validated.error };
+  const entry = validated.registry.hosts.find(h => h.id === id);
+  if (!entry) return { ok: false, error: `host '${id}' did not survive validation` };
+  const writeError = writeRegistryFile(path, validated.registry.hosts);
+  if (writeError) return { ok: false, error: writeError };
+  return { ok: true, path, entry, previous };
+}
+
+/** Remove ONE entry — only the registry entry; nothing else is touched. */
+export function removeHostEntry(id: string, path: string = defaultHostRegistryPath()): RemoveHostResult {
+  const loaded = loadExistingForWrite(path);
+  if (!loaded.ok) return loaded;
+  const idx = loaded.existing.findIndex(h => h.id === id);
+  if (idx === -1) return { ok: false, error: unknownHostIdError(loaded.existing, id) };
+  const entry = loaded.existing[idx]!;
+  const remaining = loaded.existing.filter((_, i) => i !== idx);
+  const validated = validateHostRegistry({ hosts: remaining }, path);
+  if (!validated.ok) return { ok: false, error: validated.error };
+  const writeError = writeRegistryFile(path, validated.registry.hosts);
+  if (writeError) return { ok: false, error: writeError };
+  return { ok: true, path, entry };
+}

packages/cli/src/index.ts

@@ -1,4 +1,5 @@
 #!/usr/bin/env node
+import { environmentCommand, type EnvironmentCommandDeps } from "./commands/environment.js";
 import { Command } from "commander";
 import { realpathSync } from "node:fs";
 import { fileURLToPath } from "node:url";
@@ -143,6 +144,7 @@ export interface ProgramDeps {
   skillDeps?: StatusDeps;
   agentImageDeps?: StatusDeps;
   communityHarnessDeps?: StatusDeps;
+  environmentDeps?: EnvironmentCommandDeps;
   forkDeps?: StatusDeps;
   workspaceDeps?: WorkspaceDeps;
   whoamiDeps?: StatusDeps;
@@ -227,6 +229,7 @@ export function createProgram(depsOverride?: ProgramDeps): Command {
   program.addCommand(skillCommand(depsOverride?.skillDeps));
   program.addCommand(agentImageCommand(depsOverride?.agentImageDeps));
   program.addCommand(communityHarnessCommand(depsOverride?.communityHarnessDeps));
+  program.addCommand(environmentCommand(depsOverride?.environmentDeps));
   program.addCommand(forkCommand(depsOverride?.forkDeps));
   program.addCommand(workspaceCommand(depsOverride?.workspaceDeps));
   program.addCommand(rigModeCommand(depsOverride?.rigModeDeps));

packages/cli/src/lazy-program.ts

@@ -65,6 +65,7 @@ export const COMMAND_LOADERS: ReadonlyMap<string, Load> = new Map<string, Load>(
   ["skill", async () => (await import("./commands/skill.js")).skillCommand()],
   ["agent-image", async () => (await import("./commands/agent-image.js")).agentImageCommand()],
   ["community-harness", async () => (await import("./commands/community-harness.js")).communityHarnessCommand()],
+  ["environment", async () => (await import("./commands/environment.js")).environmentCommand()],
   ["fork", async () => (await import("./commands/fork.js")).forkCommand()],
   ["workspace", async () => (await import("./commands/workspace.js")).workspaceCommand()],
   ["mode", async () => (await import("./commands/rig-mode.js")).rigModeCommand()],

packages/cli/src/lib/account-home-switch.ts

@@ -0,0 +1,7 @@
+/** An older daemon can ignore new fields. A read-only plan must acknowledge the exact selection before a mutation. */
+export function accountHomePlanMatches(plan: Record<string, unknown>, home: string | null,
+  token?: { label: string; tokenFile: string }): boolean {
+  if (plan["ok"] !== true || plan["dryRun"] !== true || plan["accountHome"] !== home) return false;
+  const selected = plan["accountToken"] as { label?: unknown; tokenFile?: unknown } | null | undefined;
+  return token ? selected?.label === token.label && selected?.tokenFile === token.tokenFile : selected == null;
+}

packages/cli/src/lib/scope/scope-audit.ts

@@ -166,6 +166,11 @@ export function deriveMissionDependencyGraph(input: MissionDependencyGraphInput)
 // without this file changing too).
 export const MISSION_BRIEF_HEADERS = ["What & why", "Building", "Progress", "Proven", "Needs you", "Pointers"];
 
+// A slice directory's name: two or more digits, a dash, then the slug. `rig scope slice create` numbers past 99
+// (100-…), so a two-digit-only check flags names OpenRig made. The CLI audit and the daemon's audit route both read
+// this one pattern.
+export const SLICE_DIRECTORY_NAME = /^\d{2,}-/;
+
 function childPath(parent: string, child: string): string {
   return parent.endsWith("/") ? `${parent}${child}` : `${parent}/${child}`;
 }

packages/cli/src/lib/scope/scope-fs.ts

@@ -554,8 +554,9 @@ export function findSlice(
         continue;
       }
       const mission = buildMissionInfo(missionsRoot, owningMissionPath);
-      const sliceRoot = path.dirname(candidate);
-      return buildSliceInfo(mission, sliceRoot, path.basename(candidate));
+      // The directory and the name both come from the resolved path, so `<slice>/.` and `<slice>/sub/..` select the slice.
+      const resolved = path.resolve(candidate);
+      return buildSliceInfo(mission, path.dirname(resolved), path.basename(resolved));
     }
   }
   if (firstContextError) throw firstContextError;
@@ -567,9 +568,18 @@ export function findSlice(
 }
 
 function findOwningMission(missionsRoot: string, slicePath: string): string | null {
+  // path.dirname never ends in a separator, so compare against the normalized root
+  // (a configured or --workspace root may be written as `/w/missions/`); resolve the
+  // slice the same way so a relative root and a relative candidate stay comparable.
+  // A path that leaves the root, such as `<root>/../other/m1/slices/x`, then falls back
+  // to the walk as written, where it reads as before.
+  return walkToMission(path.resolve(missionsRoot), path.resolve(slicePath)) ?? walkToMission(missionsRoot, slicePath);
+}
+
+function walkToMission(root: string, slicePath: string): string | null {
   let dir = path.dirname(slicePath);
-  while (dir.startsWith(missionsRoot) && dir !== missionsRoot) {
-    if (path.dirname(dir) === missionsRoot) return dir;
+  while (dir.startsWith(root) && dir !== root) {
+    if (path.dirname(dir) === root) return dir;
     dir = path.dirname(dir);
   }
   return null;

packages/cli/src/mcp-server.ts

@@ -18,6 +18,15 @@ const WAIT_FOR_IDLE_REQUEST_OVERHEAD_MS = 5_000;
 /** The most text rig_context_get and rig_artifact_read return in one call; a longer file or section is cut, and says so. */
 const CONTEXT_TEXT_LIMIT = 48_000;
 
+/** Where a cut of text at end goes: one unit earlier when end falls inside a surrogate pair (an emoji, say), so the
+ *  cut text and the rest after it both hold whole characters. */
+function wholeCharEnd(text: string, end: number): number {
+  if (end <= 0 || end >= text.length) return end;
+  const before = text.charCodeAt(end - 1);
+  const after = text.charCodeAt(end);
+  return before >= 0xd800 && before <= 0xdbff && after >= 0xdc00 && after <= 0xdfff ? end - 1 : end;
+}
+
 type TextResult = {
   content: Array<{ type: "text"; text: string }>;
   isError?: true;
@@ -399,7 +408,7 @@ function artifactView(source: string, read: FileRead, offset: number): Record<st
   const base = { available: true, source, root: read.root, path: read.path, size: read.size, mtime: read.mtime };
   if (read.binary) return { ...base, binary: true, note: "A binary file: not shown as text." };
   const text = read.content ?? "";
-  const end = offset + CONTEXT_TEXT_LIMIT;
+  const end = wholeCharEnd(text, offset + CONTEXT_TEXT_LIMIT);
   return {
     ...base, binary: false, offset, text: text.slice(offset, end), totalChars: text.length,
     truncated: text.length > end || read.truncated === true,
@@ -514,6 +523,7 @@ export const MCP_OPERATIONS: ReadonlyArray<{ tool: string; routes: readonly stri
   { tool: "rig_queue_list", routes: ["GET /api/queue/list"], effect: "read", access: "work" },
   { tool: "rig_queue_claim", routes: ["POST /api/queue/:qitemId/claim"], effect: "destructive", access: "work" },
   { tool: "rig_queue_update", routes: ["POST /api/queue/:qitemId/update"], effect: "destructive", access: "work" },
+  { tool: "rig_environment_export", routes: ["GET /api/queue/:qitemId", "POST /api/environments/:bindingId/export"], effect: "additive", access: "work" },
   { tool: "rig_telemetry_events", routes: ["GET /api/telemetry/v1/events"], effect: "read", access: "owner" },
   { tool: "rig_whoami", routes: ["GET /api/whoami"], effect: "read", access: "work" },
   { tool: "rig_team", routes: ["GET /api/rigs/:rigId", "GET /api/rigs/:rigId/nodes", "GET /healthz"], effect: "read", access: "work" },
@@ -530,12 +540,13 @@ export const MCP_OPERATIONS: ReadonlyArray<{ tool: string; routes: readonly stri
 
 /** How a member works its inbox, and asks a person while keeping the work: one versioned text, carried in this server's
  *  instructions and quoted in docs/reference/hosted-members.md. Change it here, bump the version, and update the quote. */
-export const MEMBER_WORKFLOW_VERSION = "member-workflow/v1";
+export const MEMBER_WORKFLOW_VERSION = "member-workflow/v2";
 /** What the connection itself authorizes. */
 const MEMBER_AUTHORITY = "If the person connected you, that's basically authorized: claim the items in your inbox and do them without asking whether you may.";
 export const MEMBER_WORKFLOW = [
   `Member workflow (${MEMBER_WORKFLOW_VERSION}). Your inbox is the items addressed to your seat: rig_queue_list with owned set. It's yours, not the team's queue. A wake, a ring or a message only points at it: list your inbox, then act.`,
-  "For each item: claim it (rig_queue_claim), do it, then close it done with the answer in resolution (rig_queue_update), or hand it on (rig_queue_handoff). Listing or reading an item doesn't take it.",
+  "For each item: claim it (rig_queue_claim), do it, then close it done with closure reason no-follow-on and the answer in resolution (rig_queue_update), or hand it on (rig_queue_handoff). Listing or reading an item doesn't take it.",
+  "Say what you can reach from your grant, not from the tool list: a listed tool isn't access. rig_artifact_read's answer for an item names the shared roots you may read, and none means no shared files.",
   MEMBER_AUTHORITY,
   "When an item needs a person's approval before you can finish it, don't close it and don't take the action: keep the item and ask.",
   "If it came from a person (an address ending in @external that isn't a shared channel; the item's 'Shared channel:' line names one), set it blocked with blockedOn that person and, as its summary, the question with a short form of what you'll do: the person sees the summary and the item, not your note, which keeps the record. Link a longer draft as an https evidenceRef. Their answer returns the item to you in progress.",
@@ -1349,7 +1360,7 @@ export function createMcpServer(client: DaemonClient, options: { seat?: string;
       inputSchema: z.object({
         qitemId: z.string().describe("The queue item to update"),
         state: z.string().optional().describe("New state (e.g. in-progress, blocked, done)"),
-        closureReason: z.string().optional().describe("Why it closed (as rig queue update --closure-reason)"),
+        closureReason: z.string().optional().describe("Why it closed: no-follow-on for finished work, or handed_off_to, denied, canceled, escalation or superseded (as rig queue update --closure-reason)"),
         closureTarget: z.string().optional().describe("Closure target, when the reason needs one"),
         note: z.string().optional().describe("Transition note: progress, or why this transition happened (required to reopen)"),
         resolution: z.string().optional().describe("The outcome the item keeps, stored only by the update that closes an open item (as rig queue update --resolution)"),
@@ -1374,6 +1385,28 @@ export function createMcpServer(client: DaemonClient, options: { seat?: string;
     },
   );
 
+  // Artifact publication uses this actual authenticated work caller, never a body actor or activity token.
+  if (offered("rig_environment_export")) server.registerTool(
+    "rig_environment_export",
+    {
+      description: "Transfer a bound worker's actual JSON artifact to the host, verify its bytes/schema/context and append a durable receipt to its existing work. Does not close work, stop compute or restore native history. Requires an authenticated HTTP work connection.",
+      inputSchema: z.object({
+        bindingId: z.string(), operationId: z.string(), taskId: z.string(), generation: z.string(), path: z.string(),
+        sha256: z.string(), bytes: z.number().int().positive().max(1048576),
+        schema: z.object({ format: z.literal("json-object/v1"), fields: z.record(z.string(), z.enum(["string", "number", "boolean"])), required: z.array(z.string()) }),
+        context: z.record(z.string(), z.union([z.string(), z.number(), z.boolean()])),
+      }),
+      annotations: ADDITIVE,
+    },
+    async ({ bindingId, ...input }) => {
+      if (!access) return { content: [{ type: "text" as const, text: "This export requires a validated HTTP work grant; a declared stdio/local identity is not authentication." }], isError: true };
+      try {
+        const denied = await itemOutOfScope(input.taskId, "owner"); if (denied) return denied;
+        return mapResult(await client.post(`/api/environments/${encodeURIComponent(bindingId)}/export`, input), undefined, { structured: true });
+      } catch (error) { return caught(error); }
+    },
+  );
+
   // 24. rig_telemetry_events — one bounded page of event metadata, the page `rig telemetry events` reads
   if (offered("rig_telemetry_events")) server.registerTool(
     "rig_telemetry_events",
@@ -1580,7 +1613,7 @@ export function createMcpServer(client: DaemonClient, options: { seat?: string;
         return mapResult({ status: 200, data: {
           address: found.address ?? address, packRef: found.packRef, filePath: found.filePath,
           ...(found.headerPath !== undefined ? { headerPath: found.headerPath } : {}),
-          text: cut ? text.slice(0, CONTEXT_TEXT_LIMIT) : text,
+          text: cut ? text.slice(0, wholeCharEnd(text, CONTEXT_TEXT_LIMIT)) : text,
           ...(cut ? { truncated: true, totalChars: text.length } : {}),
         } }, undefined, { structured: true });
       } catch (err) {

packages/cli/src/openrig-compat.ts

@@ -36,8 +36,8 @@ export function isFixtureScopedHome(home: string): boolean {
   return underTemp && /(^|\/)openrig-qa[^/]*/.test(home);
 }
 
-export function getOpenRigHome(): string {
-  const configured = readOpenRigEnv("OPENRIG_HOME", "RIGGED_HOME");
+export function getOpenRigHome(env: NodeJS.ProcessEnv = process.env): string {
+  const configured = readOpenRigEnv("OPENRIG_HOME", "RIGGED_HOME", env);
   if (configured !== undefined) return configured;
   return join(homedir(), ".openrig");
 }
@@ -55,12 +55,12 @@ function warnOnce(key: string, message: string): void {
   console.warn(message);
 }
 
-export function readOpenRigEnv(primary: string, legacy?: string): string | undefined {
-  const primaryValue = process.env[primary];
+export function readOpenRigEnv(primary: string, legacy?: string, env: NodeJS.ProcessEnv = process.env): string | undefined {
+  const primaryValue = env[primary];
   if (primaryValue !== undefined && primaryValue !== "") return primaryValue;
 
   if (legacy) {
-    const legacyValue = process.env[legacy];
+    const legacyValue = env[legacy];
     if (legacyValue !== undefined && legacyValue !== "") {
       warnOnce(`env:${legacy}`, `Warning: ${legacy} is deprecated; use ${primary} instead.`);
       return legacyValue;
@@ -70,6 +70,23 @@ export function readOpenRigEnv(primary: string, legacy?: string): string | undef
   return undefined;
 }
 
+/** An explicit daemon URL with trailing slashes removed, so `http://h:7433/` addresses the
+ *  same daemon as `http://h:7433`. A URL with a query or fragment is returned exactly as
+ *  written, and so is a value that would strip down to no host (such as `/`), which stays
+ *  an explicit (if invalid) target. */
+export function normalizeOpenRigUrl(raw: string): string {
+  if (/[?#]/.test(raw)) return raw;
+  const stripped = raw.replace(/\/+$/, "");
+  return /:\/\/[^/]/.test(stripped) ? stripped : raw;
+}
+
+/** OPENRIG_URL (legacy RIGGED_URL), normalized by normalizeOpenRigUrl. `env` defaults to
+ *  process.env; callers with an injected environment pass theirs. */
+export function readOpenRigUrl(env: NodeJS.ProcessEnv = process.env): string | undefined {
+  const raw = readOpenRigEnv("OPENRIG_URL", "RIGGED_URL", env);
+  return raw === undefined ? undefined : normalizeOpenRigUrl(raw);
+}
+
 export function getPreferredOpenRigHome(): string {
   const openrigHome = getOpenRigHome();
   const legacyRiggedHome = getLegacyRiggedHome();

packages/cli/src/shared-tui.ts

@@ -1,13 +1,14 @@
-import { spawn } from "node:child_process";
 import { DaemonClient } from "./client.js";
+import { assertLocalTmux, attachTmux, readTmuxServerArgs, type TmuxAttachment } from "./tmux-attachment.js";
 
 interface SharedTuiClient {
   readonly baseUrl?: string;
+  isLocalInstance(): Promise<boolean>;
   get<T>(path: string): Promise<{ status: number; data: T }>;
 }
 
 /** Resolve the terminal on the selected instance, not a remembered tmux name. */
-export async function sharedTuiTarget(client: SharedTuiClient = new DaemonClient()): Promise<string> {
+export async function sharedTuiTarget(client: SharedTuiClient = new DaemonClient()): Promise<TmuxAttachment> {
   if (client.baseUrl && !["localhost", "127.0.0.1", "[::1]"].includes(new URL(client.baseUrl).hostname)) {
     throw new Error("Shared attachment uses local tmux. Run rig tui --shared on the selected daemon's machine; standalone rig tui can use this remote connection.");
   }
@@ -20,23 +21,18 @@ export async function sharedTuiTarget(client: SharedTuiClient = new DaemonClient
   const nodes = await client.get<Array<{
     logicalId: string; runtime: string | null;
     canonicalSessionName: string | null; tmuxAttachCommand: string | null;
+    tmuxServerArgs?: string[];
+    tmuxServerMode?: "inherit" | "isolated";
   }>>(`/api/rigs/${encodeURIComponent(kernel.rigId)}/nodes`);
   if (nodes.status >= 400) throw new Error(`Cannot read kernel terminals (HTTP ${nodes.status}); run rig status.`);
   const terminal = nodes.data.find((n) => n.logicalId === "operator.human" && n.runtime === "terminal");
   if (!terminal?.canonicalSessionName || !terminal.tmuxAttachCommand) {
     throw new Error("The kernel has no bound shared terminal. Inspect rig ps --nodes --rig kernel; restore the existing terminal rather than launching another kernel. Standalone: rig tui.");
   }
-  return terminal.canonicalSessionName;
+  await assertLocalTmux(client, terminal.tmuxServerMode, "Shared attachment uses local tmux. The selected daemon's instance identity could not be confirmed locally; run rig tui --shared on that daemon's machine. Standalone rig tui can use this remote connection.");
+  return { session: terminal.canonicalSessionName, serverArgs: readTmuxServerArgs(terminal.tmuxServerArgs) };
 }
 
-export async function attachSharedTui(target: string): Promise<number> {
-  // Attach a new client even inside tmux. switch-client would retarget somebody
-  // else's client when invoked from an agent's child shell.
-  return new Promise((resolve, reject) => {
-    const env = { ...process.env };
-    delete env.TMUX;
-    const child = spawn("tmux", ["attach-session", "-t", `=${target}`], { stdio: "inherit", env });
-    child.once("error", reject);
-    child.once("exit", (code) => resolve(code ?? 1));
-  });
+export async function attachSharedTui(target: TmuxAttachment): Promise<number> {
+  return attachTmux(target);
 }

packages/cli/src/terminal-window.ts

@@ -1,22 +1,24 @@
 import { execFile, spawn } from "node:child_process";
 import { existsSync } from "node:fs";
-import { homedir, hostname, networkInterfaces } from "node:os";
+import { homedir } from "node:os";
 import path from "node:path";
 import { randomUUID } from "node:crypto";
 import { promisify } from "node:util";
-import type { DaemonClient } from "./client.js";
+import { terminalAuthHeaders, type DaemonClient } from "./client.js";
 import { shellQuote } from "./cross-host-executor.js";
 import { prepareHerdrLaunchConfig } from "./herdr-launch-config.js";
+import { assertLocalTmux, isLocalTmuxHost, readTmuxServerArgs } from "./tmux-attachment.js";
 import { openRexView } from "./terminal-window-rex.js";
-import type { OpenViewResult } from "./commands/terminal.js";
+import type { FollowControls, OpenViewResult } from "./commands/terminal.js";
 
 interface Pane { seat: string; label: string; paneCommand: string; recoveryCommand?: string }
 interface HerdrTab { workspace_id: string; tab_id: string; label: string }
 interface Preview {
   planId: string;
   session?: string;
+  tmux?: { mode: "inherit" | "isolated" };
   status: { launch?: { socketPath: string; session?: string } };
-  composed: { id: string; opened: Pane[]; pages: Pane[][]; columns?: number; kernelLayout?: string; spaceLabel?: string; absent: OpenViewResult["absent"]; degraded: OpenViewResult["degraded"] };
+  composed: { id: string; localTmuxServer?: readonly string[]; opened: Pane[]; pages: Pane[][]; columns?: number; kernelLayout?: string; spaceLabel?: string; absent: OpenViewResult["absent"]; degraded: OpenViewResult["degraded"] };
 }
 
 export interface WindowDeps {
@@ -82,13 +84,6 @@ export function defaultWindowDeps(): WindowDeps {
   };
 }
 
-function localDaemon(url: string): boolean {
-  const host = new URL(url).hostname.replace(/^\[|\]$/g, "");
-  return ["localhost", "::1", "::", "0.0.0.0", hostname()].includes(host)
-    || /^127\./.test(host)
-    || Object.values(networkInterfaces()).flat().some(iface => iface?.address === host);
-}
-
 /** A tmux started inside a Herdr pane sets TERM_PROGRAM=tmux but inherits Herdr's pane markers. */
 function inHerdrPane(env: NodeJS.ProcessEnv): boolean {
   return env["HERDR_ENV"] === "1" && !!env["HERDR_PANE_ID"] && !!env["HERDR_SOCKET_PATH"];
@@ -210,7 +205,7 @@ function failure(provider: string, error: string): OpenViewResult {
 }
 
 /** Saved Herdr labels survive reboot; only current clients prove a local attachment. */
-async function hasLiveAttachments(existing: HerdrTab, tabs: HerdrTab[], pages: Pane[][], pageLabel: (index: number) => string, herdrArgs: string[], deps: WindowDeps): Promise<boolean> {
+async function hasLiveAttachments(existing: HerdrTab, tabs: HerdrTab[], pages: Pane[][], pageLabel: (index: number) => string, herdrArgs: string[], deps: WindowDeps, serverArgs: readonly string[]): Promise<boolean> {
   try {
     const tmux = (await deps.exec("/bin/sh", ["-c", "command -v tmux"])).trim();
     if (!path.isAbsolute(tmux)) return false;
@@ -220,13 +215,15 @@ async function hasLiveAttachments(existing: HerdrTab, tabs: HerdrTab[], pages: P
     const clients = new Map<string, string[]>();
     // Match view-composer's serialization (these strings are compared, never executed).
     const quote = (value: string) => "'" + value.replace(/'/g, "'\"'\"'") + "'";
-    const inventory = await deps.exec(tmux, ["list-clients", "-F", "#{client_pid}\t#{session_name}\t#{client_readonly}"]);
+    const server = serverArgs.map((arg, index) => index % 2 ? quote(arg) : arg).join(" ");
+    const inventory = await deps.exec(tmux, [...serverArgs, "list-clients", "-F", "#{client_pid}\t#{session_name}\t#{client_readonly}"]);
     for (const line of inventory.trim().split("\n")) {
       const [pid, session, readOnly] = line.split("\t");
       if (!pid || !/^\d+$/.test(pid) || !session || !["0", "1"].includes(readOnly ?? "")) continue;
       // Compare the composer's exact command, including session aliases and read-only mode.
       // SSH/custom commands cannot be confirmed from this local client inventory.
-      clients.set(pid, ["tmux", quote(tmux)].map(binary => `${binary} attach ${readOnly === "1" ? "-r " : ""}-t ${quote(session)}`));
+      const flags = readOnly === "1" ? ["-r "] : ["", "-f ignore-size "];
+      clients.set(pid, ["tmux", quote(tmux)].flatMap(binary => flags.map(flag => `${binary}${server ? ` ${server}` : ""} attach ${flag}-t ${quote(session)}`)));
     }
     const used = new Set<string>();
     for (const [index, page] of pages.entries()) {
@@ -259,12 +256,12 @@ async function hasLiveAttachments(existing: HerdrTab, tabs: HerdrTab[], pages: P
 }
 
 /** Render the daemon's existing composition; never rediscover/relaunch kernel seats here. */
-export async function openTerminalWindow(client: DaemonClient, view: string, requestedProvider?: string, deps = defaultWindowDeps(), expectedPlan?: string, requestedSession?: string): Promise<OpenViewResult> {
+export async function openTerminalWindow(client: DaemonClient, view: string, requestedProvider?: string, deps = defaultWindowDeps(), expectedPlan?: string, requestedSession?: string, following?: FollowControls): Promise<OpenViewResult> {
   const session = requestedSession?.trim();
-  if (session === undefined && (requestedProvider === "rex" || (requestedProvider === undefined && inRex(deps.env)))) {
+  if (!following && session === undefined && (requestedProvider === "rex" || (requestedProvider === undefined && inRex(deps.env)))) {
     return openRexView(client, view, deps, expectedPlan);
   }
-  const selectedProvider = requestedProvider ?? (session !== undefined ? "herdr" : undefined);
+  const selectedProvider = requestedProvider ?? (session !== undefined || following ? "herdr" : undefined);
   const sessionFlag = session === undefined ? "" : ` --session ${shellQuote(session)}`;
   const sessionQuery = session === undefined ? "" : `&session=${encodeURIComponent(session)}`;
   let provider = selectedProvider ?? "herdr";
@@ -281,7 +278,8 @@ export async function openTerminalWindow(client: DaemonClient, view: string, req
         : `No terminal window was opened. ${reason} On the daemon's desktop, run: ${recovery}`), windowAttempted, ...(session !== undefined ? { session } : {}) });
   const windowNotes: string[] = [];
   try {
-    if (!localDaemon(client.baseUrl)) throw new Error("The window launcher must run on the daemon's own desktop; the configured daemon is remote.");
+    if (!isLocalTmuxHost(client.baseUrl)) throw new Error("The window launcher must run on the daemon's own desktop; the configured daemon is remote.");
+    if (following && selectedProvider !== "herdr") throw new Error("Following requires the qualified Herdr provider.");
     if (session !== undefined && selectedProvider !== "herdr") throw new Error("--session selects a Herdr session; use --provider herdr.");
     if (selectedProvider && !["herdr", "tmux"].includes(selectedProvider)) throw new Error("--window supports herdr, rex or tmux. Use cmux without --window.");
     const launchWindow = await windowLauncher(deps, windowNotes);
@@ -303,19 +301,25 @@ export async function openTerminalWindow(client: DaemonClient, view: string, req
     }
     const launchTarget = preview.data.status.launch;
     let { composed, planId } = preview.data;
+    await assertLocalTmux(client, preview.data.tmux?.mode, "The window launcher must run on the daemon's own desktop; its instance identity could not be confirmed locally.");
+    const tmuxServer = readTmuxServerArgs(composed.localTmuxServer);
     if (expectedPlan !== undefined && expectedPlan !== planId) {
       return failure(provider, "The terminal view changed since preview. Refresh the preview before opening.");
     }
-    if (!composed.opened.length) return { ...failed("No conversations are attachable."), absent: composed.absent, degraded: composed.degraded };
+    if (!composed.opened.length && !following) return { ...failed("No conversations are attachable."), absent: composed.absent, degraded: composed.degraded };
     const measureWindow = async () => {
       if (window && "columns" in window && window.columns !== viewportColumns && composed.kernelLayout) {
         viewportColumns = window.columns;
         const next = await client.get<Preview | OpenViewResult>(previewUrl());
         if (next.status >= 400 || !("composed" in next.data)) throw new Error("Could not compose the measured terminal view.");
+        await assertLocalTmux(client, next.data.tmux?.mode, "The window launcher must run on the daemon's own desktop; its instance identity could not be confirmed locally.");
         if (session !== undefined && (next.data.session !== session || next.data.status.launch?.socketPath !== launchTarget?.socketPath)) {
           throw new Error("The selected Herdr endpoint changed while opening the window. Inspect it before retrying.");
         }
         if (expectedPlan !== undefined && expectedPlan !== next.data.planId) throw new Error("The measured terminal layout differs from the expected preview; refresh the preview before opening.");
+        if (JSON.stringify(readTmuxServerArgs(next.data.composed.localTmuxServer)) !== JSON.stringify(tmuxServer)) {
+          throw new Error("The daemon's tmux server changed while opening the window; refresh the preview before retrying.");
+        }
         ({ composed, planId } = next.data);
         if (!composed.opened.length) throw new Error("No conversations are attachable after opening the terminal.");
       }
@@ -358,9 +362,9 @@ export async function openTerminalWindow(client: DaemonClient, view: string, req
     }
     const directCommand = herdr && socket
       ? `${configPrefix}env -u TMUX -u HERDR_SESSION -u HERDR_SOCKET_PATH HERDR_SOCKET_PATH=${shellQuote(socket)}${configEnv} ${shellQuote(herdr)}`
-      : `env -u TMUX ${composed.opened[0]!.paneCommand.replace(/^ssh /, "ssh -t ")}`;
+      : `env -u TMUX ${composed.opened[0]!.paneCommand.replace(/^ssh (?!-t(?:\s|$))/, "ssh -t ")}`;
     manualCommand = herdr && socket ? directCommand
-      : `env -u TMUX ${(composed.opened[0]!.recoveryCommand ?? composed.opened[0]!.paneCommand).replace(/^ssh /, "ssh -t ")}`;
+      : `env -u TMUX ${(composed.opened[0]!.recoveryCommand ?? composed.opened[0]!.paneCommand).replace(/^ssh (?!-t(?:\s|$))/, "ssh -t ")}`;
     if (typeof launchWindow === "string") {
       const headless = deps.env["SSH_CONNECTION"] || deps.env["SSH_CLIENT"] || deps.env["SSH_TTY"]
         || (deps.platform === "linux" && !deps.env["DISPLAY"] && !deps.env["WAYLAND_DISPLAY"]);
@@ -392,6 +396,26 @@ export async function openTerminalWindow(client: DaemonClient, view: string, req
       // Inventory is read-only: a failed read must not prevent the first open.
       // Scope each tab list to its workspace, on the daemon's exact endpoint.
       const herdrArgs = ["-u", "TMUX", "-u", "HERDR_SESSION", "-u", "HERDR_SOCKET_PATH", `HERDR_SOCKET_PATH=${endpoint.socketPath}`, herdr];
+      if (following) {
+        // Follow has a stable journaled workspace. A changed membership plan
+        // must never enter the one-shot plan-label/new-workspace path below.
+        const answer = await client.post<OpenViewResult>("/api/terminal/follow", { view, provider: "herdr", ...following,
+          ...(session !== undefined ? { session } : {}), ...(expectedPlan !== undefined ? { expectedPlan } : {}) },
+          { timeoutMs: 45_000, headers: terminalAuthHeaders() });
+        if (!answer.data.ok || !answer.data.workspaceId) return { ...answer.data, window, notes: [...(answer.data.notes ?? []), ...windowNotes] };
+        const selectedArgs = herdrArgs;
+        const selectedStatus = JSON.parse(await deps.exec("/usr/bin/env", [...selectedArgs, "status", "server", "--json"]));
+        const selectedBody = selectedStatus?.result ?? selectedStatus;
+        if (selectedStatus.error || selectedBody?.running !== true || selectedBody.socket !== endpoint.socketPath) {
+          throw new Error("The Herdr CLI did not confirm the daemon's exact following endpoint; no tab lookup or focus was requested.");
+        }
+        const listed = JSON.parse(await deps.exec("/usr/bin/env", [...selectedArgs, "tab", "list", "--workspace", answer.data.workspaceId]));
+        const tab = (listed?.result?.tabs as HerdrTab[] | undefined)?.find(t => t.workspace_id === answer.data.workspaceId);
+        if (!tab) throw new Error("The stable following workspace has no confirmed tab. Inspect it before retrying.");
+        const focused = JSON.parse(await deps.exec("/usr/bin/env", [...selectedArgs, "tab", "focus", tab.tab_id]));
+        if (focused?.result?.tab?.tab_id !== tab.tab_id || focused.result.tab.workspace_id !== tab.workspace_id) throw new Error("Following tab selection was not confirmed.");
+        return { ...answer.data, window, notes: [...(answer.data.notes ?? []), ...windowNotes] };
+      }
       const closeCommand = (id: string) => `${session === undefined ? "herdr" : `env -u HERDR_SESSION HERDR_SOCKET_PATH=${shellQuote(endpoint.socketPath)} ${shellQuote(herdr)}`} workspace close ${shellQuote(id)}`;
       let tabs: HerdrTab[] = [];
       const spaceLabels = new Map<string, string>();
@@ -445,7 +469,7 @@ export async function openTerminalWindow(client: DaemonClient, view: string, req
       }
       let existing: HerdrTab | undefined;
       for (const candidate of candidates) {
-        if (await hasLiveAttachments(candidate, tabs, composed.pages, pageLabel(candidate), herdrArgs, deps)) { existing = candidate; break; }
+        if (await hasLiveAttachments(candidate, tabs, composed.pages, pageLabel(candidate), herdrArgs, deps, tmuxServer)) { existing = candidate; break; }
         windowNotes.push(`Could not confirm live attachments in workspace ${candidate.workspace_id}. Its contents were kept. After inspecting it, close it if no longer needed: ${closeCommand(candidate.workspace_id)}`);
       }
       if (existing) {
@@ -472,14 +496,17 @@ export async function openTerminalWindow(client: DaemonClient, view: string, req
     if (typeof launchWindow !== "function") throw new Error("No terminal window launcher is available.");
     const tmux = (await deps.exec("/bin/sh", ["-c", "command -v tmux"]).catch(() => "")).trim();
     if (!tmux) throw new Error("tmux is unavailable; run rig setup first.");
+    const runTmux = (args: string[]) => deps.exec(tmux, [...tmuxServer, ...args]);
+    const serverWords = tmuxServer.map((arg, index) => index % 2 ? shellQuote(arg) : arg).join(" ");
+    const tmuxCommand = `${shellQuote(tmux)}${serverWords ? ` ${serverWords}` : ""}`;
     viewer = `openrig-view-${deps.id()}`;
     let placeholder: string | undefined;
     if (composed.kernelLayout) {
       // This is our fresh, empty viewer, never a source seat. Attach it first so Terminal can
       // report the real new window width before we select the OpenRig view layout.
-      placeholder = (await deps.exec(tmux, ["new-session", "-d", "-s", viewer, "-n", "view-1", "-P", "-F", "#{pane_id}"])).trim();
+      placeholder = (await runTmux(["new-session", "-d", "-s", viewer, "-n", "view-1", "-P", "-F", "#{pane_id}"])).trim();
       windowAttempted = true;
-      window = await launchWindow(`env -u TMUX ${shellQuote(tmux)} attach-session -t ${shellQuote(`=${viewer}`)}`);
+      window = await launchWindow(`env -u TMUX ${tmuxCommand} attach-session -t ${shellQuote(`=${viewer}`)}`);
       await measureWindow();
     }
     for (const [pageIndex, page] of composed.pages.entries()) {
@@ -489,27 +516,27 @@ export async function openTerminalWindow(client: DaemonClient, view: string, req
       const args = pageIndex === 0 ? ["new-session", "-d", "-s", viewer, "-n", name] : ["new-window", "-d", "-t", `${viewer}:`, "-n", name];
       let pane: string;
       if (pageIndex === 0 && placeholder) {
-        await deps.exec(tmux, ["respawn-pane", "-k", "-t", placeholder, `env -u TMUX ${first.paneCommand}`]);
+        await runTmux(["respawn-pane", "-k", "-t", placeholder, `env -u TMUX ${first.paneCommand}`]);
         pane = placeholder;
-      } else pane = (await deps.exec(tmux, [...args, "-P", "-F", "#{pane_id}", `env -u TMUX ${first.paneCommand}`])).trim();
+      } else pane = (await runTmux([...args, "-P", "-F", "#{pane_id}", `env -u TMUX ${first.paneCommand}`])).trim();
       // Nested source clients on this server can make a new window too narrow to split.
       // Use the viewer's configured detached size, then restore normal client resizing.
-      const [width, height] = (await deps.exec(tmux, ["show-options", "-Av", "-t", viewer, "default-size"])).trim().split("x");
-      const sizing = (await deps.exec(tmux, ["show-options", "-Awv", "-t", pane, "window-size"])).trim();
-      await deps.exec(tmux, ["resize-window", "-t", pane, "-x", width!, "-y", height!]);
-      await deps.exec(tmux, ["select-pane", "-t", pane, "-T", first.label]);
+      const [width, height] = (await runTmux(["show-options", "-Av", "-t", viewer, "default-size"])).trim().split("x");
+      const sizing = (await runTmux(["show-options", "-Awv", "-t", pane, "window-size"])).trim();
+      await runTmux(["resize-window", "-t", pane, "-x", width!, "-y", height!]);
+      await runTmux(["select-pane", "-t", pane, "-T", first.label]);
       for (const item of page.slice(1)) {
-        pane = (await deps.exec(tmux, ["split-window", "-d", "-h", "-t", pane, "-P", "-F", "#{pane_id}", `env -u TMUX ${item.paneCommand}`])).trim();
-        await deps.exec(tmux, ["select-pane", "-t", pane, "-T", item.label]);
-        await deps.exec(tmux, ["select-layout", "-t", `${viewer}:${name}`, "even-horizontal"]);
+        pane = (await runTmux(["split-window", "-d", "-h", "-t", pane, "-P", "-F", "#{pane_id}", `env -u TMUX ${item.paneCommand}`])).trim();
+        await runTmux(["select-pane", "-t", pane, "-T", item.label]);
+        await runTmux(["select-layout", "-t", `${viewer}:${name}`, "even-horizontal"]);
       }
-      await deps.exec(tmux, ["select-layout", "-t", `${viewer}:${name}`, composed.columns === page.length ? "even-horizontal" : "tiled"]);
-      await deps.exec(tmux, ["set-option", "-w", "-t", `${viewer}:${name}`, "window-size", sizing]);
+      await runTmux(["select-layout", "-t", `${viewer}:${name}`, composed.columns === page.length ? "even-horizontal" : "tiled"]);
+      await runTmux(["set-option", "-w", "-t", `${viewer}:${name}`, "window-size", sizing]);
     }
-    await deps.exec(tmux, ["select-window", "-t", `${viewer}:view-1`]);
+    await runTmux(["select-window", "-t", `${viewer}:view-1`]);
     if (!window) {
       windowAttempted = true;
-      window = await launchWindow(`env -u TMUX ${shellQuote(tmux)} attach-session -t ${shellQuote(`=${viewer}`)}`);
+      window = await launchWindow(`env -u TMUX ${tmuxCommand} attach-session -t ${shellQuote(`=${viewer}`)}`);
     }
     return { provider, ok: true, opened: composed.opened.map(pane => pane.seat), absent: composed.absent, degraded: composed.degraded, pages: composed.pages.length, window, notes: [`Viewing session: ${viewer}. Existing conversations were preserved.`, ...windowNotes, "Check the new terminal shows the intended view; window creation alone is not visual confirmation."] };
   } catch (err) {

packages/cli/src/tmux-attachment.ts

@@ -0,0 +1,94 @@
+import { spawn } from "node:child_process";
+import { readFileSync, realpathSync } from "node:fs";
+import { hostname, networkInterfaces } from "node:os";
+import path from "node:path";
+import type { DaemonClient } from "./client.js";
+
+export interface TmuxAttachment {
+  session: string;
+  serverArgs: string[];
+  readOnly?: boolean;
+}
+
+/** The daemon supplies server options as data; clients never parse a shell command. */
+export function readTmuxServerArgs(value: unknown): string[] {
+  if (value === undefined) return []; // Compatibility with a daemon predating server metadata.
+  if (!Array.isArray(value) || (value.length !== 0 && (value.length !== 2
+    || (value[0] !== "-S" && value[0] !== "-L")
+    || typeof value[1] !== "string" || !value[1] || value[1].includes("\0")))) {
+    throw new Error("The daemon returned invalid tmux server options.");
+  }
+  return [...value] as string[];
+}
+
+/** Offline cleanup pins only active isolation/migration; ordinary inherit uses tmux's environment. */
+export function readRecordedTmuxServerArgs(home: string): string[] {
+  let record: { version?: unknown; home?: unknown; mode?: unknown; socketPath?: unknown; isolatedSocketPath?: unknown; pinned?: unknown } | null = null;
+  let readError: unknown;
+  try { record = JSON.parse(readFileSync(path.join(home, "tmux-server.json"), "utf8")); }
+  catch (error) { readError = error; }
+  const pinned = record?.mode === "isolated" || record?.pinned === true;
+  let matchesHome = false;
+  try { matchesHome = record?.home === realpathSync(home); } catch { /* A missing/unreadable home is not a valid record. */ }
+  const valid = record?.version === 1 && matchesHome
+    && (record.mode === "inherit" || record.mode === "isolated")
+    && (record.pinned === undefined || typeof record.pinned === "boolean")
+    && !(record.mode === "isolated" && record.pinned === false)
+    && typeof record.socketPath === "string" && path.isAbsolute(record.socketPath) && !record.socketPath.includes("\0")
+    && (record.isolatedSocketPath === undefined || (typeof record.isolatedSocketPath === "string"
+      && path.isAbsolute(record.isolatedSocketPath) && !record.isolatedSocketPath.includes("\0")));
+  if (valid) {
+    // An explicit opt-out record is committed before the marker is removed.
+    // Older inherit records are advisory too, never a remembered cleanup socket.
+    return pinned ? readTmuxServerArgs(["-S", record!.socketPath]) : [];
+  }
+
+  let marked = false;
+  if (!pinned) {
+    try { marked = readFileSync(path.join(home, "tmux-server.initialized"), "utf8").trim() === "openrig-tmux-pinned-v1"; }
+    catch (error) {
+      if ((error as NodeJS.ErrnoException).code !== "ENOENT") {
+        throw new Error(`Cannot read the tmux server pin marker: ${error instanceof Error ? error.message : String(error)}`);
+      }
+    }
+    if (!marked) return []; // No active pin: missing/bad advisory records cannot change inheritance.
+  }
+  if (readError) {
+    if ((readError as NodeJS.ErrnoException).code === "ENOENT") {
+      throw new Error("Missing tmux server record after pinned server selection; restore it before destroying this instance's sessions.");
+    }
+    throw new Error(`Cannot read the tmux server record: ${readError instanceof Error ? readError.message : String(readError)}`);
+  }
+  throw new Error("Invalid tmux server record; inspect it before destroying this instance's sessions.");
+}
+
+/** Host locality keeps the existing window-launcher behaviour in inherit mode. */
+export function isLocalTmuxHost(url: string): boolean {
+  const host = new URL(url).hostname.replace(/^\[|\]$/g, "");
+  return ["localhost", "::1", "::", "0.0.0.0", hostname()].includes(host)
+    || /^127\./.test(host)
+    || Object.values(networkInterfaces()).flat().some(iface => iface?.address === host);
+}
+
+/** Only an effective isolated server needs proof beyond the legacy host check. */
+export async function assertLocalTmux(
+  client: Pick<DaemonClient, "isLocalInstance">,
+  mode: "inherit" | "isolated" | undefined,
+  message = "Terminal attachment uses local tmux. The selected daemon's instance identity could not be confirmed locally; run this command on that daemon's machine.",
+): Promise<void> {
+  if (mode === "isolated" && !await client.isLocalInstance()) throw new Error(message);
+}
+
+/** Attach only; creating or retargeting somebody else's tmux client is not part of this operation. */
+export function attachTmux(target: TmuxAttachment): Promise<number> {
+  const server = readTmuxServerArgs(target.serverArgs);
+  return new Promise((resolve, reject) => {
+    const env = { ...process.env };
+    delete env.TMUX;
+    delete env.TMUX_PANE;
+    const child = spawn("tmux", [...server, "attach-session", ...(target.readOnly ? ["-r"] : []), "-t", `=${target.session}`],
+      { stdio: "inherit", env, windowsHide: true });
+    child.once("error", reject);
+    child.once("exit", code => resolve(code ?? 1));
+  });
+}

packages/cli/test/acp-connector.test.ts

@@ -122,6 +122,26 @@ describe("rig acp run", () => {
     expect(q.task().notes.join("\n")).toMatch(/cancel requested at \S+; upstream inference and spend are unknown/);
   });
 
+  it("keeps waiting for cancellation when recording the cancel request fails", async () => {
+    const q = memoryQueue("scenario:slow");
+    const update = q.queue.update;
+    q.queue.update = async (id, change) => {
+      if (/^ACP run \S+: cancel requested at/.test(change.note)) throw new Error("queue unavailable");
+      return update(id, change);
+    };
+    const errors = vi.spyOn(console, "error").mockImplementation(() => {});
+    let cancel!: () => void;
+    const cancelRequested = new Promise<void>((resolve) => { cancel = resolve; });
+    const outcome = await runAcpAssignment(q.queue, options({
+      cancelRequested,
+      progress: (line) => { if (line.startsWith("step ")) cancel(); },
+    }));
+    expect(outcome.kind).toBe("cancelled");
+    expect(errors).toHaveBeenCalledWith(expect.stringContaining("queue unavailable"));
+    expect(outcomeNote(q)).toContain("outcome cancelled");
+    errors.mockRestore();
+  });
+
   it("closes an unanswered permission row as stale when the turn is cancelled, and tells the agent cancelled", async () => {
     const q = memoryQueue("scenario:permission");
     let cancel!: () => void;

packages/cli/test/attach-reply-address.test.ts

@@ -31,7 +31,10 @@ import type { TmuxAdapter } from "../../daemon/src/adapters/tmux.js";
 
 vi.mock("node:child_process", async (actual) => ({ ...await actual<typeof import("node:child_process")>(), execSync: (cmd: string) => {
   if (!cmd.startsWith("tmux display-message ")) throw new Error(`Unexpected shell command in inert fixture: ${cmd}`);
-  return `${process.env.TEST_PHYSICAL_SESSION}\n@12\n%42\n`;
+  // Render the requested format as real tmux does, which writes a control character as "_".
+  const format = /'([^']*)'\s*$/.exec(cmd)?.[1] ?? /"([^"]*)"\s*$/.exec(cmd)?.[1] ?? "";
+  return `${format.replace(/#\{session_name\}/g, process.env.TEST_PHYSICAL_SESSION ?? "").replace(/#\{window_id\}/g, "@12")
+    .replace(/#\{pane_id\}/g, "%42").replace(/[\u0000-\u001f]/g, "_")}\n`;
 } }));
 
 it.each(["0", "person-terminal"])("real attach -> returned env -> actual From -> reply to original %s pane", async (physical) => {

packages/cli/test/attach.test.ts

@@ -1,5 +1,6 @@
 import { describe, it, expect, vi, beforeAll, afterAll } from "vitest";
 import http from "node:http";
+import { execFileSync, execSync, spawnSync } from "node:child_process";
 import { Command } from "commander";
 import { attachCommand, resolveAttachContext } from "../src/commands/attach.js";
 import { DaemonClient } from "../src/client.js";
@@ -170,15 +171,8 @@ describe("Attach CLI", () => {
   });
 
   it("resolveAttachContext detects tmux session metadata when TMUX_PANE is set", () => {
-    const savedPane = process.env["TMUX_PANE"];
-    process.env["TMUX_PANE"] = "%42";
-    const mockTmuxExec = vi.fn(() => "dev1-impl2@rigged-buildout\n@12\n%42");
-
-    const result = resolveAttachContext(mockTmuxExec);
-
-    if (savedPane === undefined) delete process.env["TMUX_PANE"];
-    else process.env["TMUX_PANE"] = savedPane;
-
+    const result = withTmuxPane("%42", () =>
+      resolveAttachContext(fakeTmux({ session: "dev1-impl2@rigged-buildout", window: "@12", pane: "%42" })));
     expect(result).toEqual({
       attachmentType: "tmux",
       tmuxSession: "dev1-impl2@rigged-buildout",
@@ -186,4 +180,72 @@ describe("Attach CLI", () => {
       tmuxPane: "%42",
     });
   });
+
+  it("includes the actual socket without trimming literal path whitespace", () => {
+    const result = withTmuxPane("%42", () => resolveAttachContext(fakeTmux({
+      pane: "%42", window: "@1", session: "team_1 rig_b @7 %9", socket: "/private/socket name ",
+    })));
+    expect(result).toMatchObject({ attachmentType: "tmux", tmuxSocketPath: "/private/socket name ", tmuxSession: "team_1 rig_b @7 %9" });
+  });
+
+  // Real tmux rewrote the newlines of a combined format as "_" ("unrelated-shell_@2_%2"), and the session fell back to
+  // external_cli, so replies to the self-attached member had no transport.
+  it.each([
+    ["the shape real tmux returned", "unrelated-shell"],
+    ["a name holding underscores, spaces and id-like words", "team_1 rig_b @7 %9"],
+  ])("resolveAttachContext resolves the pane from %s", (_label, session) => {
+    const result = withTmuxPane("%2", () => resolveAttachContext(fakeTmux({ session, window: "@2", pane: "%2" })));
+    expect(result).toEqual({ attachmentType: "tmux", tmuxSession: session, tmuxWindow: "@2", tmuxPane: "%2" });
+  });
+
+  it("resolveAttachContext keeps external_cli outside tmux, and when tmux can't answer", () => {
+    const notCalled = vi.fn(() => "");
+    expect(withTmuxPane(undefined, () => resolveAttachContext(notCalled))).toEqual({ attachmentType: "external_cli" });
+    expect(notCalled).not.toHaveBeenCalled();
+    for (const tmux of [vi.fn(() => { throw new Error("no server running"); }), vi.fn(() => ""), vi.fn(() => "unexpected")]) {
+      expect(withTmuxPane("%2", () => resolveAttachContext(tmux))).toEqual({ attachmentType: "external_cli" });
+    }
+  });
 });
+
+const TMUX_SOCKET = `openrig-attach-test-${process.pid}`;
+const tmuxAvailable = spawnSync("tmux", ["-V"]).status === 0;
+
+describe.runIf(tmuxAvailable)("resolveAttachContext against a real tmux server (private socket)", () => {
+  afterAll(() => { spawnSync("tmux", ["-L", TMUX_SOCKET, "kill-server"]); });
+
+  it("resolves the session, window and pane of a session whose name holds spaces and underscores", () => {
+    const session = "unrelated_shell b";
+    execFileSync("tmux", ["-L", TMUX_SOCKET, "-f", "/dev/null", "new-session", "-d", "-s", session]);
+    const read = (format: string) => execFileSync("tmux", ["-L", TMUX_SOCKET, "list-panes", "-t", `=${session}`, "-F", format],
+      { encoding: "utf8" }).trim();
+    const pane = read("#{pane_id}");
+    const window = read("#{window_id}");
+    const onSocket = (cmd: string) => execSync(cmd.replace(/^tmux /, `tmux -L ${TMUX_SOCKET} `), { encoding: "utf8" }).trim();
+    expect(withTmuxPane(pane, () => resolveAttachContext(onSocket)))
+      .toEqual({ attachmentType: "tmux", tmuxSession: session, tmuxWindow: window, tmuxPane: pane, tmuxSocketPath: read("#{socket_path}") });
+  });
+});
+
+/** A `tmux display-message -p` double that renders the requested format as real tmux does, control characters
+ *  included: tmux writes each one as "_". */
+function fakeTmux(values: { session: string; window: string; pane: string; socket?: string }) {
+  return vi.fn((cmd: string) => {
+    const format = /'([^']*)'\s*$/.exec(cmd)?.[1] ?? /"([^"]*)"\s*$/.exec(cmd)?.[1];
+    if (format === undefined) throw new Error(`unexpected tmux command: ${cmd}`);
+    return format.replace(/#\{session_name\}/g, values.session).replace(/#\{window_id\}/g, values.window)
+      .replace(/#\{pane_id\}/g, values.pane).replace(/#\{socket_path\}/g, values.socket ?? "").replace(/[\u0000-\u001f]/g, "_");
+  });
+}
+
+function withTmuxPane<T>(pane: string | undefined, run: () => T): T {
+  const saved = process.env["TMUX_PANE"];
+  if (pane === undefined) delete process.env["TMUX_PANE"];
+  else process.env["TMUX_PANE"] = pane;
+  try {
+    return run();
+  } finally {
+    if (saved === undefined) delete process.env["TMUX_PANE"];
+    else process.env["TMUX_PANE"] = saved;
+  }
+}

packages/cli/test/capture.test.ts

@@ -156,6 +156,17 @@ describe("Capture CLI", () => {
       expect(out).toEqual(["Do you want to\n"]);
       expect(err).toEqual([`[rig capture: small@my-rig: ${SMALL_PANE_HINT}]`]);
     });
+
+    it("sends 20 lines for non-positive --lines values", async () => {
+      for (const lines of ["0", "-1"]) {
+        const before = requests.length;
+        await captureLogs(async () => {
+          await makeCmd().parseAsync(["node", "rig", "capture", "tall@my-rig", `--lines=${lines}`]);
+        });
+        expect(requests.length).toBeGreaterThan(before);
+        expect(requests[requests.length - 1]!.lines).toBe(20);
+      }
+    });
   });
 
   describe("--readable reads a small seat at a readable size", () => {

packages/cli/test/chatroom-watch-tmux-args.test.ts

@@ -0,0 +1,89 @@
+import { describe, it, expect, beforeAll, afterAll, vi } from "vitest";
+import { execFileSync, spawnSync } from "node:child_process";
+import { mkdtempSync, readFileSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import path from "node:path";
+import { chatroomWatchTmuxArgs } from "../src/commands/chatroom.js";
+
+const sessionArg = (rig: string) => { const args = chatroomWatchTmuxArgs(rig); return args[args.indexOf("-s") + 1]; };
+
+describe("chatroom watch --tmux arguments", () => {
+  it("hands tmux the watcher as separate arguments, with the rig name after --, and asks for the name it used", () => {
+    expect(chatroomWatchTmuxArgs("my rig")).toEqual(["new-session", "-d", "-P", "-F", "#{session_name}", "-s", "chatroom@my rig",
+      "rig", "chatroom", "watch", "--", "my rig"]);
+  });
+
+  // tmux reads -s as a format: `##` is one `#`, and a run of `#` before `[` is a style it keeps as written.
+  it.each([
+    ["my-rig", "chatroom@my-rig"],
+    ["a#b", "chatroom@a##b"],
+    ["a##b", "chatroom@a####b"],
+    ["a#", "chatroom@a##"],
+    ["a#{b}", "chatroom@a##{b}"],
+    ["a#[fg=red]b", "chatroom@a#[fg=red]b"],
+    ["a##[b", "chatroom@a##[b"],
+  ])("encodes %j for tmux's session-name format as %j", (rig, session) => {
+    expect(sessionArg(rig)).toBe(session);
+  });
+
+  // tmux ends a command at an argument ending in `;` unless a `\` comes before it.
+  it.each([
+    ["a;", "chatroom@a\\;", "a\\;"],
+    [";", "chatroom@\\;", "\\;"],
+    ["a\\;", "chatroom@a\\\\;", "a\\\\;"],
+    ["a;b", "chatroom@a;b", "a;b"],
+  ])("puts a \\ before a final ; in %j", (rig, session, watcherArg) => {
+    const args = chatroomWatchTmuxArgs(rig);
+    expect(sessionArg(rig)).toBe(session);
+    expect(args[args.length - 1]).toBe(watcherArg);
+  });
+});
+
+const TMUX_SOCKET = `openrig-chatroom-test-${process.pid}`;
+const tmuxAvailable = spawnSync("tmux", ["-V"]).status === 0;
+const names = ["my-rig", "my rig", "it's", 'say "hi"', "$(echo pwned)", "`id`", "a;b", "-rig",
+  "a#b", "a##b", "a#{b}", "a#[fg=red]b", "a##[b", "a;", ";"];
+
+describe.runIf(tmuxAvailable)("chatroom watch --tmux against a real tmux server (private socket)", () => {
+  let dir = "";
+  let recorder = "";
+  beforeAll(() => {
+    dir = mkdtempSync(path.join(tmpdir(), "openrig-chatroom-watch-"));
+    // Stands in for `rig`: records the arguments the watcher gets, then stays up so the session lives.
+    recorder = path.join(dir, "record.cjs");
+    writeFileSync(recorder, "require('node:fs').writeFileSync(process.argv[2], JSON.stringify(process.argv.slice(3)));\nsetTimeout(() => {}, 30000);\n");
+  });
+  afterAll(() => {
+    spawnSync("tmux", ["-L", TMUX_SOCKET, "kill-server"]);
+    rmSync(dir, { recursive: true, force: true });
+  });
+
+  // The exact arguments, with only the `rig` program swapped for the recorder; tmux prints the session it made.
+  function start(rig: string, out: string): string {
+    const args = chatroomWatchTmuxArgs(rig);
+    const program = args.indexOf("-s") + 2;
+    return execFileSync("tmux", ["-L", TMUX_SOCKET, "-f", "/dev/null", ...args.slice(0, program), process.execPath, recorder, out,
+      ...args.slice(program + 1)], { encoding: "utf8" }).trim();
+  }
+
+  it.each(names.map((rig, i) => [rig, i] as const))("keeps the session chatroom@%s and hands the watcher the name as one argument", async (rig, i) => {
+    const out = path.join(dir, `argv-${i}.json`);
+    expect(start(rig, out)).toBe(`chatroom@${rig}`);
+    await vi.waitFor(() => expect(JSON.parse(readFileSync(out, "utf8"))).toEqual(["chatroom", "watch", "--", rig]), { timeout: 5000 });
+  });
+
+  // tmux versions differ in how they normalize dots and literal dollars in session names.
+  // The reported name must address the real session, while the watcher still gets the original rig name.
+  it.each([
+    ["release-0.7", ["[email protected]", "chatroom@release-0_7"]],
+    ["a$HOME", ["chatroom@a$HOME", "chatroom@a\\$HOME"]],
+  ] as const)("reports the name tmux used for %s", async (rig, expected) => {
+    const out = path.join(dir, `argv-normalized-${rig}.json`);
+    const reported = start(rig, out);
+    expect(expected).toContain(reported);
+    expect(execFileSync("tmux", ["-L", TMUX_SOCKET, "list-sessions", "-F", "#{session_name}"], { encoding: "utf8" }).split("\n"))
+      .toContain(reported);
+    expect(spawnSync("tmux", ["-L", TMUX_SOCKET, "has-session", "-t", `=${reported}:`]).status).toBe(0);
+    await vi.waitFor(() => expect(JSON.parse(readFileSync(out, "utf8"))).toEqual(["chatroom", "watch", "--", rig]), { timeout: 5000 });
+  });
+});

packages/cli/test/config-store-extended.test.ts

@@ -44,7 +44,7 @@ function clearEnv(): () => void {
     // its custom home into default-path resolution; UNSETTING it would fall through
     // to the REAL ~/.openrig (foreign on-box state) — so tests set a fresh tmp home.
     "OPENRIG_HOME", "RIGGED_HOME",
-    "OPENRIG_PORT", "OPENRIG_HOST", "OPENRIG_DB", "OPENRIG_DAEMON_DB_CONFLICT",
+    "OPENRIG_PORT", "OPENRIG_HOST", "OPENRIG_DB", "OPENRIG_TMUX_SERVER", "OPENRIG_DAEMON_DB_CONFLICT",
     "OPENRIG_TRANSCRIPTS_ENABLED", "OPENRIG_TRANSCRIPTS_PATH",
     "OPENRIG_TRANSCRIPTS_LINES", "OPENRIG_TRANSCRIPTS_POLL_INTERVAL_SECONDS",
     "OPENRIG_WORKSPACE_ROOT", "OPENRIG_WORKSPACE_SLICES_ROOT",
@@ -109,6 +109,16 @@ describe("ConfigStore — extended namespaces (User Settings v0)", () => {
     restoreEnv();
   });
 
+  it("resolves tmux isolation from the shared config and environment with a strict mode", () => {
+    const store = new ConfigStore(configPath);
+    expect(store.resolve().terminal.tmuxServer).toBe("inherit");
+    store.set("terminal.tmux_server", "isolated");
+    expect(store.resolve().terminal.tmuxServer).toBe("isolated");
+    process.env.OPENRIG_TMUX_SERVER = "inherit";
+    expect(store.resolve().terminal.tmuxServer).toBe("inherit");
+    expect(() => store.set("terminal.tmux_server", "isolate")).toThrow('must be "inherit" or "isolated"');
+  });
+
   it("VALID_KEYS includes legacy, user-settings, ui.preview, recovery, agents, feed, and transcript-rotation keys", () => {
     const expected = [
       "daemon.port", "daemon.host", "daemon.db_conflict",
@@ -174,6 +184,7 @@ describe("ConfigStore — extended namespaces (User Settings v0)", () => {
       "snapshots.periodic.retention_keep",
       // OPR.0.4.6.02 S1 — inner-tmux status-bar launch default (static bool).
       "terminal.status_bar",
+      "terminal.tmux_server",
       "terminal.history_limit",
       "terminal.inherit_tmux_environment",
       // Slice 15 (OPR.0.7.0.15) — compact command-output defaults switch.

packages/cli/test/cross-host-commands.test.ts

@@ -285,6 +285,32 @@ describe("capture --host (cross-host short-circuit)", () => {
     expect(captureCalls.argv).toEqual(["rig", "capture", "dev-impl@my-rig", "--lines", "5x"]);
   });
 
+  it("forwards a fractional --lines as given: a positive parseInt means the remote decides", async () => {
+    const captureCalls: { argv?: readonly string[] } = {};
+    const cmd = captureCommand(deps({
+      run: async (_h, argv) => {
+        captureCalls.argv = argv;
+        return { ok: true, failedStep: "none", stdout: "", stderr: "", remoteExitCode: 0 };
+      },
+    }));
+    await cmd.parseAsync(["--host", "vm-a", "dev-impl@my-rig", "--lines=3.5"], { from: "user" });
+    expect(captureCalls.argv).toEqual(["rig", "capture", "dev-impl@my-rig", "--lines", "3.5"]);
+  });
+
+  it("forwards a non-positive or non-numeric --lines as the documented default of 20", async () => {
+    for (const value of ["0", "-1", "abc"]) {
+      const captureCalls: { argv?: readonly string[] } = {};
+      const cmd = captureCommand(deps({
+        run: async (_h, argv) => {
+          captureCalls.argv = argv;
+          return { ok: true, failedStep: "none", stdout: "", stderr: "", remoteExitCode: 0 };
+        },
+      }));
+      await cmd.parseAsync(["--host", "vm-a", "dev-impl@my-rig", `--lines=${value}`], { from: "user" });
+      expect(captureCalls.argv).toEqual(["rig", "capture", "dev-impl@my-rig", "--lines", "20"]);
+    }
+  });
+
   it("forwards --history-plus-pane with the --lines it applies to", async () => {
     const captureCalls: { argv?: readonly string[] } = {};
     const cmd = captureCommand(deps({