docs(daemon): scope the draft-aware gap comments to the tested schedule

Two code-comment precisions on billbenetti's paste-side mitigation
(#3097), as its reviewers asked:
- seat-delivery-guard.ts: what happens after typing lands between the
  final check and the paste is stated for the tested append-mode
  schedule only, not as a guarantee.
- tmux.ts: the guard's checks run last, rather than its read being the
  last wait before the paste.

In the tested append-mode schedule, Enter is refused and nothing is submitted or deleted; this is no guarantee for every interleaving, and #3076 stays open.

This narrows #3097's merge message, which stated that outcome without the limit.

CHANGELOG.md

@@ -8,187 +8,6 @@ deprecations, and behavioral changes. Breaking changes are called out explicitly
 
 ---
 
-## [0.7.14]
-
-OpenRig 0.7.14 is a patch for steadier circulation on a busy team. A burst of node
-reads now takes turns, so a health check, a queue read or a send waits for one
-step of it instead of the whole burst, and bounded queue listings over a long
-retained history use indexes instead of scanning it. An assistant connected to
-your team can reach a seat by the name its team shows, connected assistants
-appear in `rig ps` and the TUI, and setup guidance tells the agent to fit a
-first team to your logins. There are fixes across delivery, seats and the TUI,
-many contributed by people using OpenRig. See Known limits for what this
-release doesn't fix.
-
-### Before you upgrade
-
-- **Stop the daemon, back up the database, install 0.7.14, then start the
-  daemon.** Your agents keep running through a daemon restart; don't use
-  `rig down`, which ends them.
-
-  ```sh
-  rig daemon stop
-  cp "$(rig config get db.path)" "$(rig config get db.path).before-0.7.14"
-  npm install -g @openrig/cli@0.7.14   # or your existing installation method
-  rig daemon start
-  ```
-
-- **Three migrations run before the daemon listens.** Migration 163 adds the
-  `network_installations` table and four columns to `network_bindings`.
-  Migrations 164 and 166 build indexes. The author's measurement of 166 took
-  3.47 s on 28,976 rows, not a real installation's size, so a large database can
-  take longer. The write overhead, and going back to 0.7.13 after these three
-  migrations, aren't established; keep the backup. ([#3007](https://github.com/mvschwarz/openrig/pull/3007), [#3008](https://github.com/mvschwarz/openrig/pull/3008), [#3090](https://github.com/mvschwarz/openrig/pull/3090))
-- **If the start is reported pending:** the CLI waits 20 s, then may print
-  "Daemon startup is pending after the readiness deadline". Readiness remains
-  unconfirmed. It leaves the child running and the startup reservation
-  retained, without recording the daemon. Inspect the address and log it
-  prints. The child may later answer, and status can then report it running,
-  but that does not finish this start's recording or release its reservation.
-  Don't start it again until the steps below are done. The message's restart
-  steps stop that child and clear the reservation: stop the named child PID and
-  wait for it to exit, inspect `daemon-start.lock`, and only when both recorded
-  launcher and child PIDs are absent, archive the lock and run
-  `rig daemon start`. On a large database the new start can be reported pending
-  again; reaching a recorded start this way hasn't been tested.
-  `rig daemon stop` cannot stop that unpublished child.
-- **The `openrig` plugin carries the CLI's version, 0.7.14; openrig-core stays
-  0.1.29.**
-
-### Steadier under load
-
-- Node reads (`rig ps --nodes`, the TUI, several agents at once) take turns,
-  one step at a time, so other reads and sends are no longer held behind a
-  whole burst. ([#3101](https://github.com/mvschwarz/openrig/pull/3101))
-- Bounded queue listings over retained history use indexes for their ordering
-  and for live and archived owner notifications. ([#3090](https://github.com/mvschwarz/openrig/pull/3090))
-- Codex seats re-checked in one readiness poll share process listings where
-  each seat's observation order and the 2 s freshness allow; each keeps its own
-  later observation and its own fallback. ([#3061](https://github.com/mvschwarz/openrig/pull/3061))
-- New tmux servers raise their soft open-file limit to 8,192 or the hard limit
-  where they can, keeping an already higher limit. This makes descriptor
-  exhaustion less likely as panes and viewers add up; servers already running
-  keep their limits. ([#3070](https://github.com/mvschwarz/openrig/pull/3070))
-- A delegation parent gets one standing notice per child seat for unclaimed
-  rows, not one per row, and held work gets its notice back when its typing
-  hold clears. ([#3049](https://github.com/mvschwarz/openrig/pull/3049), [#3082](https://github.com/mvschwarz/openrig/pull/3082))
-
-### Your team and the people it works for
-
-- An assistant connected over MCP reaches a seat by its plain name, stored
-  under the seat's actual session; an unknown name, or one several seats match,
-  says so. ([#3088](https://github.com/mvschwarz/openrig/pull/3088))
-- Connected assistants appear, with their check-in, in `rig ps`, `rig_team` and
-  the TUI's tables, and their person sees plain tool titles. ([#3057](https://github.com/mvschwarz/openrig/pull/3057), [#3058](https://github.com/mvschwarz/openrig/pull/3058), [#3092](https://github.com/mvschwarz/openrig/pull/3092))
-- `rig doctor` and setup say a login is needed when a seat you start runs on
-  that harness. The operator is taught to check each seat's runtime against
-  your logins before launch, name any missing sign-in and let you choose.
-  ([#3077](https://github.com/mvschwarz/openrig/pull/3077))
-- When the readable person registry holds nobody, `rig daemon start` and
-  `rig gateway participants` tell the setup agent to offer registration. The
-  kernel operator is taught to include that offer in its greeting and register
-  only on your yes. ([#3041](https://github.com/mvschwarz/openrig/pull/3041), [#3071](https://github.com/mvschwarz/openrig/pull/3071))
-- A team your agent hands work to knows how to ask you a question, and your
-  answer wakes it. A click on an ask with several questions records that
-  choice; the ask stays open until the last one. ([#3059](https://github.com/mvschwarz/openrig/pull/3059), [#3045](https://github.com/mvschwarz/openrig/pull/3045))
-- A pairing request's description of itself is quoted wherever it's shown. ([#3067](https://github.com/mvschwarz/openrig/pull/3067))
-- `rig queue list` takes exact `--tag`, `--mission` and `--slice` filters, and
-  several tags must all match. ([#3066](https://github.com/mvschwarz/openrig/pull/3066))
-- A skill-only edition of the `openrig` plugin, for Claude's plugin directory,
-  follows handed-on work and keeps its reply route. ([#3056](https://github.com/mvschwarz/openrig/pull/3056), [#3075](https://github.com/mvschwarz/openrig/pull/3075))
-
-### Fixes
-
-- An ordinary send is refused as needing input while an open Claude Code or
-  Codex `/model` picker is positively identified; nothing is typed. Close the
-  picker before trying to deliver the message. ([#3083](https://github.com/mvschwarz/openrig/pull/3083))
-- On a resume of a Codex seat's own conversation in a rig with pods, including
-  `rig up --existing`, OpenRig avoids adding missing guidance blocks to the
-  repository's tracked `AGENTS.md` when Codex is 0.164.0 or later, no guidance
-  file or developer instructions are configured, and the configuration and
-  block sources can be read. ([#3039](https://github.com/mvschwarz/openrig/pull/3039))
-- With `recovery.auto_drive_provider_prompts` on, Claude Code's gated-grants
-  trust check is never answered for you, and the plain trust prompt is answered
-  "Yes". ([#3073](https://github.com/mvschwarz/openrig/pull/3073))
-- Codex seats with linked `CODEX_HOME` folders or config files get hook trust
-  under the keys Codex reads. ([#3079](https://github.com/mvschwarz/openrig/pull/3079))
-- `rig terminal open --provider herdr` reuses a live Herdr workspace, and Rex
-  tiles show the seat's agent and its activity. ([#3068](https://github.com/mvschwarz/openrig/pull/3068), [#2992](https://github.com/mvschwarz/openrig/pull/2992))
-- `rig broadcast` prints a held recipient as retained, not sent. ([#3054](https://github.com/mvschwarz/openrig/pull/3054))
-- Account erasure also removes OpenRig's own wakes and notices for that
-  account's seats. ([#3089](https://github.com/mvschwarz/openrig/pull/3089))
-- A snapshot without sessions returns its `invalid_snapshot_data` blocker
-  instead of throwing a TypeError; the restore still stops there. A retried
-  no-launch import recovers the original rig by its operation ID. ([#3069](https://github.com/mvschwarz/openrig/pull/3069), [#3042](https://github.com/mvschwarz/openrig/pull/3042))
-- A rebuilt briefing limits optional superseded recap paths to the newest
-  three by default when one folder covers that history; the current recap and
-  all artifact records remain. Other layouts keep the full list, and
-  `output.rebuild_recaps` set to 0 lists all paths. ([#3055](https://github.com/mvschwarz/openrig/pull/3055))
-- `rig compact --state` names the input check that stopped a managed
-  compaction, and the TUI's graph shows a seat the daemon reads as unknown as
-  unknown. ([#3053](https://github.com/mvschwarz/openrig/pull/3053), [#3046](https://github.com/mvschwarz/openrig/pull/3046))
-- Setting a context-pressure threshold explains when an environment override
-  inverts the pair, and `rig workflow validate` warns that harness pins aren't
-  checked against seats. ([#2567](https://github.com/mvschwarz/openrig/pull/2567), [#3102](https://github.com/mvschwarz/openrig/pull/3102))
-- The config browser's entries say each setting's type, choices, range, format
-  and write route. ([#3050](https://github.com/mvschwarz/openrig/pull/3050))
-
-### Experimental
-
-- `rig network` is experimental, unsupported and at your own risk. Its connector
-  isn't inert: the daemon starts it, sweeps leftover credential folders and
-  schedules local rounds even with no binding. ([#3007](https://github.com/mvschwarz/openrig/pull/3007), [#3084](https://github.com/mvschwarz/openrig/pull/3084))
-- A grant service can hold back only new work under its fair-use limit, and an
-  app can read what needs a person, and the answers to it, by link. ([#3048](https://github.com/mvschwarz/openrig/pull/3048), [#3008](https://github.com/mvschwarz/openrig/pull/3008))
-
-### Behaviour that changed
-
-- **Node reads:** a nodes response is no longer one snapshot across its steps,
-  and in a burst each nodes response comes later: about 43 ms against 21 ms
-  median in a 24-request burst. One very large rig is still one step. ([#3101](https://github.com/mvschwarz/openrig/pull/3101))
-- **An assistant naming a seat now hears three errors:** `destination_ambiguous`
-  (several seats match; the candidates are listed), `destination_not_found` (no
-  seat it can reach has that name) and `destination_unchecked` (the team
-  couldn't be read; try again). ([#3088](https://github.com/mvschwarz/openrig/pull/3088))
-
-### Known limits
-
-- **Queue listings:** the new indexes aren't a cure for every stall. Listings of
-  active work showed no gain, one case was slower, and writes weren't measured. ([#3090](https://github.com/mvschwarz/openrig/pull/3090))
-- **Draft-aware delivery ([#3076](https://github.com/mvschwarz/openrig/issues/3076)):** an automatic OpenRig notice can still be left beside
-  a draft started during delivery. In the reproduced paste-window cases the person's draft
-  stayed intact and no Enter was sent, but OpenRig's text stayed in the input.
-  Those cases don't cover a later window. On a conditional send, one that names
-  the conversation it expects, an isolated test with a scripted terminal showed
-  characters typed while OpenRig ran its last checks before Enter submitted
-  together with OpenRig's text, and the send reported success. The same test
-  reproduces it on 0.7.13. It hasn't been tested with a real client or model.
-- **The TUI's tables** can drop a connected assistant's check-in line when its
-  rig's seat inventory can't be read; `rig roster` still prints it. ([#3105](https://github.com/mvschwarz/openrig/pull/3105) isn't in this release)
-- **Repository guidance on restore:** rigs without pods, old or unknown Codex
-  versions, configured guidance and unreadable configuration keep the existing
-  file path. Fresh, fork and handover launches are outside this resume fix.
-  Existing managed blocks may still refresh, and the seat-skills line still
-  belongs to the file writer; check `git status`. A block whose source can't be
-  read is reported missing.
-- **From 0.7.13:** [its known limits](https://github.com/mvschwarz/openrig/releases/tag/v0.7.13)
-  still apply, except where this release fixes them. The runtime fixes cover
-  adding missing guidance on the eligible Codex resumes described above
-  ([#3039](https://github.com/mvschwarz/openrig/pull/3039)), recognized model pickers ([#3083](https://github.com/mvschwarz/openrig/pull/3083)) and the unquoted
-  pairing description ([#3067](https://github.com/mvschwarz/openrig/pull/3067)).
-  This release also changes the operator-first registration ([#3041](https://github.com/mvschwarz/openrig/pull/3041),
-  [#3071](https://github.com/mvschwarz/openrig/pull/3071)) and pre-launch login guidance ([#3077](https://github.com/mvschwarz/openrig/pull/3077)); their observed
-  outcomes aren't established yet. They weren't re-checked for this release
-  unless named here.
-
-### Credits
-
-Thanks to @Kayhusk, @billbenetti, @cavalcanticampos, @Aummadour, @Tankard,
-@w4rc0n and @Totopo27 for merged contributions. Thanks also for the reports
-behind these fixes: @Kayhusk ([#3035](https://github.com/mvschwarz/openrig/issues/3035) and [#3078](https://github.com/mvschwarz/openrig/issues/3078)), @Tankard ([#2583](https://github.com/mvschwarz/openrig/issues/2583)),
-@dajiaohuang ([#2529](https://github.com/mvschwarz/openrig/issues/2529)), @billbenetti ([#2728](https://github.com/mvschwarz/openrig/issues/2728)), @emilio-gagliardi ([#3087](https://github.com/mvschwarz/openrig/issues/3087)), @korallis
-([#161](https://github.com/mvschwarz/openrig/issues/161)) and @w4rc0n ([#3017](https://github.com/mvschwarz/openrig/issues/3017)).
-
 ## [0.7.13]
 
 OpenRig 0.7.13 is a patch for people running their own agent teams. When you answer a runner's permission request, you now get a

README.md

@@ -26,9 +26,9 @@ use:
 
 ```sh
 # Claude Code
-npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.14/skills/rigs --skill rigs --agent claude-code
+npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.12/skills/rigs --skill rigs --agent claude-code
 # Codex
-npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.14/skills/rigs --skill rigs --agent codex
+npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.12/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 `v0.7.14` 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 `v0.7.12` 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/v0.7.14/scripts/install.sh | sh -s -- --dry-run
-curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.14/scripts/install.sh | sh
+curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.12/scripts/install.sh | sh -s -- --dry-run
+curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.12/scripts/install.sh | sh
 ```
 
 If the last step isn't ready, the installer reports `INSTALLED, SETUP NOT COMPLETE` and points to the details from `rig setup` above. Follow that output and any reported Fix, then rerun `rig setup`; the CLI does not need reinstalling. If the only remaining failures are provider sign-ins, sign in to each selected provider as below and continue.

docs/reference/daemon-http-clients.md

@@ -182,9 +182,8 @@ With `needs-me.read` supported:
   about the open rows blocked on it (at most 100; empty when none are recorded).
 - **A snapshot watermark:** the response's `watermark` is the newest event `seq` read before the snapshot. Follow
   `GET /api/events` from it with `Last-Event-ID: <watermark>`: every event after the snapshot is replayed, some possibly
-  already reflected in it. A click that records or changes an answer emits `queue.answer_recorded` even while the set
-  is partial. An identical repeated choice, or a later click after a pending ask's answer set is complete, records
-  nothing new and emits none. Events are best effort; re-read the snapshot or row after a stream gap.
+  already reflected in it. Recording one answer of several emits no event yet, so re-read the snapshot (or the row)
+  rather than relying on events alone for partial answers.
 - **The answers returned to a request:** `GET /api/queue/<qitemId>` adds `answers`, oldest first, each
   `{ qitemId, by, at, text, episode, recordedIdentity }`. They are found by the answer row's link to the request, not
   by its text, so don't parse the `In reply to:` line in an answer's body. `text` is the person's answer without that

docs/reference/getting-started.md

@@ -185,10 +185,10 @@ agent doing the setup. Both use this release's tag:
 
 ```sh
 # Claude Code
-npx skills add "https://github.com/mvschwarz/openrig/tree/v0.7.14/skills/rigs" --skill rigs --agent claude-code
+npx skills add "https://github.com/mvschwarz/openrig/tree/v0.7.12/skills/rigs" --skill rigs --agent claude-code
 
 # Codex
-npx skills add "https://github.com/mvschwarz/openrig/tree/v0.7.14/skills/rigs" --skill rigs --agent codex
+npx skills add "https://github.com/mvschwarz/openrig/tree/v0.7.12/skills/rigs" --skill rigs --agent codex
 ```
 
 If OpenRig is already installed, use the release tag matching your installed `rig --version` instead: replace the tag
@@ -203,11 +203,11 @@ 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/v0.7.14/scripts/install.sh | sh -s -- --dry-run
+curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.12/scripts/install.sh | sh -s -- --dry-run
 ```
 
 The same command without `--dry-run`,
-`curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.14/scripts/install.sh | sh`, installs the
+`curl -fsSL https://raw.githubusercontent.com/mvschwarz/openrig/v0.7.12/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

docs/reference/quickstart.md

@@ -12,9 +12,9 @@ In your project folder, add OpenRig's `rigs` skill to the agent you use:
 
 ```sh
 # Claude Code
-npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.14/skills/rigs --skill rigs --agent claude-code
+npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.12/skills/rigs --skill rigs --agent claude-code
 # Codex
-npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.14/skills/rigs --skill rigs --agent codex
+npx skills add https://github.com/mvschwarz/openrig/tree/v0.7.12/skills/rigs --skill rigs --agent codex
 ```
 
 These use this release's tag, for a new install. If OpenRig is already installed, replace the tag after `tree/` with

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "openrig",
-  "version": "0.7.14",
+  "version": "0.7.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "openrig",
-      "version": "0.7.14",
+      "version": "0.7.1",
       "license": "Apache-2.0",
       "workspaces": [
         "packages/daemon",
@@ -7298,7 +7298,7 @@
     },
     "packages/cli": {
       "name": "@openrig/cli",
-      "version": "0.7.14",
+      "version": "0.7.1",
       "hasInstallScript": true,
       "license": "Apache-2.0",
       "dependencies": {
@@ -7336,7 +7336,7 @@
     },
     "packages/daemon": {
       "name": "@openrig/daemon",
-      "version": "0.7.14",
+      "version": "0.7.1",
       "license": "Apache-2.0",
       "dependencies": {
         "@hono/node-server": "^1.13.0",
@@ -7380,7 +7380,7 @@
     },
     "packages/ui": {
       "name": "@openrig/ui",
-      "version": "0.7.14",
+      "version": "0.7.1",
       "license": "Apache-2.0",
       "dependencies": {
         "@fontsource-variable/jetbrains-mono": "^5.2.8",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "openrig",
-  "version": "0.7.14",
+  "version": "0.7.1",
   "private": true,
   "license": "Apache-2.0",
   "type": "module",

packages/acp/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@openrig/acp",
-  "version": "0.7.14",
+  "version": "0.7.1",
   "description": "OpenRig as an agent in ACP apps: starts rig acp serve from the @openrig/cli it pins.",
   "type": "module",
   "bin": {
@@ -15,7 +15,7 @@
     "node": ">=22"
   },
   "dependencies": {
-    "@openrig/cli": "0.7.14"
+    "@openrig/cli": "0.7.1"
   },
   "license": "Apache-2.0",
   "author": {

packages/acp/registry/agent.json

@@ -1,7 +1,7 @@
 {
   "id": "openrig",
   "name": "OpenRig",
-  "version": "0.7.14",
+  "version": "0.7.1",
   "description": "Send work to your OpenRig team of agents and follow it through to the result",
   "repository": "https://github.com/mvschwarz/openrig",
   "website": "https://openrig.dev",
@@ -12,7 +12,7 @@
   "license_url": "https://github.com/mvschwarz/openrig/blob/main/LICENSE",
   "distribution": {
     "npx": {
-      "package": "@openrig/[email protected]4"
+      "package": "@openrig/[email protected]"
     }
   }
 }

packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@openrig/cli",
-  "version": "0.7.14",
+  "version": "0.7.1",
   "type": "module",
   "description": "Build your own network of agents from Claude Code, Codex and Pi: persistent teams with roles, shared context and owned work.",
   "keywords": [

packages/daemon/assets/plugins/openrig/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "openrig",
-  "version": "0.7.14",
+  "version": "0.1.0",
   "description": "Work with your OpenRig team from your coding agent: hand it work, follow that work, answer its questions and read what it returns.",
   "author": {
     "name": "Mike Schwarz",

packages/daemon/assets/plugins/openrig/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "openrig",
-  "version": "0.7.14",
+  "version": "0.1.0",
   "description": "Work with your OpenRig team from your coding agent: hand it work, follow that work, answer its questions and read what it returns.",
   "author": {
     "name": "Mike Schwarz",

packages/daemon/assets/plugins/openrig/.cursor-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "openrig",
-  "version": "0.7.14",
+  "version": "0.1.0",
   "description": "Work with your OpenRig team from your coding agent: hand it work, follow that work, answer its questions and read what it returns.",
   "logo": "assets/logo.svg"
 }

packages/daemon/assets/plugins/openrig/plugin.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
   "name": "openrig",
-  "version": "0.7.14",
+  "version": "0.1.0",
   "description": "Work with your OpenRig team from your coding agent: hand it work, follow that work, answer its questions and read what it returns.",
   "author": {
     "name": "Mike Schwarz",

packages/daemon/context-packs-src/onboarding-width/public-what-you-can-do.md

@@ -10,8 +10,8 @@ already does it, or two verbs crossed already answer it. **You cannot look up a
 not know exists.** So the point of what follows is not skill. It is recognition: that when a task
 lands, something rhymes, and you go check instead of building.
 
-Ninety-three top-level verbs ship (capability canon refreshed through
-`capability-delta-v0.7.14`). This marker describes the pack's teaching, not publication or live
+Ninety-two top-level verbs ship (capability canon refreshed through
+`capability-delta-v0.7.12`). This marker describes the pack's teaching, not publication or live
 adoption. Read this once for shape, and let it make you suspicious that a thing already exists.
 What changed in the installed version, and what to do about it, is one read away:
 `rig context get reference/whats-new.md`.
@@ -33,10 +33,7 @@ one command away. Your memory of the fleet is a claim about the past.
   *including the broken ones*. **The bare form is the question you usually want:** `--active`
   filters to running and **hides detached, exited and attention states** — which are exactly the
   cells you were scanning for. A seat whose tmux session is confirmed gone reads `detached`
-  (activity reason `session_missing`), and `rig ps --nodes` says to restore or relaunch it. Since 0.7.14 `rig ps` ends with connected
-  assistants and their check-in line, and node reads take turns through their three synchronous
-  steps, so other reads and sends aren't held behind a burst of them; a nodes response is no longer
-  one snapshot across those steps, and in a burst each one comes later.
+  (activity reason `session_missing`), and `rig ps --nodes` says to restore or relaunch it.
   A verified exited process in a retained pane reads `pane_exited`, instead of running and ready;
   outstanding work uses the existing recovery-finding route. An unknown observation is not an exit.
   `rig ps --nodes` also shows the model each seat was last seen using (`modelObservation`, with its
@@ -112,8 +109,7 @@ one command away. Your memory of the fleet is a claim about the past.
   Closed work's content and metadata can also be erased on a schedule
   (`retention.closed_content_days`, `retention.closed_metadata_days`; `0`, the default, keeps them),
   with an optional `retention.erasure_ledger`, for a retention policy or an account deletion. It's
-  opt-in, and a retried create can't bring erased work back. Since 0.7.14 an account's erasure also
-  removes OpenRig's own wakes and notices for that account's seats.
+  opt-in, and a retried create can't bring erased work back.
 - **Check the exit code.** Many commands used to print nothing and exit 0 when a daemon read
   failed; now they exit 1 or 2. Empty output from a failed read is not empty state.
 
@@ -128,9 +124,7 @@ unseen — which makes messages the one delivery channel that never gets skipped
   consequential delivery gets an effect check at the far end rather than a blind retry.
   `--verify` exits 1 when it can't confirm what its Enter did (`submission unconfirmed`, or text
   staged but not submitted): the message may still have arrived, so inspect with `rig capture`
-  before any resend, and don't let a script resend on that exit code alone. Since 0.7.14 an ordinary
-  send is refused as needing input while an open Claude Code or Codex `/model` picker is
-  positively identified; nothing is typed. A
+  before any resend, and don't let a script resend on that exit code alone. A
   producer-link advisory reading `no_activity_signal` means activity could not be determined;
   confirm by effect, never relaunch on it. A send or wake to a seat whose latest hook is a
   question waiting for a person is refused with `target_needs_input`, even when its screen
@@ -181,8 +175,7 @@ unseen — which makes messages the one delivery channel that never gets skipped
   and the box was empty after Enter; it only reports.
 - **`rig broadcast`** — one fact to everyone at once, rather than relaying it N times and getting
   the wording wrong on the fourth. **Blast radius is real** — on a large rig this lands in every
-  seat's turn. A broadcast that reached no one says so, and a recipient a seat's setting held
-  reads as retained, not sent.
+  seat's turn. A broadcast that reached no one says so.
 - **`rig capture <seat>`** — read what is on their screen right now. **`--rig` / `--pod`** captures
   every seat in one call. `--lines N` is the last N lines (stderr says how many earlier lines were
   cut); `--history-plus-pane` gives the older N history lines plus the whole pane. A short capture
@@ -262,7 +255,7 @@ unseen — which makes messages the one delivery channel that never gets skipped
   default. **Opening `rig terminal open saved:kernel --window` is the step after install or start**:
   say you're opening it, run it on the daemon's desktop, then tell the person where it opened. Don't
   wait for a phrase like “show me my agents”; the same command brings their agents back later. No
-  YAML, herdr when installed or plain tmux. Rex tiles show each seat's agent and its activity. The layout follows the terminal's width: on a narrow
+  YAML, herdr when installed or plain tmux. The layout follows the terminal's width: on a narrow
   one the operator gets the first tab; on a wide one the dashboard and operator sit side by side
   and the advisor gets its own tab. A `kernel` view the person saved still wins.
   Tiles show their seat's screen without attaching, so whole lines show at any tile width; a
@@ -487,8 +480,7 @@ relay is a reason to inspect the available routing, not to surrender judgment to
 - **`rig workflow validate`** — will this spec instantiate at all, before a run finds out for you. A
   step with no allowed exit that can finish it or route onward is reported as `step_cannot_finish`.
   An allowed mapped exit is valid; in a dependency graph, `handoff` can finish a sink without a next
-  step. Since 0.7.14 each step with a `harness:` pin gets a `harness_pin_unchecked` warning:
-  validation doesn't check pins against seats.
+  step.
 - **`rig workflow compile <mission>`** / **`instantiate-lifecycle`** — derive one executable graph
   from `project.yaml` → `mission.yaml` → `slice.yaml`, inspect it without writing, then start an
   eligible graph with an opaque replay key. Typed acceptance candidate, verdict, and evidence
@@ -640,8 +632,6 @@ not have to tear one down to change it, and you rarely have to start from nothin
   steps): hand it work, follow it, answer the team's questions in their own words, and read the
   result. VS Code, Copilot CLI and Cursor load it with no seat: give the app its own
   `rig mcp serve --front-door` server with `OPENRIG_SESSION_NAME` set and turn the plugin's off.
-  A skill-only edition for Claude's plugin directory, in `plugins/claude-directory/`, follows
-  handed-on work and keeps its reply route.
 - **A Claude member can bring its own setup** in the rig spec: `agent_file`,
   `append_system_prompt_file`, `plugin_dirs` and `advisor` (an Advisor model). They're copied into
   that seat's own storage, never the project, so one seat in a shared repository can have its own
@@ -759,10 +749,7 @@ not have to tear one down to change it, and you rarely have to start from nothin
   conversation continuity stays unverified. Both refuse while typing protection is on.
 - **`rig unclaim` / `rig release`** — stop managing one adopted session, or all of them,
   **without killing the process or the agent inside it**. Handing something back to its human is a
-  first-class move, not an abandonment. To stop an adopted session after unclaiming it, end its
-  process and its tmux session; closing only the viewing window detaches it. Since 0.7.13 a fresh
-  `rig launch` after `rig unclaim` ended the server's last session reopens the server (a released
-  session that's still running blocks it, as before).
+  first-class move, not an abandonment.
 
 > **Two words that mean different things at different altitudes.** `unclaim` and `release` each
 > exist both here — *stop managing a live session* — and in the queue — *put a work row back down*.
@@ -879,9 +866,7 @@ version, from checking the environment to a useful report to the OpenRig team.
   effectful handover path.** Both forms perform the operation by default; pass `--dry-run`
   to plan without changing the seat. Check the selected source and continuity evidence first. A
   seat can be handed over again into its own pane, so a seat left half-handed-over recovers by
-  running the handover again. A handover whose successor was launched but not ready says it may be
-  running: look at the pane before retrying. A handover cleans up the retiring occupant's own claims
-  and jobs. A Claude seat started through a `#!` wrapper that runs Claude without
+  running the handover again. A Claude seat started through a `#!` wrapper that runs Claude without
   `exec`, or through the installer's `~/.local/bin/claude` symlink, rebuilds with `--source
   rebuild`; a failed rebuild leaves the resume target on the earlier conversation, so rebuild again
   rather than relaunching it fresh.
@@ -972,8 +957,7 @@ version, from checking the environment to a useful report to the OpenRig team.
 - **`rig compact-plan`** → **`rig compact`** — who is near the context wall, then act on it.
   **Ordering matters: running `compact` without the plan is guessing which seat needed it.**
   Managed compaction waits for the seat's restore map before compacting; `rig compact --state`,
-  `--cancel` and `--skip-map` inspect, stop or skip that wait; since 0.7.14 `--state` also names the
-  input check behind a `draft_input_unknown` stop (`inputReadFailure`). If preparation stops at an ordinary
+  `--cancel` and `--skip-map` inspect, stop or skip that wait. If preparation stops at an ordinary
   approval, a manual `rig compact` keeps waiting for someone to answer it, within its 180-second
   budget; OpenRig never answers it, and the request times out after 240 seconds. The `claude-compaction-restore`
   skill covers writing and reading the map; `claude-compact-in-place` no longer ships. After a
@@ -1247,9 +1231,6 @@ reaching them is ordinary work rather than an escalation.
   see which host you are acting on, and check a registered host is *actually* usable end to end.
   `rig host edit <id>` corrects an entry (its id and transport stay fixed), and `rig host remove <id>` deletes
   only the registry entry: credential files, the learned identity and the selection stay as they are.
-- **`rig network`** — **experimental, unsupported and at your own risk.** It connects one local seat
-  to a hub you already have. The daemon starts the connector, sweeps leftover credential folders and
-  schedules local rounds even with no binding, so it isn't inert, and its migrations apply on upgrade.
 - **A cross-host queue write that times out reads INDETERMINATE, not failed.** The CLI waits 15
   seconds for the daemon's answer, and `rig queue create` with no answer reads the row back once by
   its ID. Reconcile by that ID before any retry; don't send the write twice.
@@ -1294,11 +1275,6 @@ reaching them is ordinary work rather than an escalation.
   a person can decide directly in the TUI's pairing settings. A relay connector that can't accept
   incoming connections can pull instead (`OPENRIG_RELAY_MODE=pull` in the relay secrets file); push
   stays the default, and a person's connector binding can be unset.
-  **Since 0.7.13 a person's runner answer gets a note back:** who decides and in what state, or that
-  the answer is kept and the tool hasn't run, and a runner waiting on it wakes on the decision. Read
-  that note before asking the person to answer again. The answer a receiving agent reads says what an
-  answer typed at the machine's terminal means, to act on a clear one with the request's own decision
-  command, and how to ask the person back, with a new ask to them, when it's unclear.
 - **The `relay` gateway connector** carries a registered person's questions, updates and notices to a
   bridge process you run, over loopback HTTP, and takes their replies back with the sender resolved
   against their bindings. It's off until a person has a relay binding (`rig gateway human add <id> …
@@ -1437,8 +1413,7 @@ your circumstances is configuration, and the ones that are not, another agent ca
   conversation it is in the middle of. **`provider signals` reports anomalies rather than a
   listing** — unbound seats, accounts shared across several seats. A Codex seat showing its
   usage-limit banner reads activity `unknown` with reason `usage_limit`, and keeps it while its
-  screen redraws (a banner with a curly apostrophe included); a Claude seat whose usage-limit message
-  states a reset time is parked until then, instead of being retried every few minutes; `provider signals` can show a provider-event signal for that seat's account that
+  screen redraws; `provider signals` can show a provider-event signal for that seat's account that
   allows a switch decision. With `provider.usage_windows` on (off by default), `rig provider status`
   shows Codex's windows and Claude's 5-hour and weekly windows from the seats' status lines, each
   seat's configured account, and with `--all-hosts` or `--hosts <ids>` every registered machine's
@@ -1509,14 +1484,7 @@ your circumstances is configuration, and the ones that are not, another agent ca
   (empty, the default, means every signed-in one). `--router-line yes|no` decides whether OpenRig
   adds its one line to shared guidance files (`skills.router_line.enabled`).
   Setup offers to register the person when none is registered, and a team started from the
-  person's own assistant fits their own logins, with a review before anything is committed. Since
-  0.7.13, `rig daemon start` ends with that offer when the person registry can be read and holds
-  nobody. The kernel operator's startup is also taught to run `rig gateway participants` before its
-  first greeting and offer registration when nobody is registered. The agent asks first and
-  registers only on the person's yes. Since 0.7.14 setup guidance tells the agent to fit a first
-  team to the person's logins: a login `rig doctor` reports missing matters when a seat you'd start
-  runs on that harness, so say which one before launch
-  (`rig specs preview <team> --kind rig --json` shows each seat's runtime).
+  person's own assistant fits their own logins, with a review before anything is committed.
 - **`rig usage series`** — what a seat's token curve has looked like over time: climbing steadily,
   reset, or stopped reporting entirely. The last one is a signal, not a gap.
 - **`rig usage top`** — which of this machine's seats used the most tokens over a window (`--window
@@ -1601,12 +1569,7 @@ your circumstances is configuration, and the ones that are not, another agent ca
   Clients on MCP protocol revision 2026-07-28 and on the 2025 revisions share the same endpoint.
   With a grant service (experimental and off by default; you serve it to the internet behind
   your own tunnel), `--task-destination <member@rig>` names where a caller's `rig_queue_create`
-  goes when it gives no destination, and a seat's name alone resolves within the grant's team. Since
-  0.7.14 a plain name finds the seat's actual session (its own name, then its pod-member form, then
-  that name's slug, among seats the grant can address): an ambiguous name returns
-  `destination_ambiguous` with the candidates, an unknown one `destination_not_found`, and an
-  unreadable team `destination_unchecked`. Its person sees each tool's plain title, and under a
-  grant service's fair-use limit only new work is held back. A
+  goes when it gives no destination, and a seat's name alone resolves within the grant's team. A
   grant that names a person adds `rig_question_list` and `rig_question_answer`, so that person
   answers the team's questions from their assistant, recorded as their answer through that
   connection.
@@ -1618,10 +1581,6 @@ your circumstances is configuration, and the ones that are not, another agent ca
   with secret names, never values): OpenRig then POSTs a pointer to the row when work lands, and a
   2xx records only that the ring was accepted, not that anything woke. When work waits for a
   member nothing can wake, or a rung member doesn't claim it in time, its person is told.
-  A hosted member's roster entry shows whether it's checking in: its expected cadence and its last
-  check-in, read as checking in, overdue, not yet checked in or checking in on request, and unknown
-  when that can't be read. It's a status, never a gate on delivery. Since 0.7.14 `rig ps`,
-  `rig_team` and the TUI's tables list connected assistants with that check-in line.
   **Internet-facing hosted members are experimental, unsupported, off by default and at your
   own risk.** Local assistants can connect locally. Internet access needs your own TLS tunnel or
   proxy and a narrowly scoped, revocable token; see `reference/hosted-members.md`. Task terminal
@@ -1679,8 +1638,7 @@ database. It is one more set of primitives on a machine full of them.
   original byte size and the exact expansion command. Read the complete original record with
   `rig queue show <qitemId> --full --json`; a preview or a compact list is not the whole body.
   Compact `rig queue list` rows carry `summary`; the fields they leave out are absent, not empty,
-  and `fieldsElided` names them. Since 0.7.14 `--tag` (repeatable; every tag must match),
-  `--mission` and `--slice` filter exactly, and bounded listings over retained history use indexes.
+  and `fieldsElided` names them.
   Delivery status is separate from content: inspect its receipt before claiming it was consumed.
 - **"What am I waiting on?"** — the queue row's `waiting` view derives its owner, exact blocker
   and blocker owner, last meaningful change, activity confidence and next backstop from live
@@ -1718,6 +1676,6 @@ deliberate writing.
 
 **`rig --help`**, then `rig <verb> --help`.
 
-Ninety-three verbs ship, and this page named most of them once. **You will not remember which — you
+Ninety-two verbs ship, and this page named most of them once. **You will not remember which — you
 are meant to remember only that the list is long enough to be worth reading before you build
 anything.** The question is never *how do I write this*; it is **does this already exist**.

packages/daemon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@openrig/daemon",
-  "version": "0.7.14",
+  "version": "0.7.1",
   "private": true,
   "license": "Apache-2.0",
   "type": "module",

packages/daemon/src/adapters/tmux.ts

@@ -494,7 +494,29 @@ export class TmuxAdapter {
       if (probe.state !== "present") return false;
       const panes = await this.listPanes(target.session);
       return panes.length === 1 && panes[0]!.id === target.pane && await this.isPaneDead(target.pane);
-    });
+    }, target => this.lastClientActivity(target.session));
+  }
+
+  /** When a writable tmux client attached to this session last had activity (epoch ms, whole seconds), or null when no
+   *  writable client is attached. tmux's client_activity is session-wide and covers more than text keys, so this is
+   *  recent writable-client activity, not proof of typing in the pane. A read-only client (a viewer tile) can't type
+   *  into the seat and doesn't count. An attach counts: tmux can't tell a key in the attach's second from the attach,
+   *  so a delivery can be held briefly after one. A tmux failure or an unreadable row throws: the guard then holds
+   *  the paste rather than take it for "no activity" (#3076). */
+  async lastClientActivity(session: string): Promise<number | null> {
+    const exact = exactTarget(session, "session");
+    const format = `#{client_readonly}${TMUX_FIELD_SEPARATOR}#{client_activity}`;
+    const output = await this.run(["tmux", "list-clients", "-t", exact, "-F", format],
+      `tmux list-clients -t ${shellQuote(exact)} -F "${format}"`);
+    let latest: number | null = null;
+    for (const line of output.split("\n").map(row => row.trim()).filter(row => row !== "")) {
+      const [readonly, activity, ...rest] = line.split(TMUX_FIELD_SEPARATOR);
+      if (rest.length || !/^[01]$/.test(readonly ?? "") || !/^[1-9]\d*$/.test(activity ?? "")) {
+        throw new Error(`tmux list-clients: unreadable client row for ${session}`);
+      }
+      if (readonly === "0") latest = Math.max(latest ?? 0, Number(activity) * 1000);
+    }
+    return latest;
   }
   private readonly freshProbes = new Map<string, string>();
   private readonly freshManaged = new Map<string, {nodeId: string; pane: string}>();
@@ -1081,12 +1103,13 @@ export class TmuxAdapter {
         // A paste the guard proves it left unsubmitted (#2429) is submitted first, under this lease, once this send's own
         // checks pass; they then aren't run a second time.
         beforePaste: async () => {
-          let checked = false;
-          const checks = async () => { await options?.beforePaste?.(); beforeWrite(); beforeInput?.(); checked = true; };
+          // This send's own checks can wait (a conversation lookup, an input capture), so they run first, and the
+          // guard's checks run last (#3076). A check that refuses leaves a leftover's record for the next delivery.
+          await options?.beforePaste?.();
+          const checks = async () => { beforeWrite(); beforeInput?.(); };
           // The synchronous checks run again just before the key: the guard's re-read awaits after `checks`.
           const enter = async () => { beforeWrite(); beforeInput?.(); return (await this.sendKeysUnchecked(pane, ["Enter"])).ok; };
           if (checkDraft) await guard!.beforeInput(identity, false, text, { checks, enter });
-          if (!checked) await options?.beforePaste?.();
         },
         afterPaste: async () => { if (checkDraft) await guard!.notePasted(identity, text); },
       } : undefined;

packages/daemon/src/domain/seat-delivery-guard.ts

@@ -60,6 +60,15 @@ export interface TypingGuardConfig extends TypingGuardSettings { revision: strin
 const DEFAULT_GUARD: TypingGuardConfig = { mode: "off", revision: "legacy-off" };
 /** How long the guard waits before reading again an input it couldn't read right after its own paste or Enter (#2429). */
 const OWN_INPUT_REREAD_MS = 300;
+/** Right before an automatic paste, the empty input is read once more after this wait: a draft typed (or redrawn by the
+ *  agent's TUI) since the first read is seen before the paste, not after it, where nothing is removed (#2447, #3076). */
+const PASTE_SETTLE_MS = 300;
+/** Recent activity at a writable tmux client of the seat (keys, but also other client input; session-wide, not proof of
+ *  typing in this pane) holds the paste: a later attempt can deliver once it reads quiet and the input empty. tmux
+ *  reports client activity in whole seconds, so this is three of them (#3076). */
+const INPUT_QUIET_MS = 3000;
+/** tmux's client_activity is in whole seconds; the activity may have come up to this much after the second it reports. */
+const CLIENT_ACTIVITY_PRECISION_MS = 999;
 /** A terminal member with nothing set holds OpenRig's own notices and takes deliberate sends. */
 const TERMINAL_DEFAULT: TypingGuardConfig = { mode: "off", level: "sends-only", revision: "terminal-default" };
 /** The actor a guard change records when its request carries no seat identity, as from an ordinary terminal (#2350).
@@ -122,6 +131,9 @@ export class SeatDeliveryGuard {
   /** A paste this guard kept from being submitted because it couldn't read the input, per node (#2429). */
   private readonly leftovers = new Map<string, { target: GuardTarget; humanEpoch: number; text: string; collapsedPaste: string | null }>();
   private inspectInput?: (target: GuardTarget, owned?: ComposerOwnership) => Promise<ComposerInput | null>;
+  /** When a writable tmux client of the seat last had activity (epoch ms, whole seconds); null when no writable client
+   *  is attached; a rejection when that can't be known. */
+  private inputActivity?: (target: GuardTarget) => Promise<number | null>;
   private inputAbsent?: (target: GuardTarget) => Promise<boolean>;
   private readonly hasModeConfig: boolean;
 
@@ -133,9 +145,11 @@ export class SeatDeliveryGuard {
     this.hasModeConfig = (db.prepare("PRAGMA table_info(seat_delivery_guards)").all() as Array<{ name: string }>).some(c => c.name === "desired_config");
   }
 
-  attachInputInspection(inspect: (target: GuardTarget, owned?: ComposerOwnership) => Promise<ComposerInput | null>, absent: (target: GuardTarget) => Promise<boolean>): void {
+  attachInputInspection(inspect: (target: GuardTarget, owned?: ComposerOwnership) => Promise<ComposerInput | null>, absent: (target: GuardTarget) => Promise<boolean>,
+    activity?: (target: GuardTarget) => Promise<number | null>): void {
     this.inspectInput = inspect;
     this.inputAbsent = absent;
+    this.inputActivity = activity;
   }
 
   /** Startup-only, before exposing routes or starting writers. A stopped operation cannot
@@ -483,6 +497,9 @@ export class SeatDeliveryGuard {
     } else if (input.state !== "empty") {
       throw new DeliveryGuardError("draft_input_busy", "The seat has an unsubmitted input. Automatic delivery is held.");
     } else {
+      // Right before the paste (the adapter's check passes the text): the read above is a snapshot, and a draft typed
+      // after it would get the paste appended and then lose the Enter, with the paste left in the person's input (#3076).
+      if (pasteText !== undefined) input = await this.quietBeforePaste(target, read, unknown);
       const uncertainty = pasteText === undefined ? null : composerPasteUncertainty(input, pasteText);
       if (uncertainty) {
         const message = uncertainty === "controls"
@@ -496,6 +513,31 @@ export class SeatDeliveryGuard {
     }
   }
 
+  /** No recent writable-client activity before a short wait, the input still empty when read again after it, and still
+   *  no recent activity after that read. A time tmux can't report holds the paste, as an unreadable input does. The
+   *  read it returns is the one the paste goes on. Typing after this final check and before the paste can still leave
+   *  the notice beside the person's draft. In the tested append-mode schedule the Enter is then refused and nothing is
+   *  submitted or deleted; that isn't a guarantee for every interleaving (#3076). */
+  private async quietBeforePaste(target: GuardTarget, read: () => Promise<ComposerInput | null>,
+    unknown: (reading: ComposerInput | null) => DeliveryGuardError): Promise<ComposerInput> {
+    const quiet = async () => {
+      if (!this.inputActivity) return;
+      const last = await this.inputActivity(target).then(at => ({ at }), () => null);
+      if (!last) throw new DeliveryGuardError("draft_input_unknown", "OpenRig couldn't read the recent activity of the seat's terminal clients. Automatic delivery is held; nothing was pasted.");
+      // tmux reports whole seconds: activity at N.999 s reads as N, so the quiet window counts from the end of that second.
+      if (last.at !== null && Date.now() - (last.at + CLIENT_ACTIVITY_PRECISION_MS) < INPUT_QUIET_MS) {
+        throw new DeliveryGuardError("draft_input_busy", "A writable client of this seat had recent activity. Automatic delivery is held; nothing was pasted.");
+      }
+    };
+    await quiet();
+    await this.sleep(PASTE_SETTLE_MS);
+    const again = await read();
+    if (!again || again.state === "unknown") throw unknown(again);
+    if (again.state !== "empty") throw new DeliveryGuardError("draft_input_busy", "The input changed just before the paste. Automatic delivery is held; nothing was pasted.");
+    await quiet();
+    return again;
+  }
+
   /** A leftover paste is OpenRig's only while it's exactly there: the same occupant, no person's input through OpenRig
    * since, and its text or the label recorded for it. A label first seen later proves nothing (#1871). */
   private ownsLeftover(target: GuardTarget, input: ComposerInput): boolean {

packages/daemon/test/draft-aware-delivery.test.ts

@@ -32,7 +32,7 @@ function signal() {
   return { ready: new Promise<void>(resolve => { release = resolve; }), release: () => release() };
 }
 
-function fixture(saved?: Buffer, opts: { runtime?: string; idleHook?: boolean } = {}) {
+function fixture(saved?: Buffer, opts: { runtime?: string; idleHook?: boolean; appendPaste?: boolean } = {}) {
   const db = new Database(saved ?? ":memory:");
   if (!saved) {
     migrate(db, ALL_MIGRATIONS);
@@ -42,23 +42,27 @@ function fixture(saved?: Buffer, opts: { runtime?: string; idleHook?: boolean }
       INSERT INTO sessions(id,node_id,session_name,status) VALUES ('sa','a','worker@test','running'),('sb','b','sibling@test','running');
       INSERT INTO occupant_tenures(id,node_id,generation_ordinal,generation_uuid,kind) VALUES ('ga','a',1,'g1','fresh'),('gb','b',1,'g2','fresh');`);
   }
-  const guard = new SeatDeliveryGuard(db, name => resolveGuardTarget(db, name));
+  const hooks: { load?: () => void | Promise<void>; paste?: () => void | Promise<void>; sleep?: (ms: number) => void | Promise<void>;
+    settle?: (ms: number) => void | Promise<void>; clients?: () => string } = {};
+  // The guard's own waits (its re-reads) go to `settle`, kept apart from the transport's `sleep`; none is real time.
+  const guard = new SeatDeliveryGuard(db, name => resolveGuardTarget(db, name), async ms => { await hooks.settle?.(ms); });
   guard.recoverActivation();
   const bodies: Record<string, string | null> = { a: "", b: "" };
   const history: Record<string, string> = { a: "", b: "" };
   const writes: string[] = [], submissions: string[] = [];
   const observations: unknown[] = [];
-  const hooks: { load?: () => void | Promise<void>; paste?: () => void | Promise<void>; sleep?: (ms: number) => void | Promise<void> } = {};
   let staged = "";
   // A terminal pane shows a submitted line as typed; an agent pane echoes a sent message under its prompt, as Claude does.
   const echo = (node: string, body: string) => (db.prepare("SELECT runtime FROM nodes WHERE id=?").get(node) as { runtime: string }).runtime === "terminal"
     ? `${body}\n` : `${body.split("\n").map((row, i) => (i ? "  " : "❯ ") + row).join("\n")}\n`;
   const tmux = new TmuxAdapter(async command => {
+    if (command.includes("list-clients")) return hooks.clients?.() ?? "";
     if (command.includes("load-buffer")) await hooks.load?.();
     const node = command.includes("%2") ? "b" : "a";
     if (command.includes("paste-buffer")) {
       await hooks.paste?.();
-      writes.push(`paste:${node}:${staged}`); bodies[node] = staged;
+      // appendPaste: a paste lands after whatever the input already holds, as in a real composer (#3076 tests).
+      writes.push(`paste:${node}:${staged}`); bodies[node] = opts.appendPaste ? (bodies[node] ?? "") + staged : staged;
     }
     if (command.includes("send-keys")) {
       writes.push(`keys:${node}:${command}`);
@@ -1156,3 +1160,158 @@ describe("draft-aware at a full held quota (#2430)", () => {
     expect(f.outbox.retentionUsage("a").count).toBe(100);
   });
 });
+
+
+describe("#3076: a draft typed between the guard's read and the paste", () => {
+  const draft = "Please send it to the team. Also";
+  const notice = "[OpenRig watchdog scheduler · policy: periodic-reminder · job: j]\nWake timer fired.";
+  const pasted = (f: ReturnType<typeof fixture>) => f.writes.filter(w => w.startsWith("paste:"));
+  const setActivity = (f: ReturnType<typeof fixture>, at: () => number | null) => {
+    vi.spyOn(f.tmux, "lastClientActivity").mockImplementation(async () => at());
+  };
+
+  it("is seen by the re-read after the settle wait: the notice is held and nothing is pasted into the person's input", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    const waits: number[] = [];
+    // The person starts typing during the guard's wait right before the paste (the first read saw an empty input).
+    f.hooks.settle = ms => { waits.push(ms); if (!pasted(f).length) f.bodies.a = draft; };
+    const result = await f.send("watchdog-wake", notice, { openrigNotice: true, verify: false });
+    expect(result).toMatchObject({ outcome: "retained", sent: false, reason: "draft_input_busy" });
+    expect(waits[0]).toBe(300);
+    expect(pasted(f)).toEqual([]);
+    expect(f.submissions).toEqual([]);
+    expect(f.bodies.a).toBe(draft);
+  });
+
+  it("holds on recent writable-client activity, also activity during the wait, and a later attempt delivers once quiet", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    let last: number | null = Date.now() - 1000;
+    setActivity(f, () => last);
+    const busy = await f.send("wake-1", "Wake timer fired.", { openrigNotice: true, verify: false });
+    expect(busy).toMatchObject({ outcome: "retained", sent: false, reason: "draft_input_busy" });
+
+    last = Date.now() - 60_000;
+    f.hooks.settle = () => { last = Date.now(); };   // activity during the settle wait, nothing drawn yet
+    f.advance();
+    const during = await f.send("wake-2", "Wake timer fired again.", { openrigNotice: true, verify: false });
+    expect(during).toMatchObject({ outcome: "retained", sent: false, reason: "draft_input_busy" });
+    expect(pasted(f)).toEqual([]);
+
+    last = Date.now() - 60_000;
+    f.hooks.settle = undefined;
+    f.advance();
+    const sent = await f.send("wake-3", "Wake timer fired a third time.", { openrigNotice: true, verify: false });
+    expect(sent).toMatchObject({ ok: true });
+    expect(f.submissions).toHaveLength(1);
+  });
+
+  it("counts the quiet window from the end of the whole second tmux reports: 3.5 s ago still holds, 4.5 s ago delivers", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    let last = Date.now() - 3500;   // activity at N.999 s reads as N: it may have been 2.5 s ago
+    setActivity(f, () => last);
+    const held = await f.send("whole-second", "Wake timer fired.", { openrigNotice: true, verify: false });
+    expect(held).toMatchObject({ outcome: "retained", sent: false, reason: "draft_input_busy" });
+    last = Date.now() - 4500;
+    f.advance();
+    const sent = await f.send("whole-second-later", "Wake timer fired again.", { openrigNotice: true, verify: false });
+    expect(sent).toMatchObject({ ok: true });
+  });
+
+  it("holds as unreadable when tmux can't report recent writable-client activity; with no writable client it delivers", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    vi.spyOn(f.tmux, "lastClientActivity").mockRejectedValue(new Error("command timed out"));
+    const held = await f.send("activity-unreadable", notice, { openrigNotice: true, verify: false });
+    expect(held).toMatchObject({ sent: false, reason: "draft_input_unknown" });
+    expect(pasted(f)).toEqual([]);
+    setActivity(f, () => null);
+    f.advance();
+    const sent = await f.send("no-client", "Wake timer fired again.", { openrigNotice: true, verify: false });
+    expect(sent).toMatchObject({ ok: true });
+  });
+
+  it("holds as unreadable when the input can't be read after the wait", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    f.hooks.settle = () => { if (!pasted(f).length) f.bodies.a = null; };
+    const result = await f.send("unreadable", notice, { openrigNotice: true, verify: false });
+    expect(result).toMatchObject({ sent: false, reason: "draft_input_unknown" });
+    expect(pasted(f)).toEqual([]);
+  });
+
+  it("runs the send's own waiting checks before the guard's last read, so a draft typed while they wait is seen", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    // A conditional send's checks await (a conversation lookup); the person types meanwhile. The adapter returns the
+    // guard's refusal as its result.
+    const result = await f.guard.operation("worker@test", () =>
+      f.tmux.sendText("worker@test", notice, undefined, { beforePaste: async () => { f.bodies.a = draft; } }));
+    expect(result).toMatchObject({ ok: false, code: "draft_input_busy" });
+    expect(pasted(f)).toEqual([]);
+    expect(f.bodies.a).toBe(draft);
+  });
+
+  it("applies to every automatic input under draft-aware, a deliberate send included, as the empty check already did", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    f.hooks.settle = () => { if (!pasted(f).length) f.bodies.a = draft; };
+    const result = await f.send("deliberate", "a message from another seat", { verify: false });
+    expect(result).toMatchObject({ sent: false, reason: "draft_input_busy" });
+    expect(pasted(f)).toEqual([]);
+    expect(f.bodies.a).toBe(draft);
+  });
+
+  it("the remaining gap: typing after the final check and before the paste leaves the notice beside the draft, never submitted or deleted", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    let typed = false;
+    // The person types between the guard's last check and tmux's paste: the paste lands after their text.
+    f.hooks.paste = () => { if (!typed) { typed = true; f.bodies.a = draft; } };
+    const result = await f.send("final-boundary", notice, { openrigNotice: true, verify: false });
+    // Contamination: exactly one paste, and the input is the person's text followed by it, nothing removed.
+    expect(pasted(f)).toHaveLength(1);
+    const staged = pasted(f)[0]!.slice("paste:a:".length);
+    expect(f.bodies.a).toBe(draft + staged);
+    // Submission: none. No key at all reached the pane (no Enter, C-m or C-u), and the guard says why.
+    expect(f.writes.filter(w => w.startsWith("keys:a"))).toEqual([]);
+    expect(f.submissions).toEqual([]);
+    expect(result).toMatchObject({ ok: false });
+    expect(JSON.stringify(result)).toContain("draft_input_changed");
+  });
+
+  it("checks the activity again after the final read: activity that appears at that read holds the paste", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    let last: number | null = Date.now() - 60_000;
+    setActivity(f, () => last);
+    const real = f.tmux.captureComposerSnapshot.getMockImplementation()!;
+    let armed = false;
+    f.hooks.settle = () => { armed = true; };
+    // The composer still reads empty at the final read, but a writable client has activity right then.
+    vi.spyOn(f.tmux, "captureComposerSnapshot").mockImplementation(async (name: string) => {
+      const shot = await real(name);
+      if (armed) { armed = false; last = Date.now(); }
+      return shot;
+    });
+    const result = await f.send("activity-at-final-read", notice, { openrigNotice: true, verify: false });
+    expect(result).toMatchObject({ outcome: "retained", sent: false, reason: "draft_input_busy" });
+    expect(pasted(f)).toEqual([]);
+  });
+
+  it("reads tmux's real list-clients output end to end: an attach in this second holds, a quiet client delivers", async () => {
+    const f = fixture(undefined, { runtime: "claude-code", appendPaste: true });
+    await f.enable();
+    const now = () => Math.floor(Date.now() / 1000);
+    f.hooks.clients = () => `1|${now()}\n0|${now()}\n`;   // a viewer tile and a writable client attached this second
+    const held = await f.send("attach-now", notice, { openrigNotice: true, verify: false });
+    expect(held).toMatchObject({ outcome: "retained", sent: false, reason: "draft_input_busy" });
+    expect(pasted(f)).toEqual([]);
+    f.hooks.clients = () => `1|${now()}\n0|${now() - 10}\n`;   // only the read-only tile is active
+    f.advance();
+    const sent = await f.send("attach-quiet", "Wake timer fired again.", { openrigNotice: true, verify: false });
+    expect(sent).toMatchObject({ ok: true });
+  });
+});

packages/daemon/test/tmux-adapter.test.ts

@@ -51,6 +51,47 @@ describe("TmuxAdapter.listSessionClientTerminals", () => {
   });
 });
 
+describe("TmuxAdapter.lastClientActivity (#3076)", () => {
+  const activity = (stdout: string) => adapterFor(mockExec({ "list-clients": { stdout } })).lastClientActivity("op@kernel");
+
+  it("is the latest activity of a writable client, in milliseconds", async () => {
+    expect(await activity("0|1790000100\n0|1790000105\n")).toBe(1790000105000);
+  });
+
+  it("does not count a read-only client (a viewer tile)", async () => {
+    expect(await activity("1|1790000200\n0|1790000100\n")).toBe(1790000100000);
+    expect(await activity("1|1790000200\n")).toBeNull();
+  });
+
+  it("counts an attach as recent activity: a key in the same second as the attach can't be told apart from it", async () => {
+    // tmux would report client_created == client_activity for both an attach alone and an attach plus a key that second.
+    expect(await activity("0|1790000300\n")).toBe(1790000300000);
+  });
+
+  it("is null with no client", async () => {
+    expect(await activity("")).toBeNull();
+    expect(await activity("\n")).toBeNull();
+  });
+
+  it("throws on a row it can't read, alone or next to a readable one, so the guard holds", async () => {
+    for (const stdout of ["0|abc\n", "0|0\n", "0\n", "2|1790000100\n", "0|1790000000|1790000100\n", "0|1790000100\n0|abc\n"]) {
+      await expect(activity(stdout)).rejects.toThrow();
+    }
+  });
+
+  it("throws when tmux can't say", async () => {
+    for (const error of [new Error("can't find session: op@kernel"), NO_SERVER_ERROR, new Error("command timed out")]) {
+      await expect(adapterFor(mockExec({ "list-clients": { error } })).lastClientActivity("op@kernel")).rejects.toThrow();
+    }
+  });
+
+  it("asks tmux for the read-only flag and the activity of each client of the session", async () => {
+    const commands: string[] = [];
+    await adapterFor(async command => { commands.push(command); return ""; }).lastClientActivity("op@kernel");
+    expect(commands.some(c => c.includes("list-clients") && c.includes("#{client_readonly}|#{client_activity}"))).toBe(true);
+  });
+});
+
 describe("TmuxAdapter", () => {
   it("starts only an empty terminal server and reconciles repeated requests", async () => {
     let live = false;

packages/daemon/test/tmux-argv-legacy-equivalence.test.ts

@@ -158,6 +158,7 @@ async function driveEveryMethod(): Promise<RunCall[]> {
   await a.readPaneLastActivity(SESSION);
   await a.listClients();
   await a.listSessionClientTerminals(SESSION);
+  await a.lastClientActivity(SESSION);  // #3076: the client activity read right before an automatic paste
   await a.switchClient("/dev/ttys001", SESSION);
   // A runtime switch's exchange: two session-id reads, then the guarded if-shell swap (nested quoting); and its env writes.
   await a.swapRuntimeSwitchPanes({ nodeId: "n1", stagingSession: "staging'q a", canonicalSession: SESSION, canonicalPane: "%1", stagedPane: "%1" });

packages/tui/src/connected-assistants.ts

@@ -10,7 +10,7 @@ export function toConnectedAssistants(body: unknown): ConnectedAssistant[] | nul
   if (!Array.isArray(members)) return null;
   return members.flatMap((member: Record<string, unknown>) => {
     if (typeof member?.seat !== "string" || !member.seat) return [];
-    const known = typeof member.status === "string" && member.status in CHECK_IN_DISPLAY;
+    const known = typeof member.status === "string" && Object.hasOwn(CHECK_IN_DISPLAY, member.status);
     return [{ seat: member.seat, checkIn: {
       status: known ? member.status as CheckInStatus : "unknown",
       lastCheckInAt: typeof member.lastCheckInAt === "string" ? member.lastCheckInAt : null,

packages/tui/src/render.ts

@@ -958,7 +958,11 @@ function contentLines(state: ViewState, snap: FleetSnapshot, contentWidth: numbe
       { text: `Its seats aren't read in this view: rig ps --host ${rig.remoteHost} --nodes --rig ${rig.name}` },
     ];
     if (rig.inventoryNotLoaded) return [{ text: `Reading ${rig.name}…` }];
-    if (rig.inventoryUnavailable) return [{ text: `Inventory unavailable for ${rig.name} · refresh to Retry` }];
+    if (rig.inventoryUnavailable) return [
+      { text: `Inventory unavailable for ${rig.name} · refresh to Retry` },
+      ...(leaf?.kind === "pod" ? [] : connectedAssistantLines(host.connectedAssistants, host.connectedAssistantsUnavailable,
+        rig.name, state.filter, contentWidth, state.timeZone)),
+    ];
     const podFilter = leaf?.kind === "pod" ? leaf.name : null;
     const all = rig.pods.flatMap((p) => p.agents.map((a) => ({ pod: p.name, ...a })));
     const rows = all

packages/tui/test/hydrate.test.ts

@@ -782,6 +782,34 @@ describe("snapshot hydration over the §4.A reads (Phase 2)", () => {
     expect(instanceLines).not.toContain("not shown");
   });
 
+  it("shows inherited-property check-in statuses as unknown", async () => {
+    const hosted = { members: ["constructor", "toString"].map((status, i) => ({
+      seat: `guest${i}@myrig`, status, reason: "unrecognized reason",
+    })) };
+    const snap = await hydrateSnapshot(fixtureClient({}, { "/api/queue/hosted-members": hosted }));
+    const view = createViewState({ instanceId: "t", getSnapshot: () => snap });
+    view.dispatch({ type: "drill", resource: "rig", name: "myrig" });
+    const lines = renderScreen(view.get(), snap, { cols: 160, rows: 60 }).lines.join("\n");
+    expect(lines).toContain("guest0@myrig  unknown");
+    expect(lines).toContain("guest1@myrig  unknown");
+    expect(lines).not.toContain("unrecognized reason");
+  });
+
+  it("shows connected check-ins beside a failed rig nodes read", async () => {
+    const hosted = { members: [
+      { seat: "chat@myrig", status: "on-request", lastCheckInAt: null },
+      { seat: "other@downrig", status: "checking-in", lastCheckInAt: null },
+    ] };
+    const snap = await hydrateSnapshot(fixtureClient({ "/api/rigs/01JRIG/nodes": { status: 500 } },
+      { "/api/queue/hosted-members": hosted }));
+    const view = createViewState({ instanceId: "t", getSnapshot: () => snap });
+    view.dispatch({ type: "drill", resource: "rig", name: "myrig" });
+    const lines = renderScreen(view.get(), snap, { cols: 160, rows: 60 }).lines.join("\n");
+    expect(lines).toContain("Inventory unavailable for myrig");
+    expect(lines).toContain("chat@myrig  on request:");
+    expect(lines).not.toContain("other@downrig");
+  });
+
   it("a known-empty list of connected assistants stays quiet", async () => {
     const snap = await hydrateSnapshot(fixtureClient());
     expect(snap.hosts[0]!.connectedAssistants).toEqual([]);

packages/ui/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@openrig/ui",
-  "version": "0.7.14",
+  "version": "0.7.1",
   "private": true,
   "license": "Apache-2.0",
   "type": "module",

scripts/fixtures/packaged-plugins-last-release.json

@@ -1,14 +1,14 @@
 {
-  "release": "0.7.13",
-  "archiveSha256": "534e9d5d626368e585d1ba19df20df7d97577153aabef61b9fca68a05edc86e8",
+  "release": "0.7.12",
+  "archiveSha256": "e90b2487fb744d7ca69e52e13b2aaef2a204fcfa76db00c373248acc0795eb52",
   "plugins": {
     "openrig": {
-      "version": "0.7.13",
-      "sha256": "62a5c308ec64b77d678b4276fc81dce933a7ef610da93790bc6a30a3dc3be990"
+      "version": "0.1.0",
+      "sha256": "ee8a1300119576f36889cc6494988c69faa3dfb2c3172cfd856a7ecc3edc0506"
     },
     "openrig-core": {
-      "version": "0.1.29",
-      "sha256": "5e08deffe33feed6f7beff6df39d70402248632f87e3b63f074835b400fff54d"
+      "version": "0.1.28",
+      "sha256": "59b2fb34be931a423583283696c22034fa417530ea682268a68156a70730fb5f"
     }
   }
 }