Replace custom media generation routing with native Agent skill execution and update FileCard-style attachment icons, inline audio playback, and image/video previews. Verified with frontend tests and desktop checks.
The direct-conversation branch in promptFor already delivers only the
raw user query (no group framing, no 'Current user task' label, no
history) for private 1:1 chats. The legacy-session-recovery test still
expected the rebuilt prompt to carry prior agent conclusions, which
contradicts the query-only contract and failed on a clean tree. Align
the assertion so session rebuilds also omit prior conclusions.
Run sequential mode strictly in configured CLI order, let auto-beta route later turns from canonical mentions, and persist one native Agent session per task across rounds and restarts. Remove rigid review framing from natural prompts and add recovery, routing, and continuity coverage.
Extract V4 outcome validation helpers from the local auto runner, add recursive desktop test discovery suites, and harden CLI stdin pipe failures.
Checks: npm --prefix desktop test -- --suite workspace --test-name-pattern 'auto runner support|v4'; npm --prefix desktop test
Commit each validated proposal and challenge as soon as its Agent finishes so later failures or stalls cannot hide healthy round output. Show the failing Agent logo on structured failure messages.
Checks: npm --prefix frontend test; node --test desktop/test/workspace/local-workspace-auto.test.cjs; npm --prefix desktop test; npm --prefix frontend run build; npm --prefix frontend run build:desktop; npm --prefix desktop run pack; git diff --check
Keep round-one proposals visible and durable when later agents fail or a run is recovered. Enforce bounded and unlimited round semantics, normalize collaboration roles, and preserve healthy agents after invalid control output.
Update scheduler, failure handling, runtime trace UI, connector compatibility, and focused regression coverage.
Checks: npm --prefix desktop test; npm --prefix frontend test; npm --prefix frontend run build:desktop; git diff --check; node --check on modified workspace runtime files.
Register Pi Agent, treat detected compatible local CLIs as usable when their native configuration is not inspectable, and add collapsible Agents and Groups sections with the official Pi logo.\n\nChecks: npm --prefix frontend test; npm --prefix frontend run build; npm --prefix frontend run build:desktop; node --test test/agents/cli/cli-adapters-discovery.test.cjs test/workspace/local-workspace-agent-catalog.test.cjs
Bump frontend and desktop packages to 0.1.3, update download and distribution documentation, and add user-facing release notes for concurrent group targeting, conversation refinements, and stop durability.
Checks: frontend 306/306; desktop 1338/1338; deterministic eval 6 cases/18 results; web and desktop builds; pack; dist; DMG/ZIP integrity; codesign verification; packaged Electron UI acceptance; git diff --check. The prerelease remains ad-hoc signed and is not notarized.
Treat an empty Concurrent responses selection as the full group while preserving explicit Agent toggles. Remove the special No round limit composer surface so the standard chat styling remains consistent, while keeping the infinity status cue.
Checks: frontend 306/306; desktop 1338/1338; focused composer 26/26; Electron packaged acceptance; git diff --check.
Polish user messages, inline mentions, Markdown rendering, no-limit feedback, and jump-to-latest behavior. Default concurrent responses to every group Agent, raise scheduler capacity for four-member batches, preserve visible manual proposals without valid receipts, and persist completed or streamed replies when the user stops a batch.
Checks: npm --prefix frontend test; npm --prefix frontend run build; npm --prefix frontend run build:desktop; npm --prefix desktop test; npm --prefix desktop run pack; git diff --check. Packaged Electron acceptance covered four Agents running together without mentions and retained stopped partial output across app restart.
Register the SIGTERM observer before publishing the helper readiness marker so CI cannot cancel between those operations. Product cancellation behavior is unchanged.\n\nChecks: focused cancellation test 30/30; frontend 302/302; desktop 1335/1335; deterministic eval 6 cases/18 results; web and desktop builds; desktop pack; git diff --check.
Keep return-to-latest jumps synchronous even when the timeline uses smooth scrolling, and retain the reader's upward position while provisional output becomes durable.
Checks: npm --prefix frontend test; npm --prefix frontend run build; npm --prefix frontend run build:desktop; npm --prefix desktop test; npm --prefix desktop run eval:deterministic; git diff --check
Reuse the LocalWorkspace waiting-run validator during Cloud runtime recovery so a pending local permission Gate is not downgraded on startup. Cover the valid Gate and a forged same-shape Run in the Cloud runtime regression suite.\n\nVerified: frontend 300/300; desktop 1335/1335; deterministic eval 6 cases/18 results; both frontend builds; desktop pack; packaged restart/reject/no-replay acceptance; git diff --check.
Track permission outcome promises per ACP turn and wait for pending handlers through the abort-aware operation barrier before rejecting a terminal response.
Checks: named permission race test 20/20; ACP protocol tests 55/55; node --check; git diff --check.
Document the terminal permission barrier, monotonic checkpoint recovery semantics, and current 132-test focused evidence while keeping full-suite, packaged-app, signing, and release-asset claims pending for the exact final commit.
Checks: 38 documented test paths exist; git diff --check passed.
Reject duplicate active ACP permission request IDs, restore the complete V4 live round on known checkpoint failure, and retain monotonic state when a checkpoint callback has an unknown post-write outcome.
Checks: 132 related ACP, Gate, Human Gate, and Auto durability tests passed; production syntax checks and git diff --check passed.
Track pending permission callbacks across the common invocation and ACP transport, prevent Auto V4 phase advancement while an exact Agent-slot Gate is active, and roll back live orchestration when a checkpoint cannot persist.
Checks: node --test desktop/test/agents/cli/cli-adapters-agent-protocols.test.cjs desktop/test/workspace/local-workspace-v4-gate-pause.test.cjs (71/71); node --check on the three changed production modules; git diff --check.
Persist exact Harness attempt bindings for active V4 slots and private Human Gates, keep retry/recovery transitions crash-durable, and fail closed on stale or colliding identities.\n\nChecks: focused crash windows 2/2; Auto 74/74; exact binding 131/131; recovery/state 109/109; node --check; git diff --check.
Record the latest candidate test counts while marking final packaged Hermes/OpenClaw, V4, stop, mobile UI, archive, checksum, and signing checks as pending until they are rerun from the exact release commit.
Checks: git diff --check; stale SHA and test-count search.
Ryder Sun