feat(console): Fleet Room composer — @-address a member

[mabry1985]

Follow-up to #2129 (Fleet Room).

Fixes the regression

In the DM rework the composer became broadcast-only, so you could no longer address a specific member from the input. This restores it:

• @ in the composer → a member picker → an address chip (@name).
• Send to a member opens its DM with the message pre-sent — the wired chat streams the reply, so you see the response (not a fire-into-the-void).
• Broadcast stays the default (no chip, or ⌘↵).

Row-click still opens a member's DM directly; this just adds the composer path back.

Changes

• FleetRoom: @-mention autocomplete + address-target chip + send routing (member → its DM with the message; else broadcast). Footer updated.
• PaletteChat: optional initial prop auto-sends once on open; memberDmView threads it from ctx.props.
• e2e: @-address a member → send → its DM opens.

Testing

tsc ✓ · vitest 628 ✓ · build ✓ · e2e fleet 14 (incl. new @-address test) + keybindings 7 ✓.

Draft

UI — leaving as draft for the feel. The activity feed is still real-events-only (presence + turn.usage); wiring the GitHub/HITL sources for the richer PR/approval rows is separate.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LyQuxgkFpMUJAQbVXnqz2d

---

Update — event sources are wired + fleet controls folded in (2026-07-22)

8e0d44a: the two missing feed sources are real end to end — the executor fires the progress hook on a HITL pause and _a2a_progress publishes turn.input_required for every context (console: "⚠ needs your approval" feed row + roster pill); GitHub github.pr.opened/merged rows land via the github-plugin's emit (separate plugin PR).

2961e3e:

• HITL resume hand-back: turn_started frames carry resumed; the server publishes turn.resumed, so the pill flips needs-approval → running the moment the answer lands. 8 new tests cover the publisher + executor frames (both were untested).
• Folded Feature: Quick-chat with any fleet agent via desktop command palette #1733 quick-chat + Feature: Toggle fleet agents on/off from the command palette #1769 toggle into the room: per-member root commands and the Toggle submorph are gone; member names ride the Fleet Room command's keywords (⌘K "ava" → the room), roster rows carry DM / open / Start / Stop.
• DM header names the member ("@ava") — closes the known rough edge.
• e2e: room-based toggle coverage, fold pins, and a full HITL pill-lifecycle walk over a slug-gated mock stream.

Gates on 2961e3e: pytest 3956 ✓ · ruff ✓ · lint-imports ✓ · live-smoke ✓ · tsc/build ✓ · vitest 618 ✓ · e2e fleet 15 + keybindings 7 ✓.
QA: :7882 restarted on this build — the publisher is live, so a real member approval will light the pill.

[coderabbitai]

Warning

Review limit reached

@mabry1985, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 45 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

Run ID: fec860d1-8165-4875-8b5e-6a6c622a7021

📥 Commits

Reviewing files that changed from the base of the PR and between 18b3c88 and 7b1fbb2.

📒 Files selected for processing (14)

• a2a_impl/executor.py
• apps/web/e2e/fleet.spec.ts
• apps/web/e2e/mock-server.mjs
• apps/web/src/app/FleetActivity.tsx
• apps/web/src/app/FleetRoom.tsx
• apps/web/src/app/PaletteChat.tsx
• apps/web/src/app/fleet-activity.css
• apps/web/src/app/fleet-room.css
• apps/web/src/app/fleetPalette.test.ts
• apps/web/src/app/fleetPalette.ts
• apps/web/src/app/usePaletteRegistry.ts
• server/a2a.py
• tests/test_a2a_handler.py
• tests/test_background.py

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/fleet-room-polish

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[protoreview]

QA panel review — FAIL

code-review-structural · head 7b1fbb23e205 · formal

[review-synthesizer completed: workflow code-review-structural:report]

Verification Pass — Final Synthesis

What changed: Finding 6 was fully refuted — the verifier confirmed mapTopic does have a case "turn.usage" entry, so the comment is accurate. Finding 5's sub-claim about "10 unit tests dropped from fleetPalette.test.ts" was also refuted (this PR doesn't touch that file), but its core claim — zero unit tests for the six new state-mutation methods — is confirmed. Findings 1–4 are all confirmed as stated.

Coverage cross-check: FleetActivity.tsx (the most heavily-changed file, ~150 lines of new state-machine logic) has four surviving findings across three categories — good coverage. FleetRoom.tsx has one confirmed finding. No silent files.

---

Prose brief:

This PR replaces per-component useState tracking with a centralized Zustand store for Fleet turn state. The change is moderate risk: the cleanup-loop bug (finding 1) can leak stale "needs approval" state across remounts, and the SSE-gap counter drift (finding 2) can silently overcount. Fix finding 1 first — it's the only major and the likeliest to surface in normal usage (a turn parked in input_required at unmount). The panel disagreed on finding 6 (refuted — the turn.usage comment was correct all along), and the verifier qualified finding 5 by noting the "10 dropped tests" sub-claim isn't from this PR's diff. No structural-pass gaps to report.

[
  {
    "file": "apps/web/src/app/FleetActivity.tsx",
    "line": 310,
    "severity": "major",
    "category": "correctness",
    "claim": "The cleanup useEffect only clears zustand `running` entries on unmount, missing `awaiting`-only entries — causing stale 'needs approval' pills to persist across remounts when a turn was parked at unmount time and completed while the component was unmounted.",
    "evidence": "The loop `for (const slug of Object.keys(st.running)) st.clear(slug)` iterates only `running` keys. When a turn is in `input_required` state, `markDone` already deleted the slug from `running` and `markAwaiting` added it to `awaiting`. The `clear` method clears BOTH maps, but the loop never reaches the slug because it's no longer in `running`. On remount, `awaiting[slug]` remains `true` with no future frame to clear it if the turn already completed.",
    "source": "removed-behavior",
    "verdict": "confirmed",
    "note": "Cleanup at lines ~306-310 iterates only `Object.keys(st.running)`. A slug that was in `input_required` at unmount (markDone removed it from `running`, markAwaiting added to `awaiting`) is never passed to `clear()`, so its `awaiting[slug]` survives unmount."
  },
  {
    "file": "apps/web/src/app/FleetActivity.tsx",
    "line": 245,
    "severity": "minor",
    "category": "correctness",
    "claim": "If an SSE reconnection gap causes the client to miss a `turn.input_required` frame but receive the subsequent `turn.resumed` frame, the running counter permanently overcounts by +1, leaving a stuck 'running' pill on the roster.",
    "evidence": "The state machine decrements on `turn.input_required` (`markDone`) and increments on `turn.resumed` (`markRunning`), balancing only when both arrive. If `input_required` is dropped: running = started(+1) → [missed input_required] → resumed(+1) → usage(−1) = 1 stuck. `markDone` clamps at 0 but nothing clamps the overflow from an unmatched `turn.resumed`.",
    "source": "correctness",
    "verdict": "confirmed",
    "note": "`turn.resumed` handler unconditionally calls `markRunning(slug)` (increment), but `turn.input_required` is the only path that decrements to offset it. A missed `input_required` means an unmatched +1. `markDone` clamps at 0 but `markRunning` has no ceiling."
  },
  {
    "file": "apps/web/src/app/FleetActivity.tsx",
    "line": 54,
    "severity": "minor",
    "category": "tests",
    "claim": "The PR adds six state-mutation methods to the FleetActivity zustand store (markRunning, markDone, clear, markAwaiting, clearAwaiting) with non-trivial ref-counting, clamping, and bidirectional state transitions — but includes zero unit tests for any of them, while 10 unit tests for removed functions were dropped from fleetPalette.test.ts.",
    "evidence": "`markDone` clamps: `const n = Math.max(0, (s.running[slug] ?? 0) - 1);` — the comment in useFleetStreams says 'a missed pause/resume still settles at zero (markDone clamps)' but this edge case is never unit-tested. Only e2e happy-path coverage exists.",
    "source": "tests",
    "verdict": "confirmed",
    "note": "Zero unit tests for the six new methods — confirmed (diff adds only e2e tests). The '10 unit tests dropped from fleetPalette.test.ts' sub-claim is refuted: this PR's diff does NOT touch fleetPalette.test.ts; the current file has 2 tests, but the reduction is not attributable to this PR."
  },
  {
    "file": "apps/web/src/app/FleetActivity.tsx",
    "line": 100,
    "severity": "minor",
    "category": "conventions",
    "claim": "The exported `clearMemberRunning` name is misleading: it calls `clear(slug)` which wipes BOTH `running` and `awaiting` state, not just running — contrary to its name and JSDoc ('Reset a member's in-flight count entirely').",
    "evidence": "`export const clearMemberRunning = (slug: string) => useFleetActivity.getState().clear(slug);` — but `clear()` deletes both `running[slug]` AND `awaiting[slug]`, while a separate `clearAwaiting` export already exists for the awaiting half.",
    "source": "conventions",
    "verdict": "confirmed",
    "note": "`clearMemberRunning` calls `clear()` which deletes both `running[slug]` and `awaiting[slug]`. The name and JSDoc ('Reset a member's in-flight count entirely') don't mention the awaiting side, and a separate `clearAwaiting` export exists. Naming is indeed misleading."
  },
  {
    "file": "apps/web/src/app/FleetRoom.tsx",
    "line": 174,
    "severity": "minor",
    "category": "correctness",
    "claim": "When `target` points to a member who has left the roster, `submit()` resets to broadcast via `setTarget('broadcast')` and returns — but the composed `draft` may still contain a stripped `@name` prefix from a prior `pickMention` call, and the next Enter press will broadcast text that was originally addressed to someone else.",
    "evidence": "User types `@ava ship it`, presses Enter (picks ava, draft becomes `ship it`), then ava leaves the roster before the second Enter: `targetAgent` goes undefined, `target !== 'broadcast'` is true, the error toast fires, `setTarget('broadcast')` runs, and `submit` returns. On the next Enter, `resolveTypedMention(raw)` finds no agent (draft is `ship it` — no `@` prefix), falls through to `broadcast(raw)`, sending `ship it` to everyone.",
    "source": "correctness",
    "verdict": "confirmed",
    "note": "After `setTarget('broadcast')` the draft still holds `ship it`. Next Enter: target IS 'broadcast' so the error-guard at `!targetAgent && target !== 'broadcast'` no longer fires. `resolveTypedMention('ship it')` finds no `@` prefix, no agent → falls through to `broadcast(raw)`. Text that was meant for ava gets broadcast to all."
  }
]