fix: stop the busy indicator looping forever after a mid-turn model switch

[saucam]

The bug

The web "thinking…/drafting…" indicator sometimes loops forever after the model's response has finished. Root cause is daemon-side and deterministic: it happens whenever you run /model or /rotate while a turn is in flight.

#teardownProvider (src/daemon/session.ts) nulls #activeRun before draining the event consumer, so the consumer's finally skips its idle reset (#activeRun === run is now false). #setStatus is the only thing that broadcasts session.status_change, so #status stays thinking/tool_running forever and setModel even re-broadcasts the stale status via #broadcastInfoUpdate. The indicator (WorkerIndicator.tsx, gated purely on status) loops indefinitely.

Fix

• Daemon (primary): #teardownProvider emits the idle transition itself when it tore down an active run and nothing took over — guarded against the onRecoveryNeeded recovery-run path and the terminal error state, so it's a no-op for destroy() and safe for all callers.
• Web (defensive): WorkerIndicator adds a staleness cap — if a session claims thinking but produces no activity for 90s and no tool is executing, the indicator clears. The daemon fix is the real fix; this guarantees the UI can never loop forever again even if a status update is ever missed.

Test

src/tests/session-integration.test.ts — a mid-turn setModel now broadcasts status_change: idle and the session lands at idle. Without the daemon fix, waitForIdle times out.

Full daemon suite green (905), web typecheck + lint clean.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes
  • Improved session status handling during model switches so the UI returns to idle correctly after an interrupted turn.
  • Prevented the worker/activity indicator from staying visible indefinitely when session activity has stalled.
• Tests
  • Added integration coverage for mid-turn provider shutdowns to verify the idle status is broadcast as expected.

[coderabbitai]

Warning

Review limit reached

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

Next review available in: 27 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: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: a0746f65-729c-4ddb-b9d7-10a76a150cd9

📥 Commits

Reviewing files that changed from the base of the PR and between 8a3b560 and bf52451.

📒 Files selected for processing (2)

• web/src/components/transcript/WorkerIndicator.tsx
• web/src/state/messages.ts

📝 Walkthrough

Walkthrough

Session teardown now explicitly transitions status to idle when a mid-turn provider teardown leaves no active run and the session isn't in error, replacing implicit finally-path handling. A new integration test verifies this. Separately, WorkerIndicator visibility now hides after a stall timeout unless a tool is actively running.

Changes

Session Teardown Idle Status

Layer / File(s)	Summary
Teardown sequencing and idle emission src/daemon/session.ts	#teardownProvider records prior active-run state, clears run/consumer task before awaiting provider.teardown(), then explicitly calls #setStatus("idle") when no replacement run exists and status isn't error.
Integration test for teardown idle broadcast src/tests/session-integration.test.ts	New test starts an approval-blocked tool turn, calls setModel mid-turn to tear down the provider, and asserts the session becomes idle with an "idle" status_change broadcast.

WorkerIndicator Visibility Update

Layer / File(s)	Summary
Stall threshold and staleness-aware visibility web/src/components/transcript/WorkerIndicator.tsx	Adds STALL_MS constant, tracks lastActiveAt via createEffect on focusedSessionId/status, and rewrites visible to hide stale "thinking/tool_running" states unless liveTool() is active.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Test as Integration Test
  participant Session
  participant Provider

  Test->>Session: setModel(...) mid-turn
  Session->>Provider: teardown()
  Session->>Session: clear activeRun / eventConsumerTask
  Provider-->>Session: teardown complete
  Session->>Session: setStatus("idle") if no active run & no error
  Session-->>Test: status_change broadcast ("idle")

Loading

Possibly related PRs

• saucam/codeoid#105: Both PRs modify WorkerIndicator.tsx visibility gating logic based on session status.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main user-facing fix: preventing the busy indicator from getting stuck after a mid-turn model switch.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/status-idle-indicator

---

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.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 77.83%. Comparing base (7855889) to head (bf52451).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #125      +/-   ##
==========================================
+ Coverage   77.10%   77.83%   +0.73%     
==========================================
  Files          89       89              
  Lines       14412    14385      -27     
==========================================
+ Hits        11112    11197      +85     
+ Misses       3300     3188     -112     

Flag	Coverage Δ
daemon	77.83% <100.00%> (+0.73%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/daemon/session.ts	82.62% <100.00%> (+6.50%)	⬆️

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@web/src/components/transcript/WorkerIndicator.tsx`:
- Around line 39-44: The fallback timing in WorkerIndicator is using local
component time, which makes stale sessions look fresh again on mount or refocus.
Replace the Date.now()-based lastActiveAt state and the createEffect reset with
a per-session activity timestamp sourced from the store/session model, and make
the indicator derive its stale-window from that persisted value instead of
reinitializing on focusedSessionId() or status() changes. Ensure the logic in
WorkerIndicator keeps the existing stale state across navigation/reload rather
than restarting the 90s timer.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ad56bbc6-eb92-4714-ade2-6139817f59c7

📥 Commits

Reviewing files that changed from the base of the PR and between 7855889 and 8a3b560.

📒 Files selected for processing (3)

• src/daemon/session.ts
• src/tests/session-integration.test.ts
• web/src/components/transcript/WorkerIndicator.tsx