[orchestration] Clear stale active turn when session becomes inactive

[Andrew-Forster]

Summary

This clears stale activeTurnId state when a provider reports that a session is inactive.

Concretely, session.state.changed now clears activeTurnId when it maps to:

• ready
• interrupted
• stopped
• error

It still preserves activeTurnId for starting and running.

Root Cause

The projection layer could previously write an impossible session state:

session.status = ready
session.activeTurnId = <old turn id>

That is inconsistent. A session that is ready, interrupted, stopped, or error should not also point at an active turn.

Codex exposes one concrete path into this state because T3 maps Codex session/ready into session.state.changed: ready, which is separate from turn.completed. This PR does not assume Codex normally omits turn/completed; the upstream app-server docs describe turn/completed as the normal terminal turn event. The fix is about making T3's projection invariant correct whenever an inactive session-state event is processed.

Why This Matters

A stale active turn can make the UI/sidebar keep treating old work as active after the provider has settled.

It can also affect cleanup. ProviderSessionReaper skips sessions when the projected thread still has activeTurnId != null, so preserving stale active-turn state can make a settled provider session look unreapable.

This does not try to solve every provider desync case. It fixes one small shared invariant: inactive sessions should not retain active turns.

Scope / Tradeoff

Low risk: this only changes session.state.changed handling for inactive statuses.

Shared orchestration fix: the behavior is not Codex-specific. Codex is just the clearest known path because of its separate session/ready event.

Preserved behavior: mid-turn session.started / thread.started updates still keep the active turn. There is an existing regression test for that path, and this PR keeps it passing.

Validation

• pnpm --filter t3 exec vp test run src/orchestration/Layers/ProviderRuntimeIngestion.test.ts -t 'clears active turn when provider session becomes ready' --reporter=verbose
• pnpm --filter t3 exec vp test run src/orchestration/Layers/ProviderRuntimeIngestion.test.ts -t 'does not clear active turn when session/thread started arrives mid-turn' --reporter=verbose
• pnpm --filter t3 exec vp run typecheck
• pnpm exec vp check (existing unrelated warnings only)
• git diff --check

References

• T3 maps Codex session/ready to session.state.changed: ready in CodexAdapter
• ProviderSessionReaper skips cleanup while projected activeTurnId is non-null
• OpenAI Codex app-server docs describe turn/completed as the normal terminal turn event: https://github.com/openai/codex/blob/main/codex-rs/app-server/README.md#lifecycle-overview

Related

• Fixes [Bug]: T3 code not showing model output #2695
• Fixes [Bug]: Codex provider drains plan credits while T3 Code is idle in background (Codex equivalent of #2191) #2720
• Fixes reconcile provider session state and settle stuck turns #2666
• Fixes Reconcile stopped provider session projections #3106

---

Note

Low Risk
Narrow projection invariant fix with targeted tests; only affects session.state.changed handling for inactive statuses.

Overview
Fixes inconsistent projected session state where ready (or other inactive statuses) could still carry a non-null activeTurnId—for example when Codex emits session.state.changed: ready separately from turn.completed.

ProviderRuntimeIngestion now derives session status first, then sets nextActiveTurnId via sessionStatusAllowsActiveTurn: only starting and running keep the previous active turn on session.state.changed; ready, interrupted, stopped, and error clear it. Mid-turn session.started / thread.started behavior is unchanged.

Adds a regression test for ready after turn.started, and bumps waitForThread timeouts in the existing mid-turn lifecycle test.

^{Reviewed by Cursor Bugbot for commit 88d5656. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Clear stale activeTurnId when provider session transitions to an inactive state

• Adds a sessionStatusAllowsActiveTurn helper in ProviderRuntimeIngestion.ts that returns true only for starting or running statuses.
• When a session.state.changed event produces a non-running/non-starting status (e.g. ready, error, stopped), session.activeTurnId is now set to null, preventing stale active turns from persisting.
• Adds a test case in ProviderRuntimeIngestion.test.ts asserting that activeTurnId is cleared when the session returns to ready.

^{Macroscope summarized 88d5656.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: b2a3de97-2e79-4222-bcd0-2c5dfa9e3198

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}

[macroscopeapp]

Approvability

Verdict: Approved

Straightforward bug fix that clears stale activeTurnId when a provider session transitions to an inactive state. The change is small, well-tested, and has no security or schema implications.

^{You can customize Macroscope's approvability policy. Learn more.}