fix(runner): finish stopped turns before running steered input

[mmabrouk]

Context

Steer could cancel a running tool and then lose the follow-up with “This session is already running a turn.” The runner reported Stop completion immediately after issuing the abort, allowing the API to promote the saved input while the original environment was still busy.

Changes

Keep the immediate HTTP acknowledgement, but report the stopped outcome after the execution finishes teardown and releases ownership. Duplicate deliveries wait for that same outcome. An abandoned execution reports failure rather than allowing Steer to enter an environment that may still be busy.

Tests

• Reproduced with a real running shell: cancellation succeeded, but the promoted input failed before the original turn ended.
• Repeated that exact sequence with the fix: correct execution cancelled, follow-up started after its predecessor ended, one saved agent reply, no error records, empty queue and idle session.
• Control/server/timeout regressions: 91 passed, including immediate abort, delayed outcome, duplicate delivery, abandonment, abort failure and execution identity.
• Full runner suite initially passed 2,902 tests; four failures needed generated tool-shim assets, and an existing pause-teardown test needed to release its mocked teardown before awaiting the final outcome. Both affected suites then passed all 91 tests. Strict typecheck passed.

What to QA

Start a long shell operation, use Steer while the tool is running, and verify the original stops and the follow-up runs once. Check ordinary Stop still acknowledges immediately and keeps the next message usable.

Stacked on #6591 for release/v0.115.2.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
agenta-documentation	Ready	Preview	Sep 5, 2026 11:46pm UTC

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository YAML (base), Organization UI (inherited)

Review profile: CHILL

Plan: Team

Run ID: ff971e9b-9291-4766-998c-05d652de5728

📥 Commits

Reviewing files that changed from the base of the PR and between a6f2795 and 20cd767.

📒 Files selected for processing (6)

• services/runner/src/server.ts
• services/runner/src/sessions/applied-commands.ts
• services/runner/src/sessions/control-channel.ts
• services/runner/src/sessions/execution-registry.ts
• services/runner/tests/unit/control-command-apply.test.ts
• services/runner/tests/unit/sandbox-agent-orchestration.test.ts

Included review availability: Your plan provides up to 8 included reviews per hour; 6 remain after this review.

---

📝 Summary

Summary by CodeRabbit

• Bug Fixes
  • Stop requests now wait for the active run to finish releasing resources before reporting the final outcome.
  • Duplicate stop requests remain synchronized with the original request and receive the correct final status.
  • Abandoned runs now report a failed outcome instead of leaving requests unresolved.
  • Stop handling remains reliable even when cleanup encounters an error or when a newer run starts.

Walkthrough

The runner now tracks teardown completion, exposes per-execution release signals, and delays Stop outcomes until teardown settles. Duplicate commands await the original command. Tests cover normal, abandoned, failed-abort, and successor-execution cases.

Changes

Stop teardown settlement

Layer / File(s)	Summary
Execution release tracking services/runner/src/sessions/applied-commands.ts, services/runner/src/sessions/execution-registry.ts	Commands now expose settlement promises. The execution registry resolves per-run release signals during unregistration and test reset.
Command settlement after teardown services/runner/src/sessions/control-channel.ts, services/runner/tests/unit/control-command-apply.test.ts	Stop handling waits for environment release before reporting. Duplicate, abandoned, abort-failure, and successor-execution cases are covered.
Runner teardown status integration services/runner/src/server.ts, services/runner/tests/unit/sandbox-agent-orchestration.test.ts	The streaming runner passes teardown completion to unregistration. Orchestration tests await the Stop outcome after unregistration.

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

Merge Risk: ⚪ Minimal · up to 20cd7

Stop outcomes now wait for execution teardown and ownership release, preventing follow-up input from starting against a busy environment. Covered cancellation and duplicate-delivery paths show no remaining merge-blocking risk.

Sequence Diagram(s)

sequenceDiagram
  participant Client
  participant ControlChannel
  participant ExecutionRegistry
  participant Runner
  Client->>ControlChannel: applyCommand(Stop)
  ControlChannel->>Runner: abort()
  Runner->>ExecutionRegistry: unregisterExecution(..., teardownCompleted)
  ExecutionRegistry-->>ControlChannel: release status
  ControlChannel-->>Client: settled outcome

Loading

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	Docstring coverage is 80.00% which is sufficient. The required threshold is 60.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 5 functions across 6 files.
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.
Title check	✅ Passed	The title clearly identifies the main change: completing stopped turns before processing steered input.
Description check	✅ Passed	The description directly explains the teardown ordering fix, duplicate synchronization, abandonment behavior, tests, and QA steps.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/release-1152-steer-settlement

---

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.}

[mmabrouk]

@coderabbitai full review

Please review the execution-release fence, duplicate outcome ordering, and abandonment safety. The live running-Steer regression now passes; the original reproduced failure and targeted runner regressions guided this change.

Requested by Codex (OpenAI), acting for Mahmoud.

[coderabbitai]

@mmabrouk I will perform a full review of #6592. I will focus on execution-release ordering, duplicate outcome synchronization, abandonment handling, and the stated runner regressions.

✅ Action performed

Full review finished.

[github-actions]

Railway Preview Environment

Status	Destroyed (PR closed)

Updated at 2026-09-06T02:47:22.481Z