Skip to content

fix(codex): early child completion no longer gets lost - #5723

Open
caezium wants to merge 2 commits into
pingdotgg:mainfrom
caezium:agent/preserve-preregistration-subagent-lifecycle
Open

fix(codex): early child completion no longer gets lost#5723
caezium wants to merge 2 commits into
pingdotgg:mainfrom
caezium:agent/preserve-preregistration-subagent-lifecycle

Conversation

@caezium

@caezium caezium commented Aug 8, 2026

Copy link
Copy Markdown
Contributor

What Changed

  • Retain each unregistered child thread's latest meaningful lifecycle state without creating a child identity from traffic alone.
  • Replay a compact completion, idle, error, or closed settlement after explicit child registration so the UI cannot remain on the registration event's running state.
  • Keep duplicate registration signals settled, while allowing newer active, turn-started, and interrupted lifecycle events to supersede older completion state.
  • Add runtime-stream integration coverage for early completion, idle, and terminal errors, initial idle suppression, lifecycle supersession, duplicate registration, interaction neutrality, and interruption ordering.

Why

Codex can emit a child thread's completion or idle notification before either explicit registration signal reaches T3. The runtime correctly suppresses that foreign child traffic from the parent timeline, but previously discarded it, so the later registration event could leave the child displayed as running indefinitely.

This change buffers only compact lifecycle state and never treats an unknown foreign thread as an agent. It complements #5718 while keeping the runtime ordering fix separate from adapter liveness mapping.

Validation

  • pnpm exec vp test run apps/server/src/provider/Layers/CodexCollabRuntime.integration.test.ts apps/server/src/provider/Layers/CodexCollabWire.test.ts apps/server/src/provider/Layers/CodexSessionRuntime.test.ts — 38 tests passed
  • Targeted vp lint and vp fmt --check passed for both changed files
  • pnpm --filter t3 typecheck passed
  • Standards and correctness reviews reported no findings on the final commit

All added fixture identifiers, paths, and content are deterministic mock data.

Checklist

  • This PR is small and focused
  • I explained what changed and why
  • No UI changes

Model: GPT-5.6 Sol; harness: Codex in T3 Code.

Note

Fix early child completion getting lost before registration in makeCodexSessionRuntime

  • Buffers the latest lifecycle settlement (turn completed, status changed, or closed) for a child that arrives before explicit registration, then replays it immediately after registration is received.
  • Adds a toCollabChildSettlement helper in CodexSessionRuntime.ts to map child notifications to compact synthetic settlement events.
  • Terminal child errors (error with willRetry=false) are now treated as child-scoped events, suppressed from the parent timeline, and used to update lifecycle state.
  • Stale or purely idle pre-registration states are not replayed to avoid false completions after a restart.
  • Adds integration tests in CodexCollabRuntime.integration.test.ts covering preregistration replay, duplicate registration, stale settlement avoidance, and related edge cases.

Macroscope summarized c73f5a8.


Note

Medium Risk
Touches Codex collab notification ordering and synthetic agent liveness in CodexSessionRuntime, where subtle race bugs could mis-report child status; scope is focused and heavily covered by new integration tests.

Overview
Fixes collab child agents staying running when Codex sends completion, idle, error, or close on a child thread before thread/started or subAgentActivity registers it. Foreign child traffic is still kept off the parent timeline; the runtime now keeps only a compact latest lifecycle state per unknown foreign thread (via toCollabChildSettlement and collabChildLifecycleStatesRef) and replays the matching synthetic collabAgent/* event right after explicit registration.

Newer active / turn/started lifecycle wins over a buffered completion; duplicate registration re-emits the latest settlement instead of leaving the UI on “running”; pre-registration idle is ignored when it would look like a finished run without a live turn; interrupted activity and terminal error paths update the buffered state. Integration tests add a shared mock script helper and cover pre-registration completion, error, idle, duplicate registration, restart, interaction, interruption, and superseded completion ordering.

Reviewed by Cursor Bugbot for commit c73f5a8. Bugbot is set up for automated code reviews on this repo. Configure here.

Summary by CodeRabbit

  • Bug Fixes
    • Improved reliability of child-agent status tracking across registration, restarts, interruptions, and idle periods.
    • Preserved completion and error states that occur before a child agent is registered, replaying them when appropriate.
    • Prevented unrelated child-agent notifications from appearing in the parent activity timeline.
    • Improved handling of duplicate registrations, terminal errors, and superseded completions.

@coderabbitai

coderabbitai Bot commented Aug 8, 2026

Copy link
Copy Markdown

Review Change Stack

📝 Walkthrough

Walkthrough

CodexSessionRuntime now retains child-agent lifecycle settlements that arrive before registration. It replays current settlements after registration, preserves interruptions, handles restarts and terminal errors, filters foreign child events, and adds integration coverage for these scenarios.

Changes

Codex child lifecycle

Layer / File(s) Summary
Lifecycle state model
apps/server/src/provider/Layers/CodexSessionRuntime.ts
Defines normalized child settlements and stores lifecycle state per child for later replay.
Event lifecycle integration
apps/server/src/provider/Layers/CodexSessionRuntime.ts
Updates lifecycle state during registration, activity, completion, closure, interruption, errors, and suppressed foreign-child notifications.
Lifecycle regression coverage
apps/server/src/provider/Layers/CodexCollabRuntime.integration.test.ts
Adds reusable fixtures and tests for filtering, replay, duplicate registration, restarts, interruptions, idle states, and superseded completions.

Estimated code review effort: 4 (Complex) | ~45 minutes

Sequence Diagram(s)

sequenceDiagram
  participant ChildAgent
  participant CodexSessionRuntime
  participant ProviderEvent
  ChildAgent->>CodexSessionRuntime: send registration or lifecycle notification
  CodexSessionRuntime->>CodexSessionRuntime: retain or replay child settlement
  CodexSessionRuntime->>ProviderEvent: emit filtered child status or completion
Loading

Possibly related PRs

Suggested reviewers: t3dotgg, juliusmarminge

🚥 Pre-merge checks | ✅ 5
✅ Passed checks (5 passed)
Check name Status Explanation
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.
Title check ✅ Passed The title clearly identifies the primary fix: preserving early Codex child completion events.
Description check ✅ Passed The description explains the change, rationale, validation, scope, and checklist status with sufficient detail.
✨ 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

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

@github-actions github-actions Bot added vouch:unvouched PR author is not yet trusted in the VOUCHED list. size:L 100-499 changed lines (additions + deletions). labels Aug 8, 2026
@caezium
caezium marked this pull request as ready for review August 8, 2026 17:32
@caezium

caezium commented Aug 8, 2026

Copy link
Copy Markdown
Contributor Author

@coderabbitai review

@coderabbitai

coderabbitai Bot commented Aug 8, 2026

Copy link
Copy Markdown
✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

@coderabbitai coderabbitai Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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 `@apps/server/src/provider/Layers/CodexSessionRuntime.ts`:
- Around line 1413-1439: Update the unregistered-child notification handling
around toCollabChildSettlement to explicitly process terminal error
notifications with willRetry false, creating a collabAgent/statusChanged
settlement whose status.type is systemError. Remove the corresponding
foreignThreadId from collabChildLiveTurnsRef and prevent stale active state from
being replayed during registration. Add an integration case covering
turn/started, a non-retryable terminal error, and subsequent child registration.
🪄 Autofix

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: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7493a612-8aa7-4590-87e0-3672c96723e4

📥 Commits

Reviewing files that changed from the base of the PR and between a20923c and 9958174.

📒 Files selected for processing (2)
  • apps/server/src/provider/Layers/CodexCollabRuntime.integration.test.ts
  • apps/server/src/provider/Layers/CodexSessionRuntime.ts

Comment thread apps/server/src/provider/Layers/CodexSessionRuntime.ts
@macroscopeapp

macroscopeapp Bot commented Aug 8, 2026

Copy link
Copy Markdown
Contributor

Approvability

Verdict: Needs human review

This PR adds substantial new state management and event replay mechanisms for multi-agent coordination, beyond a simple bug fix. The changes introduce lifecycle state buffering and replay logic that affect core runtime behavior of the collab system, warranting careful human review.

You can customize Macroscope's approvability policy. Learn more.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

size:L 100-499 changed lines (additions + deletions). vouch:unvouched PR author is not yet trusted in the VOUCHED list.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant