[juliusmarminge]

Problem

Turning on Resume after usage limits made the usage-limit recovery sweep arm and resume delegated child threads that had stopped on a usage limit while the setting was off, even when their parent thread was settled. When a resumed child finished, its delegated-completion wake could pull the settled parent back to active and start a new run. The candidate query only checked the child's own archivedAt and settledOverride and never looked at its lineage.

Reproduced on main (2b15ba0) with a projection-level test. It builds two parent/subagent pairs whose children failed on a usage limit and have nothing armed; one parent is settled. With autoResume on (and separately with snooze on), both children get a limit-arm: command, including the child of the settled parent.

Change

getLimitRecoveryCandidates gets one extra condition in the branch that freshly arms a recovery (both the auto-resume arm and the snooze arm). A thread whose lineage.relationshipToParent is subagent and whose direct parent has settledOverride = 'settled' is not offered for a new arm. The parent lookup goes through the threads table's primary key and only runs for rows that already passed the usage-limit pre-filter.

The resume branch is unchanged: a recovery that was already armed for the child's latest failed run still resumes at its reset. Unsettling the parent emits a thread.* event, which already makes the sweep due, so the child becomes eligible again on the next tick.

Scope and approval

Fixes #27944

Verification

• New test ProjectionStoreV2 > does not freshly arm limit recovery for a subagent of a settled parent in apps/server/src/orchestration-v2/ProjectionStore.test.ts. It checks that a child of a settled parent is not armed under autoResume or snooze, that a child of an unsettled parent still is, and that a child armed before its parent settled still produces its limit-resume: dispatch.
• Red with main's ProjectionStore.ts (2b15ba0):

  × does not freshly arm limit recovery for a subagent of a settled parent
  AssertionError: only the child of the unsettled parent is armed ({"now":"1970-01-01T00:00:00.000Z","autoResume":true,"snooze":false}): expected [ …(2) ] to deeply equal [ Array(1) ]
    actual: [ 'thread:settled-parent-limit:active:child', 'thread:settled-parent-limit:settled:child' ],
    expected: [ 'thread:settled-parent-limit:active:child' ]
  Tests  1 failed | 44 skipped (45)
  

• Green with the fix (003df33): Tests 1 passed | 44 skipped (45).
• ProjectionStore.test.ts and UsageLimitRecoveryWorker.test.ts: 46/46 pass. runtimeLayer.test.ts -t "limit": 20 pass. vp run check:changed passes (format, lint, knip, typecheck, tests).
• Server-only change with no visible surface, so there are no captures. Not checked: a live provider hitting a real usage limit, which a VM cannot trigger. The proof is the projection test above.

Claude Opus 5.5 (Claude Code in T3 Code; reviews by GPT-6.1 Sol on Codex and Claude Opus 5.5).

🤖 Generated with Claude Code

[macroscopeapp]

Approvability

Verdict: Approved at 5abe8a0

Macroscope's review found this PR approvable — Fresh usage-limit recovery will no longer arm subagents of explicitly settled parents, preventing abandoned work from waking those parents. Already-armed recoveries and product preference defaults remain unchanged.

No code changes detected at 8f9e45d. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.6 KiB	21.6 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	−18 B (−0.3%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	−18 B (−1.4%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.9 KiB	21.9 KiB	−41 B (−0.2%)	24.4 KiB	✅
Claude	Live turn messages	2	1	−1 (−50.0%)	8	✅

Baseline: f6503f0 · PR result: 8f9e45d · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

The limit-recovery candidate query now excludes subagent threads whose parent has settledOverride set to "settled". Tests check fresh recovery selection under two option sets and verify that an already-armed recovery can still produce a resume command after its reset time.

Changes: Usage-limit recovery

Layer / File(s)	Summary
Filter fresh recovery candidates apps/server/src/orchestration-v2/ProjectionStore.ts, apps/server/src/orchestration-v2/ProjectionStore.test.ts	The query excludes subagent threads when their parent has settledOverride set to "settled". Tests cover both option sets and confirm that an already-armed recovery can still yield a resume command after the reset time.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix · Severity of issue fixed: Medium

Possibly related issues

Suggested labels: size:M

Suggested reviewers: t3dotgg, macroscopeapp

Merge Risk: Minimal · up to 8f9e4

Fresh recovery selection excludes children of settled parents while preserving already-armed recoveries. No merge-blocking issue was identified.

Pre-merge checks: 3 · 1

Failed checks: 1 warning

Check name	Status	Explanation	Resolution
Linked Issues check		Issue #27944 requires that a child whose direct parent is settled is not resumed by the usage-limit sweep. ProjectionStore.ts excludes the child only when the sweep would create a fresh recovery arm…	Also exclude subagents of settled parents from the already-armed resume path, or cancel their recovery before dispatch. Keep a regression test that settles the parent after arming and verifies that no resume dispatch occurs.

Passed checks: 3 passed

Check name	Status	Explanation
Title check		The title clearly and concisely describes the main server fix: preventing usage-limit auto-resume from freshly arming children of settled parents.
Description check		The description includes all required sections: Problem, Change, Scope and approval, and Verification. It explains the bug, implementation, linked issue, focused tests, observed results, and untested …
Out of Scope Changes check		The changes stay within issue #27944. The production change updates usage-limit candidate selection, and the test covers settled and unsettled parents, auto-resume, snooze, and existing recovery behav…

Full details: Description check

Explanation

The description includes all required sections: Problem, Change, Scope and approval, and Verification. It explains the bug, implementation, linked issue, focused tests, observed results, and untested live-provider scenario.

Full details: Out of Scope Changes check

Explanation

The changes stay within issue #27944. The production change updates usage-limit candidate selection, and the test covers settled and unsettled parents, auto-resume, snooze, and existing recovery behavior.

Full details: Linked Issues check

Explanation

Issue #27944 requires that a child whose direct parent is settled is not resumed by the usage-limit sweep. ProjectionStore.ts excludes the child only when the sweep would create a fresh recovery arm. UsageLimitRecoveryWorker.ts still dispatches limit-resume: when an existing recovery matches the latest failed run, without checking the parent's settlement. The added test confirms this behavior for an already-armed child.

Resolution

Also exclude subagents of settled parents from the already-armed resume path, or cancel their recovery before dispatch. Keep a regression test that settles the parent after arming and verifies that no resume dispatch occurs.

Finishing Touches

Generate docstrings

• Commit to this branch
• Create a new PR

Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Some tools did not complete. Review the errors below.

ast-grep (0.51.1)

ast-grep could not parse rule config: /ast-grep-rules/ast-grep-essentials/rules/ruby/security/insufficient-rsa-key-size-ruby.yml

Usage-based review receipt

• Mode: Continue automatically
• Reviewed files: 2
• Charged: $0.50
• View usage details

Note

This review was completed with usage-based billing: files reviewed beyond your plan's included limits are billed at $0.25/file. View usage-based billing.

CodeRabbit Verify

• Run CodeRabbit Verify to test the changes in this pull request.

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: fix/issue-27944
• File: .coderabbit.config.ts
• Commit: 8f9e45d
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Team
• Run ID: 22161180-013d-447a-a139-ba9923d758f6

Commits

Reviewing files that changed from the base of the PR and between 5abe8a0 and 8f9e45d.

Files selected for processing: 2

• apps/server/src/orchestration-v2/ProjectionStore.test.ts
• apps/server/src/orchestration-v2/ProjectionStore.ts

Limit details: You’ve used all 10 included reviews currently available.