[voltcrash]

Problem

A failed thread with a future snooze can reappear as active after a metadata update when its failed run has completedAt: null. The client falls back to runtime.updatedAt and treats that metadata write as a fresh failure, while the server keeps the same thread snoozed.

Found while investigating #30049. The report has no client-state capture, so this fixes a reproducible remaining mismatch without claiming to establish the cause of that incident or close the issue.

Change

Keep the failed run's unknown completion time unknown instead of substituting metadata activity. Known failures after the snooze and pending approvals still wake the thread. The shared classifier covers web, Electron desktop, and mobile; the server predicate and resume scheduler need no changes.

Scope and approval

This qualifies as a small, focused obvious-bug fix: the schema allows a failed run with a null completion timestamp, and the existing server snooze predicate already preserves snooze in that case. Two additional rows extend the existing regression table. No new workflow or product default is introduced.

Verification

• Added the missing-completion regression first: it failed on the original classifier (effectiveSnoozed returned false). All 27 snooze tests pass with the fix, including the added approval-wake case and existing fresh-failure, expiry, and legacy-session cases.
• The baseline snooze, presentation, usage-limit recovery, and server settlement suites passed (79 tests). vp run check:changed passes formatting, lint, knip, affected-package typechecks, and the adjacent tests.
• Linux web: started with an app-created Replay (dev) thread, then seeded a synthetic failed-run read-model fixture in isolated worktree state with null completion, a future snooze, and newer metadata. This is not a real provider usage-limit incident or a coherent failure event history. The original client shows it active with Failed; the patched client keeps that same saved state Snoozed. Clicking the real Wake now action clears both snooze fields.
• Native desktop/mobile and remote/relay/tunnel were not exercised live. Their shared classifier receives this fix. The reporter's installation and exact incident remain unverified.

Before:

After:

Wake now verification recording

Reviewed with review-loop: 1 round with GPT-6.1-Sol on xhigh. Reviewer verdict: approved. No findings; diff stayed at 2 files, +15/-3. Reviewer usage was not returned.

Created with GPT-6.1-Sol in the Codex harness.

[macroscopeapp]

Approvability

Verdict: Approved at 7a34062

Macroscope's review found this PR approvable — Failed runs without completion timestamps now stay snoozed rather than resurfacing after metadata updates. The change preserves timer expiry and user-request wake behavior without altering execution or product defaults.

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

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

The failure edge now preserves a null completion timestamp instead of using the runtime update time. Added test cases check failure and approval wake times when no failure completion timestamp exists.

Changes: Snooze wake-time handling

Layer / File(s)	Summary
Failure edge and wake-time cases packages/client-runtime/src/state/threadSettled.ts, packages/client-runtime/src/state/threadSnoozed.test.ts	When the latest run is failed, failureEdgeAt returns its completedAt value, including null. Tests check that a failure with no completion timestamp has no wake time and that an approval uses the runtime update time.

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

Change: Bug fix

Possibly related issues

• pingdotgg/t3code#30049: This change addresses a similar UI symptom for failed runs with a null completion timestamp, but does not establish or resolve the issue’s cause.

Suggested labels: bugfix, size:XS

Suggested reviewers: rakshithbhat03

Merge Risk: Minimal · up to 7a340

The change correctly keeps incomplete failed runs snoozed while still waking approvals at the runtime update time.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title names the main change: snoozed failures stay hidden when the failed run has no completion time. It is specific, follows the conventional commit form, and matches the raw summary and objectiv…
Description check		The description follows the template. It has Problem, Change, Scope and approval, and Verification sections. The problem and the fix are clearly stated. The scope section argues the small obvious-bug …
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Full details: Title check

Explanation

The title names the main change: snoozed failures stay hidden when the failed run has no completion time. It is specific, follows the conventional commit form, and matches the raw summary and objectives. The 66-character length is acceptable.

Full details: Description check

Explanation

The description follows the template. It has Problem, Change, Scope and approval, and Verification sections. The problem and the fix are clearly stated. The scope section argues the small obvious-bug exemption and links the related issue #30049 as context without claiming it closes it. Verification gives focused test results, a manual Linux web check with a clearly labeled synthetic fixture, and before/after evidence. It states what was not checked (native desktop/mobile and remote/relay/tunnel). The description does not include a conventional commit title line, but the template comment is an HTML comment and the title itself is conventional. The description is mostly complete.

Finishing Touches

Generate unit tests (beta)

• Create a new PR

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

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: main
• File: .coderabbit.config.ts
• Commit: cefb544
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: 1e5a9f11-d7e6-46a1-91ef-beb1c437713d

Commits

Reviewing files that changed from the base of the PR and between cefb544 and 7a34062.

Files selected for processing: 2

• packages/client-runtime/src/state/threadSettled.ts
• packages/client-runtime/src/state/threadSnoozed.test.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 3 remain after this review.

[maria-rcks]

Note

Written by gpt-6.1-sol on behalf of Maria

Verified in a real dev environment

partial verification only. the missing-completion-timestamp regression remains unverified.

ran linux web with repository playwright and /usr/bin/google-chrome, first on github/main (cefb544af3), then the exact pr head 7a3406216fbdf2cfb48c51da1b9ac91572835ec1. only one dev server ran at a time. created a project for a real git repository through the app, then ran real codex gpt-6.1-sol turns. no mocks, replay turns, fixtures, or database writes.

• main: a completed thread stayed snoozed after rename and reload. an actual codex launch failure, induced with an invalid launch argument in provider settings, also stayed snoozed after rename and reload.
• pr: copied that app-created state after stopping main. the same failed thread stayed snoozed, including after another rename and reload. wake now cleared snooze and survived reload. restored valid provider settings and successfully ran another real codex turn in the same thread. a custom 0.2-minute snooze expired and showed "woke from snooze".
• blacksmith: vp run check:changed FETCH_HEAD^ passed formatting, lint, both knip steps, affected typechecks, and all 27 snooze tests. the initial default invocation lacked a merge base in the worker's shallow snapshot; fetching this pr with depth 2 and passing its parent resolved that setup issue. no source changes.

read-only runtime transcript: persisted runs and snooze metadata.

the following matching-thread captures demonstrate the known-timestamp neighboring path, not the bug fixed by this pr.

Not verified

could not reproduce the original bug on main without fabricating state. the actual failed run had completedAt: "2026-11-03T11:57:20.383Z", not null. the inspected codex timestamp conversion substitutes the current time for missing provider timestamps, and the inspected failure handling records completion times. therefore these app checks do not prove the pr's null-completion branch, a real usage-limit incident, or the cause of issue #30049. the passing regression table is separate unit evidence.

native electron, ios/android, other providers, and remote/relay/tunnel connections were not exercised. no code was changed or pushed.