A successful Codex worktree handoff detaches the old provider session, but the old run is then recorded as interrupted. On Linux web, a real handoff and automatic continuation succeeded while the timeline still showed a red “Run interrupted · Stopped by the provider” divider.
Treat an interrupted provider turn as completed when its thread moved to another workspace, using the normal checkpoint and continuation path. Keep the raw interruption for native-child cleanup so detached children do not remain running. Check for an unpaired Stop request inside the terminal-write transaction; a concurrent Stop retains the interrupted outcome and its attribution.
Fixes https://github.com/pingdotgg/t3code/issues/24513 GitCafe: https://git.cafe/pingdotgg/t3code/issues/24513
Triaged bug via the automated issue flow. Shared server lifecycle fix; no client, contract, setting, or persisted-field changes. The existing attribution fix remains untouched.
interrupted. It now asserts completed, a checkpoint, the new worktree binding, and no interrupt-result item.attached: true, and both runs completed with checkpoints and no false interrupted divider.Evidence: fixture matrix, full index, narrated steps and UTC timestamps, raw bundle, runtime extracts and reproduction helpers.
Each recording starts a real native Codex child running sleep 60, immediately hands the thread off once, and waits for continuation to report its actual cwd, branch and clean status. Before: parent falsely interrupted. After: parent completed, child stopped, continuation succeeded. After recordings also open the settled parent/child details.
Before is historical main 8176b58931, not the final merge base. After is 43cd19866d; intervening main changes affect mobile Markdown, preview certificates and Cursor terminal handling. This set proves runtime behavior, not an isolated pixel comparison.
Linux Chromium 154, DPR 2, 16px root font, desktop UA at both CSS widths. Recordings retain 2880×1600 / 780×1600 pixels; desktop preview stills are downsampled to 1280×711. Stills settled and matched twice at least 500ms apart. Diff0 below is amplified ×8; threshold is explicitly 0, unmasked.
| CSS viewport / theme | Before recording | After recording |
|---|---|---|
| 1440×800 / light | Before | After |
| 1440×800 / dark | Before | After |
| 390×800 / light | Before | After |
| 390×800 / dark | Before | After |
| CSS viewport / theme | Before · 8176b58931 | After · 43cd19866d | Diff0 ×8 | Threshold 0 |
|---|---|---|---|---|
| 1440×800 / light | Before 1440-light | After 1440-light | Difference 1440-light | Threshold 1440-light |
| 1440×800 / dark | Before 1440-dark | After 1440-dark | Difference 1440-dark | Threshold 1440-dark |
| 390×800 / light | Before 390-light | After 390-light | Difference 390-light | Threshold 390-light |
| 390×800 / dark | Before 390-dark | After 390-dark | Difference 390-dark | Threshold 390-dark |
What else changed in the captures: provider prose/capitalization, titles, durations, timestamps, hover controls, folding, spacing and scroll composition vary between independent real runs. The normal completed parent now folds. No claim that nothing else changed visually.
Unverified at runtime: Electron, native mobile, other providers and remote/relay/tunnel modes. The 390-wide captures are responsive web. Shared server projections are covered by the focused tests; client renderers and wire contracts are unchanged. Stop-race and event-ingestion closure are proven by deterministic tests, not live-provider race footage.
Independent proof audit: H/N handoff and native-child runtime rows supported at all four viewport/theme cells; full matrix remains incomplete. Missing explicit controls: recovery-lookup failure, unchanged/same-path workspace callbacks, and changed-workspace non-interrupted terminal variants. The auditor reproduced the settled renderer (transcript pixels matched; 0.131% total delta in header controls), but no uploaded browser replay driver exists for a fresh interaction rebuild. Historical-base and natural transcript differences also prevent an isolated pixel claim. These are pending proof gates; the observed runtime behavior did not contradict the fix.
Models and harnesses: Claude Opus 5.5 via Claude Code; GPT-6.1-Sol via Codex in T3 Code.
Added 6 commits
Verdict: Approved at 43cd198
Macroscope's review found this PR approvable — This is a bounded server lifecycle bug fix that corrects successful worktree handoffs while preserving interruption semantics for real Stop requests and native-child cleanup. Production changes are limited to the existing orchestration path and are covered by focused persistence, race, subagent, and integration tests.
You can add or adjust custom eligibility rules. Learn more.
✅ 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 | 20.9 KiB | 20.9 KiB | 0 B (0.0%) | 29.3 KiB | ✅ |
| Codex | Live turn messages | 2 | 2 | 0 (0.0%) | 8 | ✅ |
| Claude | Total thread wire | 5.0 KiB | 5.0 KiB | −7 B (−0.1%) | 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 | −7 B (−0.6%) | 2.0 KiB | ✅ |
| Claude | Live turn WebSocket decoded | 21.2 KiB | 21.2 KiB | 0 B (0.0%) | 29.3 KiB | ✅ |
| Claude | Live turn messages | 2 | 2 | 0 (0.0%) | 8 | ✅ |
Baseline: 3afb9b3 · PR result: 43cd198 · Source CI: success
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.
Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.
No actionable comments were generated in the recent review.
46501324-d5f2-44ff-bcda-34d1cd4906a1Reviewing files that changed from the base of the PR and between 3afb9b366f271b56a3e49fd86f2594981ec11b66 and 43cd19866dc706a4e7724f3fcafb6d4d5dd4d7c2.
apps/server/src/orchestration-v2/EventSink.tsapps/server/src/orchestration-v2/FoundationPersistence.test.tsapps/server/src/orchestration-v2/ProviderTurnStartService.tsapps/server/src/orchestration-v2/RunExecutionService.test.tsapps/server/src/orchestration-v2/RunExecutionService.tsapps/server/src/orchestration-v2/SelectionRestart.integration.test.tsIncluded review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 3 remain after this review.
The change adds a workspace-move check to root-run finalization. When an interrupted root run coincides with a workspace move, finalization attempts a completed terminal write. A pending interrupt request can block that write, causing finalization to write the interrupted terminal instead.
Changes: Workspace handoff finalization
| Layer / File(s) | Summary |
|---|---|
Guard terminal writes against pending interrupts apps/server/src/orchestration-v2/EventSink.ts, apps/server/src/orchestration-v2/RunExecutionService.ts, apps/server/src/orchestration-v2/FoundationPersistence.test.ts | writeIfRunCurrent can reject a write when an interrupt request exists without its matching result. Run finalization passes the request and result IDs and tracks whether the write committed. The persistence test covers both rejected and committed writes. |
Detect workspace moves and finalize root runs apps/server/src/orchestration-v2/ProviderTurnStartService.ts, apps/server/src/orchestration-v2/RunExecutionService.ts, apps/server/src/orchestration-v2/RunExecutionService.test.ts, apps/server/src/orchestration-v2/SelectionRestart.integration.test.ts | The provider turn start flow passes a workspace-move check to root-run execution. For an interrupted root terminal after a move, finalization attempts a completed terminal write and falls back to the interrupted terminal if the write does not commit. Tests cover handoff behavior and a Stop-request race. |
Priority: External image Normal
Estimated code review effort: 3 (Moderate) | ~25 minutes
Change: Bug fix · Severity of issue fixed: Low
Possibly related issues
Possibly related PRs
Suggested labels: bugfix, size:L
Suggested reviewers: maria-rcks
43cd1Successful worktree handoffs can settle as completed without overriding a user Stop. No actionable merge-blocking risk remains after normal checks.
| Check name | Status | Explanation |
|---|---|---|
| Title check | Passed | The title clearly and concisely describes the main fix: successful worktree handoffs now complete normally. |
| Description check | Passed | The description includes all required sections. It explains the problem, implementation, scope and approval, focused verification, test results, limitations, and evidence. |
| Linked Issues check | Passed | The changes satisfy issue #24513. RunExecutionService detects an interrupted turn after a workspace move and first writes a completed terminal state. It retains the raw interrupted terminal for nati… |
| Out of Scope Changes check | Passed | The changes remain within issue #24513. The source changes are shared server lifecycle logic in RunExecutionService, EventSink, and ProviderTurnStartService, plus focused persistence and integra… |
Explanation
The changes satisfy issue #24513. RunExecutionService detects an interrupted turn after a workspace move and first writes a completed terminal state. It retains the raw interrupted terminal for native-child cleanup. EventSink.writeIfRunCurrent checks for an unpaired interrupt request in the terminal-write transaction, so a concurrent Stop falls back to interrupted attribution. ProviderTurnStartService supplies the workspace-move probe and returns false when projection loading fails. Tests cover successful handoff completion, checkpoint and worktree binding, omitted interrupt results, native-child cleanup, and the Stop race.
Explanation
The changes remain within issue #24513. The source changes are shared server lifecycle logic in RunExecutionService, EventSink, and ProviderTurnStartService, plus focused persistence and integration tests. These tests directly verify successful handoff state and Stop-race safety. No unrelated client, contract, setting, or persisted-field changes are shown in the reviewed change summary.
Comment @coderabbitai help to get the list of available commands.