Skip to content

fix(desktop): wait for the server's database upgrade instead of reporting a failed start - #35056

Merged
juliusmarminge merged 3 commits into
mainfrom
factory/issue-29516-a1794473846
Nov 13, 2026
Merged

juliusmarminge merged 3 commits into
mainfrom
factory/issue-29516-a1794473846

Conversation

@juliusmarminge

@juliusmarminge juliusmarminge commented Nov 12, 2026 •

Copy link
Copy Markdown
Member

The desktop gave the backend a fixed 180s readiness budget, then reported "The server started but did not become ready in time." Long upgrade migrations on large databases outlast it, so users saw a false failure and quit mid-migration. The server now logs shared markers around pending migrations (packages/shared/src/databaseMigrationLog.ts): a start line before them, the existing per-migration "Upgrading database" progress line, and the existing "Migrations ran successfully" line. While the desktop's backend output tail shows a start or progress line after the last finished line, a failed readiness probe logs a warning and probes again instead of reporting a startup failure. A backend that crashes still reports through the exit path. Merged origin/main without conflicts. Main now runs migrations on a worker thread, so the listener answers 503 during an upgrade, and it logs one progress line per migration. Because the desktop keeps only the last 4,000 characters of output, an upgrade spanning many migrations pushed the start line out and brought the false failure back. My follow-up commit counts progress lines as a running upgrade and fixes a stale comment that said migrations block the server. The existing desktop readiness test now has its migrating backend print a start line followed by 80 progress lines; it fails without the follow-up and passes with it.

Fixes #29516

Reproduction on main

Not executed during screening. Prepare two independent disposable copies of a large database awaiting an expensive upgrade, like the 12.5 GB database whose migration 72 took 4m45s in #29516. Launch the desktop on main against one copy and this change against the other, ensuring migration execution exceeds three minutes and neither database has already been upgraded. Leave both apps open: main should show the startup failure at three minutes, while this change should retain the splash and open after migration completion. Record migration output and readiness timing; pausing the backend after it has already become ready does not reproduce this bug.

Evidence

Ran separately built Linux Electron desktops under Xvfb at 1920×1080 in light theme. Each received an independent copy of the same synthetic 23 GiB database awaiting migration 72. A backend-only 2% CPU quota during the first four minutes ensured actual SQLite upgrade execution exceeded three minutes; the desktop remained unrestricted. Both apps stayed open until readiness. No post-readiness pause or simulated migration was used.

Before After
3de49baa70 954404df98
Before: false failure while migration 72 is running After: splash remains beyond three minutes
Full recording: false failure at three minutes; backend ready at 462.4 seconds. Full recording: splash retained; automatically opens at 420.8 seconds.

After: desktop automatically opens when upgrades finish

The actual migration-72 CREATE INDEX took 292.724 seconds on base and 276.509 seconds on head. Head timeout output confirms it keeps waiting for the database upgrade. Runtime output and readiness timings.

Crash control: terminating the captured backend child before readiness still produces “The server could not start (killed by SIGKILL)” on base and head.

Driver, synthetic seed scripts, crash controls, and reproduction instructions. Recordings retain real-time timing; local paths and setup names are redacted.

Not shown:

  • Migration-progress recognition after enough output evicts the original start line was not separately stress-tested; the captured runs prove the requested long-upgrade behavior.
  • Crash control covers SIGKILL before readiness, not every possible backend failure mode.

Tests

vp test run apps/desktop/src/backend/DesktopBackendManager.test.ts -t 'waits out a slow startup' passes, and fails with 'expected false to be true' when the progress marker is removed from isRunningDatabaseMigrations; vp test run apps/server/src/persistence/Migrations.test.ts apps/server/src/persistence/Sqlite.test.ts: 18/18 pass; vp run check:changed: format, lint, knip, graphql, typecheck (41 packages) and tests (3 files, 25 tests) all pass

Not in this change

  • No real-app capture of the failure or the fix; that is left to the proof step (it needs a database copy with long pending migrations and a backend past the 180s budget).
  • A migration that hangs without the server exiting now keeps the desktop waiting indefinitely instead of showing the failure dialog; a crash still reports through the exit path.
  • When migrations are pending, the off-main-thread path runs hasPendingMigrations twice (once in Sqlite.ts, once in runMigrations); it is a cheap query, so I left it.

Who did the work

  • Written by Claude Opus 5.5, in Claude Code
  • Reviewed by GPT-6.1 Sol, in Codex, and by Claude Opus 5.5, each on its own
  • Proof captured by GPT-6.1 Sol and checked by Claude Opus 5.5

Run by the t3code factory in T3 Code.

…ting a failed start

The desktop gave the backend a fixed readiness budget and then showed
"could not start". Long upgrade migrations on large databases outlast it,
and the server printed nothing while migrating, so the app reported a false
failure and users quit mid-migration.

The server now logs "Running database migrations" before pending migrations
run. While that is the backend's latest migration line, the desktop keeps
waiting instead of reporting a readiness failure. A crash still reports
through the exit path.

AI-assisted; verified with the tests and tools listed above.
@github-actions github-actions Bot added vouch:trusted Author is trusted by repo permissions or the VOUCHED list. size:M 30-99 changed lines (additions + deletions). labels Nov 12, 2026
@macroscopeapp

macroscopeapp Bot commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Approvability

Verdict: Approved at 954404d

Macroscope's review found this PR approvable — Desktop startup now waits without reporting failure when server logs show an ongoing database upgrade. Migration execution, readiness retries and process-exit handling remain unchanged, keeping the impact limited to startup reporting.

You can add or adjust custom eligibility rules. Learn more.

macroscopeapp[bot]
macroscopeapp Bot previously approved these changes Nov 12, 2026
@github-actions

github-actions Bot commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

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 21.5 KiB 21.5 KiB 0 B (0.0%) 24.4 KiB ✅
Codex Live turn messages 1 1 0 (0.0%) 8 ✅
Claude Total thread wire 5.0 KiB 5.0 KiB 0 B (0.0%) 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 0 B (0.0%) 2.0 KiB ✅
Claude Live turn WebSocket decoded 21.9 KiB 21.9 KiB 0 B (0.0%) 24.4 KiB ✅
Claude Live turn messages 1 1 0 (0.0%) 8 ✅

Other scenarios

All 64 scenario measurements match main.

All 64 scenario measurements
Provider Scenario Metric Main baseline This PR Impact PR ceiling
Codex short turn resume WebSocket decoded 10.5 KiB 10.5 KiB 0 B (0.0%) 12.7 KiB ✅
Codex short turn resume WebSocket messages 2 2 0 (0.0%) 2 ✅
Codex long turn running, cold open HTTP decoded 281.7 KiB 281.7 KiB 0 B (0.0%) 302.7 KiB ✅
Codex long turn running, cold open rows 75 75 0 (0.0%) 75 ✅
Codex long turn running, history page HTTP decoded 92.4 KiB 92.4 KiB 0 B (0.0%) 107.4 KiB ✅
Codex long turn running, history page rows 75 75 0 (0.0%) 75 ✅
Codex long turn finished, cold open HTTP decoded 381.5 KiB 381.5 KiB 0 B (0.0%) 410.2 KiB ✅
Codex long turn finished, cold open rows 151 151 0 (0.0%) 151 ✅
Codex long turn finished, socket cold open WebSocket decoded 381.6 KiB 381.6 KiB 0 B (0.0%) 410.2 KiB ✅
Codex long turn finished, socket cold open WebSocket messages 2 2 0 (0.0%) 2 ✅
Codex long turn finished, socket cold open rows 151 151 0 (0.0%) 151 ✅
Codex long turn finished, work page HTTP decoded 186.2 KiB 186.2 KiB 0 B (0.0%) 200.2 KiB ✅
Codex long turn finished, work page rows 150 150 0 (0.0%) 150 ✅
Codex idle thread resume WebSocket decoded 65 B 65 B 0 B (0.0%) 200 B ✅
Codex idle thread resume WebSocket messages 1 1 0 (0.0%) 1 ✅
Codex fleet shell snapshot HTTP decoded 42.4 KiB 42.4 KiB 0 B (0.0%) 53.7 KiB ✅
Codex fleet shell snapshot rows 27 27 0 (0.0%) 27 ✅
Codex fleet shell subscribe WebSocket decoded 65 B 65 B 0 B (0.0%) 200 B ✅
Codex fleet shell subscribe WebSocket messages 1 1 0 (0.0%) 1 ✅
Codex fleet shell resume check HTTP decoded 18 B 18 B 0 B (0.0%) 100 B ✅
Codex fleet shell resume check rows 0 0 0 0 ✅
Codex stale shell reload HTTP decoded 42.4 KiB 42.4 KiB 0 B (0.0%) 53.7 KiB ✅
Codex stale shell reload rows 27 27 0 (0.0%) 27 ✅
Codex stale shell resume WebSocket decoded 140 B 140 B 0 B (0.0%) 400 B ✅
Codex stale shell resume WebSocket messages 2 2 0 (0.0%) 2 ✅
Codex stale todo list cold open HTTP decoded 72.4 KiB 72.4 KiB 0 B (0.0%) 83.0 KiB ✅
Codex stale todo list cold open rows 75 75 0 (0.0%) 75 ✅
Codex stale todo list cold open nodes 2 2 0 (0.0%) 4 ✅
Codex stale todo list cold open subagents 2 2 0 (0.0%) 4 ✅
Codex stale todo list cold open plans 1 1 0 (0.0%) 2 ✅
Codex stale todo list cold open providerTurns 1 1 0 (0.0%) 2 ✅
Codex stale todo list cold open checkpoints 1 1 0 (0.0%) 2 ✅
Claude short turn resume WebSocket decoded 10.6 KiB 10.6 KiB 0 B (0.0%) 12.7 KiB ✅
Claude short turn resume WebSocket messages 2 2 0 (0.0%) 2 ✅
Claude long turn running, cold open HTTP decoded 283.3 KiB 283.3 KiB 0 B (0.0%) 302.7 KiB ✅
Claude long turn running, cold open rows 75 75 0 (0.0%) 75 ✅
Claude long turn running, history page HTTP decoded 93.4 KiB 93.4 KiB 0 B (0.0%) 107.4 KiB ✅
Claude long turn running, history page rows 75 75 0 (0.0%) 75 ✅
Claude long turn finished, cold open HTTP decoded 384.2 KiB 384.2 KiB 0 B (0.0%) 410.2 KiB ✅
Claude long turn finished, cold open rows 151 151 0 (0.0%) 151 ✅
Claude long turn finished, socket cold open WebSocket decoded 384.3 KiB 384.3 KiB 0 B (0.0%) 410.2 KiB ✅
Claude long turn finished, socket cold open WebSocket messages 2 2 0 (0.0%) 2 ✅
Claude long turn finished, socket cold open rows 151 151 0 (0.0%) 151 ✅
Claude long turn finished, work page HTTP decoded 188.1 KiB 188.1 KiB 0 B (0.0%) 200.2 KiB ✅
Claude long turn finished, work page rows 150 150 0 (0.0%) 150 ✅
Claude idle thread resume WebSocket decoded 66 B 66 B 0 B (0.0%) 200 B ✅
Claude idle thread resume WebSocket messages 1 1 0 (0.0%) 1 ✅
Claude fleet shell snapshot HTTP decoded 42.9 KiB 42.9 KiB 0 B (0.0%) 53.7 KiB ✅
Claude fleet shell snapshot rows 27 27 0 (0.0%) 27 ✅
Claude fleet shell subscribe WebSocket decoded 66 B 66 B 0 B (0.0%) 200 B ✅
Claude fleet shell subscribe WebSocket messages 1 1 0 (0.0%) 1 ✅
Claude fleet shell resume check HTTP decoded 18 B 18 B 0 B (0.0%) 100 B ✅
Claude fleet shell resume check rows 0 0 0 0 ✅
Claude stale shell reload HTTP decoded 42.9 KiB 42.9 KiB 0 B (0.0%) 53.7 KiB ✅
Claude stale shell reload rows 27 27 0 (0.0%) 27 ✅
Claude stale shell resume WebSocket decoded 140 B 140 B 0 B (0.0%) 400 B ✅
Claude stale shell resume WebSocket messages 2 2 0 (0.0%) 2 ✅
Claude stale todo list cold open HTTP decoded 73.4 KiB 73.4 KiB 0 B (0.0%) 83.0 KiB ✅
Claude stale todo list cold open rows 75 75 0 (0.0%) 75 ✅
Claude stale todo list cold open nodes 2 2 0 (0.0%) 4 ✅
Claude stale todo list cold open subagents 2 2 0 (0.0%) 4 ✅
Claude stale todo list cold open plans 1 1 0 (0.0%) 2 ✅
Claude stale todo list cold open providerTurns 1 1 0 (0.0%) 2 ✅
Claude stale todo list cold open checkpoints 1 1 0 (0.0%) 2 ✅

Baseline: 3de49ba · PR result: 954404d · 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

coderabbitai Bot commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Explain Change →

No actionable comments were generated in the recent review.

Walkthrough

The server now logs shared database migration markers. During readiness checks, the desktop manager checks captured output for active migrations and defers startup failure handling when it finds them. Tests cover migration output and readiness timeout behavior.

Changes: Migration-aware startup

Layer / File(s) Summary
Shared migration markers and server logging
packages/shared/src/databaseMigrationLog.ts, packages/shared/package.json, apps/server/src/persistence/Migrations.ts
Shared code defines migration start, progress, and finish messages and checks which marker occurred most recently. The migration runner uses the shared messages.
Desktop readiness handling and tests
apps/desktop/src/backend/DesktopBackendManager.ts, apps/desktop/src/backend/DesktopBackendManager.test.ts
When output indicates an active migration, the desktop manager logs a warning and returns before reporting startup failure or persisting a failure snapshot. Tests cover active migration output, readiness after HTTP 200, and failure after the timeout.

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

Change: Bug fix · Severity of issue fixed: Medium

Sequence Diagram(s)

sequenceDiagram
  participant MigrationRunner
  participant DesktopBackendManager
  participant MigrationLogCheck
  MigrationRunner->>DesktopBackendManager: captured output with migration markers
  DesktopBackendManager->>MigrationLogCheck: check captured output
  MigrationLogCheck-->>DesktopBackendManager: active migration status
  DesktopBackendManager-->>DesktopBackendManager: defer readiness failure handling
Loading

Possibly related issues

  • pingdotgg/t3code#35421: The issue describes slow startup migrations. This change defers desktop readiness failure handling when migration markers show that a migration is active.

Possibly related PRs

  • pingdotgg/t3code#26247: Raises the managed backend readiness budget and tests the startup timeout exercised by this change.

Suggested labels: bugfix

Suggested reviewers: shivamhwp, yash-singh1

Merge Risk: Minimal · up to 95440

The change keeps the desktop waiting while database migrations are running instead of showing a false startup failure. No concrete merge-blocking problem was found. The author notes that the full app was not tested end to end with the fix.

Pre-merge checks: Passed 4

Check name Status Explanation
Title check Passed The title names the main change: the desktop now waits for the server's database upgrade instead of reporting a failed start. The raw summary confirms this. The desktop backend manager skips the start…
Linked Issues check Passed [#29516] The issue reports that the desktop app does not start on Windows 0.1.6. It gives no logs and no acceptance criteria. Expected behavior: the app starts. The PR attributes the failure to a data…
Out of Scope Changes check Passed All changed files serve one objective: keep desktop startup waiting while database migrations run. packages/shared/src/databaseMigrationLog.ts holds the shared message constants and the detection func…
Description check Passed The description is thorough and covers the template's required sections. Problem: the desktop reports a false startup failure when a long database migration outlasts the 180s readiness budget, and the…
Full details: Title check

Explanation

The title names the main change: the desktop now waits for the server's database upgrade instead of reporting a failed start. The raw summary confirms this. The desktop backend manager skips the startup-failure report while migrations run, and the server logs migration start and finish messages. The title uses the conventional commit form. It is specific and does not use vague terms.

Full details: Linked Issues check

Explanation

[#29516] The issue reports that the desktop app does not start on Windows 0.1.6. It gives no logs and no acceptance criteria. Expected behavior: the app starts. The PR attributes the failure to a database migration that runs longer than the fixed readiness budget. At head 954404d, the reviewed code covers that path. apps/server/src/persistence/Migrations.ts logs 'Running database migrations' when pending migrations exist. packages/shared/src/databaseMigrationLog.ts exports the start, progress, and finish strings and isRunningDatabaseMigrations(). The follow-up commit counts progress lines as migration evidence, so a long run of per-migration lines cannot push the start line out of the 4,000-character tail and cause a false failure. DesktopBackendManager.ts onReadinessFailure checks that tail. While migrations run, it logs a warning and does not report a startup failure, so the probe loop keeps waiting. Crash handling stays on the exit/restart path. The DesktopBackendManager.test.ts changes cover a migration-start line that has left the tail, and a backend that stays unready past the timeout still reports failure. The cause comes from the author's simulated reproduction, not from the reporter's logs, so the link to the reporter's exact failure is unconfirmed. The implemented objective is present in the code. Note: the current PR description says detection uses only start and finish lines. The head code also treats progress lines as evidence. The code at head governs.

Full details: Out of Scope Changes check

Explanation

All changed files serve one objective: keep desktop startup waiting while database migrations run. packages/shared/src/databaseMigrationLog.ts holds the shared message constants and the detection function. packages/shared/package.json adds the export that the server and desktop import. apps/server/src/persistence/Migrations.ts emits the start, progress, and finish logs. apps/desktop/src/backend/DesktopBackendManager.ts applies the readiness-failure guard. The test files cover that guard and the new progress-line case. The follow-up change for progress lines directly supports the same guard. No unrelated behavior change appears in the summarized changes. The assessment relies on the whole-PR summary and the listed file changes; line-level diffs were not read.

Full details: Description check

Explanation

The description is thorough and covers the template's required sections. Problem: the desktop reports a false startup failure when a long database migration outlasts the 180s readiness budget, and the description gives expected behavior and reproduction steps. Change: shared migration log markers, a desktop check on the captured output tail, and a retry in place of a failure report while migrations run, with a crash still on the exit path. Scope: it links issue #29516 as the bug, and it states the change is a bug fix. Verification: it lists focused tests with results, and it includes before/after evidence, recordings, and a crash control. It also states what was not checked, which the template asks for. It names the models that did the work, as the template asks. The description is long and includes some detail beyond the template, but no required section is missing or empty. The title line is not in the template body, so it is not checked here.

Finishing Touches

Generate unit tests (beta)

  • Commit to this branch
  • Create a new PR
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Recent review info

Run configuration

Commits

Reviewing files that changed from the base of the PR and between b91e87b and 954404d.

Files selected for processing: 3

  • apps/desktop/src/backend/DesktopBackendManager.test.ts
  • apps/server/src/persistence/Migrations.ts
  • packages/shared/src/databaseMigrationLog.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 2 remain after this review.

@juliusmarminge

Copy link
Copy Markdown
Member Author

The automated pipeline stopped on this pull request and will not merge it:

the capture step could not show the problem on the base commit, so there is nothing to prove fixed.

Built apps/desktop on each commit. Launched electron dist-electron/main.cjs --no-sandbox --disable-gpu with a fresh T3CODE_HOME and T3CODE_DEV_SEED=0. Sent SIGSTOP to the spawned backend at seven seconds, captured around 20, 175 and 190 seconds, sent SIGCONT at 200 seconds, captured after resume, and stopped around 230 seconds. Both backends became HTTP-ready before the seven-second pause. Both recordings showed Welcome past the readiness budget; neither reproduced the reported startup failure. Consequently, the head's slow-migration behavior was not proven. All started apps were stopped and both worktrees and scratch files were removed.

It stays open for a person to finish or close.

juliusmarminge and others added 2 commits November 12, 2026 23:16
…of the output tail

The server now logs one progress line per pending migration, and the
desktop keeps only the last 4,000 characters of backend output. An upgrade
spanning many migrations pushed the start line out, so the desktop reported
a failed start again. Progress lines now count as a running upgrade too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@macroscopeapp
macroscopeapp Bot dismissed their stale review November 12, 2026 23:21

Dismissing prior approval to re-evaluate 954404d

@juliusmarminge
juliusmarminge merged commit 2e741fc into main Nov 13, 2026
36 checks passed
@juliusmarminge
juliusmarminge deleted the factory/issue-29516-a1794473846 branch November 13, 2026 00:00
github-actions Bot added a commit to omarcresp/t3code-flake that referenced this pull request Nov 13, 2026
**v0.1.10-nightly.20261113.4616** · 143 changes from 68 contributors

- **Keep agents working.** Your computer stays awake while agents are running, preventing sleep from interrupting their work.
- **Codex cyber access.** Choose Standard or Daybreak cyber access for eligible Codex models.
- **Check secret references.** Verify a saved secret reference in Settings before starting a provider.
- **Faster Usage page.** See cost and token figures immediately while Usage refreshes them in the background.
- **Faster terminal opening.** Terminals with large histories attach faster over slow connections.
- **Preserve local PR commits.** Opening a pull request in a new worktree no longer drops unpushed commits on its local branch.
- **Protect nested repositories.** Worktree cleanup now preserves worktrees containing nested repositories and their uncommitted files.

### Plus 127 more bug fixes and performance improvements.

<details>
<summary>Full changelog</summary>

## What's Changed
* perf(usage): the Usage page shows cost and tokens at once instead of after a full scan by @Bil0000 in pingdotgg/t3code#33993
* feat(codex): choose a cyber access program for Daybreak-eligible models by @Mezzle in pingdotgg/t3code#27928
* fix(web): command palette footer hints wrap instead of clipping on narrow viewports by @imluckii in pingdotgg/t3code#29288
* feat(secrets): check a secret reference before saving it by @yordis in pingdotgg/t3code#31723
* fix(server): the Welcome page offers repositories you only worked on in worktrees by @dominikzabcik in pingdotgg/t3code#33923
* fix(web): a branch picked right after opening the picker no longer waits for its status read by @Yuruzuu in pingdotgg/t3code#33877
* perf(server): status refresh on a fork's default branch is 2.5x faster (median 7,826 to 3,168 ms) by @Yuruzuu in pingdotgg/t3code#33954
* fix(server): agents can see a thread's pending background work by @fjooord in pingdotgg/t3code#33858
* fix(ssh): timed-out SSH commands keep the output they already printed by @juliusmarminge in pingdotgg/t3code#35589
* fix(server): worktree dev servers no longer clean up real worktrees by @Lenoridev in pingdotgg/t3code#33902
* fix(server): server updates retry when a release download fails briefly by @shivamhwp in pingdotgg/t3code#33896
* fix(server): dev database seed no longer runs settle scripts in real worktrees by @Lenoridev in pingdotgg/t3code#33908
* perf(server): avoid loading provider drivers for terminal imports by @Adamulek123 in pingdotgg/t3code#33898
* fix(web): queued message preview keeps inline image labels in place by @Ragulsundaram in pingdotgg/t3code#33405
* fix(web): project settings can reach each checkout on one environment by @EmanuelMedinaAraujo in pingdotgg/t3code#29441
* perf(server,client): attaching to a terminal with a large history over a slow link drops from 5,555 ms to 751 ms median (7.4x) by @Yuruzuu in pingdotgg/t3code#33942
* fix(server): PR sync reads share GitHub requests again on busy servers by @only21mil in pingdotgg/t3code#28422
* ci: pin GitHub Actions to full commit SHAs by @yordis in pingdotgg/t3code#33696
* fix(mcp): answer MCP 2026-07-28 discovery instead of a 400 by @Phicks-debug in pingdotgg/t3code#28683
* fix(server): a server published through Tailscale Serve no longer reports itself as local-only by @juliusmarminge in pingdotgg/t3code#35543
* perf(server): save large agent recordings faster by @Yuruzuu in pingdotgg/t3code#33930
* fix(web): hand-started t3 servers get an update command that updates them by @hieu-lee in pingdotgg/t3code#29462
* fix(web): Select All then Copy in a read-only file preview copies the whole file by @whoisaldo in pingdotgg/t3code#33903
* perf(web): PDF Find jumps to the first match in long PDFs 3.1x faster (median) by @Yuruzuu in pingdotgg/t3code#33956
* fix(mobile): Keep a tall question's Submit button above Android's navigation bar by @mwolson in pingdotgg/t3code#33929
* fix(web): choosing a project icon again re-reads it from disk by @AlexSandilands in pingdotgg/t3code#28584
* fix(web): sidebar keeps the open thread active while Settings or Usage is open by @Mnigos in pingdotgg/t3code#28703
* fix(web): gitignore fences in the rich-text composer highlight as ini, as chat does by @juliusmarminge in pingdotgg/t3code#35603
* fix(web): keep later navigation when thread deletion completes by @Edge358 in pingdotgg/t3code#33953
* fix(server): Git status finds the default branch when the only remote isn't origin by @only21mil in pingdotgg/t3code#28815
* fix(pi): settle prompts consumed by input handlers by @Chinoman10 in pingdotgg/t3code#28798
* fix(cursor): honor HTTP/1.1 transport for proxy networks by @only21mil in pingdotgg/t3code#25338
* fix(client): hide Resume while a pending request blocks continuing the run by @only21mil in pingdotgg/t3code#28810
* fix(server): t3 doctor holds off server launches and names the holder by @only21mil in pingdotgg/t3code#24963
* fix(server): preserve filename whitespace when reading and saving files by @voltcrash in pingdotgg/t3code#29314
* fix(server): tool calls a released session cuts off no longer keep the thread Waiting by @NotReliable in pingdotgg/t3code#29036
* fix(server): keep thread tool responses small by default by @voltcrash in pingdotgg/t3code#29268
* fix(web): a chunk that fails after boot no longer reloads the page in a loop by @grodriguez-fr in pingdotgg/t3code#29306
* fix(opencode): keep ACP question choices and fold the custom answer by @QRpeachKZ in pingdotgg/t3code#25879
* fix(web): sent messages honor their theme foreground by @fancyboi999 in pingdotgg/t3code#26888
* perf(ci): run server test files in parallel on each shard by @juliusmarminge in pingdotgg/t3code#35618
* chore(deps): bump the npm_and_yarn group across 2 directories with 1 update by @dependabot[bot] in pingdotgg/t3code#33945
* fix(server): keep worktrees containing nested repositories by @maria-rcks in pingdotgg/t3code#25241
* fix(shared): Windows-style paths stay out of the slash menu by @samkelly880 in pingdotgg/t3code#27851
* fix(acp): let agents that advertise /compact compact their own context by @ParasN in pingdotgg/t3code#28822
* test(acp-registry): subscribe to snapshot changes before triggering them by @juliusmarminge in pingdotgg/t3code#35627
* fix(web): diff colors follow appearance contrast by @voltcrash in pingdotgg/t3code#28947
* fix(web): keep the child thread open when its remote parent was deleted by @ScottN-PV in pingdotgg/t3code#27436
* fix(desktop): poll for updates every 4 hours, not 4 minutes by @tao12345666333 in pingdotgg/t3code#28481
* fix(usage): bill Codex requests whose cumulative counters rewind to an earlier snapshot by @jokaperes in pingdotgg/t3code#29366
* fix(server): persist terminal output without rewriting all history by @nateEc in pingdotgg/t3code#24019
* perf(server): stop serializing terminal chunks just to measure them by @bompus in pingdotgg/t3code#24179
* fix(web): stop opening a menu when a terminal selection ends by @fixfon in pingdotgg/t3code#24112
* fix(server): agents keep named browser profiles after a server restart by @only21mil in pingdotgg/t3code#23484
* fix(web): reveal full composer command descriptions by @voltcrash in pingdotgg/t3code#27414
* fix(server): restart the service after T3 home ownership clears by @maria-rcks in pingdotgg/t3code#25237
* fix(web): Enter on an empty prompt resumes an interrupted thread by @toni92-beep in pingdotgg/t3code#27079
* fix(shared): token counts just below a unit boundary use the next unit by @juliusmarminge in pingdotgg/t3code#35634
* fix(desktop): long annotation strokes stay cheap to draw by @only21mil in pingdotgg/t3code#24573
* fix(ssh): load remote profiles with the account login shell by @nabettu in pingdotgg/t3code#23934
* fix(server): terminals recreate a worktree removed by storage cleanup by @TonybynMp4 in pingdotgg/t3code#24156
* test(web): Mod+E bold-word test no longer drops its first arrow key by @juliusmarminge in pingdotgg/t3code#35646
* fix(server): access list keeps changes made while its stream opens by @juliusmarminge in pingdotgg/t3code#35640
* fix(server): a failed Git version probe no longer advises installing Git by @juliusmarminge in pingdotgg/t3code#35638
* fix(server): PR worktrees keep local commits on an existing branch by @only21mil in pingdotgg/t3code#24674
* fix(web): branch labels keep the space before a middle-truncated name by @guryanov-a in pingdotgg/t3code#23627
* fix(server): stale provider update results clear once the provider is up to date by @fixfon in pingdotgg/t3code#24338
* perf(server): merged-PR threads no longer re-read git state every minute by @Batuhan1999 in pingdotgg/t3code#24099
* fix(server): pending request reads keep question option previews by @juliusmarminge in pingdotgg/t3code#35644
* fix(opencode): preserve owned turns during server recovery by @maria-rcks in pingdotgg/t3code#25596
* fix(server): baseline existing pull request checks when watching by @vkpdeveloper in pingdotgg/t3code#23522
* fix(server): saved preview snapshot failures name the screenshot destination by @juliusmarminge in pingdotgg/t3code#35649
* fix(ssh): distinguish incompatible Node from a missing executable by @MasonNight in pingdotgg/t3code#35607
* fix(pi): stop offering extension-only models for text generation by @only21mil in pingdotgg/t3code#24110
* fix(server): new worktrees show submodule downloads instead of a stuck checkout by @GuardedAirplane in pingdotgg/t3code#33687
* fix(web): restore composer focus when returning to a thread by @voltcrash in pingdotgg/t3code#29179
* fix(server): linked calls fail fast when a peer stalls the MCP handshake by @only21mil in pingdotgg/t3code#28876
* fix(web,mobile): keep line breaks in pending question text by @StiensWout in pingdotgg/t3code#33353
* fix(mobile): opening a search result clears Home's search on iPhone by @only21mil in pingdotgg/t3code#28324
* fix(desktop): Escape during IME composition no longer cancels a preview annotation by @juliusmarminge in pingdotgg/t3code#35645
* fix(web): a click no longer lands on the sidebar row that just slid under it by @mkantautas in pingdotgg/t3code#23517
* fix(mcp): background wait fails when none of its task IDs exist by @onlydole in pingdotgg/t3code#29773
* fix(web): the collapsed mobile composer shows Stop while a turn runs by @TanJeeSchuan in pingdotgg/t3code#29959
* test(web): split the cloud voice cancel flow from the failed-then-saved setup by @juliusmarminge in pingdotgg/t3code#35665
* fix(desktop): nightly updates no longer depend on GitHub's tag order by @t3dotgg in pingdotgg/t3code#35599
* fix(web): PDF preview prerenders pages two viewports below the visible area by @juliusmarminge in pingdotgg/t3code#35666
* perf(server): uploading a large document with seven notebooks finishes 1.8x faster (median 823 ms -> 461 ms) by @Yuruzuu in pingdotgg/t3code#33946
* Typeset equations that fail only because of an unescaped percent by @meutex in pingdotgg/t3code#29511
* fix(server): wake PR watches when the merge queue drops a pull request by @only21mil in pingdotgg/t3code#24141
* fix(web): legacy collapsed project folders stay collapsed after a save by @juliusmarminge in pingdotgg/t3code#35659
* fix(server): explain how to close unclaimed browser tabs by @voltcrash in pingdotgg/t3code#29205
* fix(editors): JetBrains IDEs open file links in the thread's project by @only21mil in pingdotgg/t3code#33558
* fix(mobile): select part of a source file's text on iOS by @binjamshid in pingdotgg/t3code#33218
* fix(mobile): show loading until existing threads are known by @shivamhwp in pingdotgg/t3code#28663
* fix(web): streamed device toolbar starts at the panel size by @juliusmarminge in pingdotgg/t3code#35664
* fix(acp): preserve OMP single-choice questions and custom answers by @nateEc in pingdotgg/t3code#24012
* fix(server): workspace search leaves out files Git ignores outside .gitignore by @juliusmarminge in pingdotgg/t3code#35673
* fix(server): worktree setup no longer hangs when the shell is Nushell by @KenDJohnson in pingdotgg/t3code#29461
* fix(secrets): notify when a delegated agent asks for a secret by @zmievsa in pingdotgg/t3code#33585
* fix(web): Cursor threads keep a stored model the catalog doesn't list by @only21mil in pingdotgg/t3code#24872
* fix(server): workspace search keeps backslashes in POSIX file names by @juliusmarminge in pingdotgg/t3code#35672
* fix(provider): make readiness probe timeouts configurable by @johanjohnthomas in pingdotgg/t3code#33275
* fix(cursor): mcp.json servers using ${workspaceFolder} now start by @only21mil in pingdotgg/t3code#24505
* fix(server): workspace search ignore filter follows the host path separator by @juliusmarminge in pingdotgg/t3code#35697
* fix(server): t3 uninstall names another home's service it leaves installed by @juliusmarminge in pingdotgg/t3code#35684
* fix(client-runtime): long MCP app download names no longer split an emoji by @juliusmarminge in pingdotgg/t3code#35680
* fix(web): collapsed reasoning previews no longer split an emoji by @juliusmarminge in pingdotgg/t3code#35677
* fix(web): a question answer restored with undo is no longer sent twice by @ixmxvii-hash in pingdotgg/t3code#35574
* fix(server): reject PR watches when hosting dependencies are unavailable by @complynx in pingdotgg/t3code#33389
* fix(chat): requeued steers no longer look delivered by @Alb11747 in pingdotgg/t3code#23420
* fix(web): reuse timeline rows while thinking and plans stream by @Batuhan1999 in pingdotgg/t3code#24121
* fix(mobile): Back from a PR opened in a chat returns to the chat on Android by @bitmvk in pingdotgg/t3code#27653
* fix(mobile): tall iOS code blocks with long lines show their code by @Muzammil-Bit in pingdotgg/t3code#33243
* fix(mobile): refreshing a review diff keeps unfolded files open by @juliusmarminge in pingdotgg/t3code#35682
* fix(server): snooze no longer waits two minutes after a steered queued message by @juliusmarminge in pingdotgg/t3code#35661
* test(server): count leaked PR cache caller data without WeakRefs by @juliusmarminge in pingdotgg/t3code#35694
* fix(cli): t3 pair's protocol mismatch hint keeps --base-dir by @juliusmarminge in pingdotgg/t3code#35690
* fix(web): a dismissed provider banner stays dismissed when you switch threads by @torztomasz in pingdotgg/t3code#25513
* fix(cli): t3 status no longer suggests --no-lan under a T3CODE_HOST override by @juliusmarminge in pingdotgg/t3code#35691
* fix(desktop): wait for the server's database upgrade instead of reporting a failed start by @juliusmarminge in pingdotgg/t3code#35056
* fix(server): a C# dictionary term no longer rewrites standalone letters by @juliusmarminge in pingdotgg/t3code#35700
* fix(client-runtime): last user message sort ignores agent-posted messages by @juliusmarminge in pingdotgg/t3code#33873
* fix(web): long screen-reader announcements keep emoji whole at the cutoff by @juliusmarminge in pingdotgg/t3code#35681
* fix(server): Claude context meter honours CLAUDE_CODE_AUTO_COMPACT_WINDOW by @juliusmarminge in pingdotgg/t3code#33806
* feat(server): watched GitHub pull requests react when a check suite completes by @juliusmarminge in pingdotgg/t3code#35709
* fix(web): triple-click selects the whole line around links and file chips by @juliusmarminge in pingdotgg/t3code#32334
* fix(projects): preserve grouped renames to repository names by @maria-rcks in pingdotgg/t3code#25244
* fix(acp): agents get T3's runtime notes once per session, not on every message by @Radu028 in pingdotgg/t3code#26959
* feat(server): keep the computer awake while agents work by @juliusmarminge in pingdotgg/t3code#34128
* fix(web): find scheduled tasks in Settings search by @juliusmarminge in pingdotgg/t3code#35715
* fix(preview): bound stalled screenshot captures by @cabamarcos in pingdotgg/t3code#25862
* test(server): idle-resume replays wait for the session release before the next prompt by @juliusmarminge in pingdotgg/t3code#35720
* fix(server): archive refuses a thread with running delegated tasks by @SouthCarpet in pingdotgg/t3code#26078
* fix(web): minimap Next turn no longer sticks on adjacent user messages by @juliusmarminge in pingdotgg/t3code#33036
* fix(server): Claude background commands show how long they have run by @SunkenInTime in pingdotgg/t3code#26668
* fix(server): honor worktree prefixes when automatic naming fails by @juliusmarminge in pingdotgg/t3code#35718
* fix(git): refuse remote switches that would silently detach HEAD by @Kirat0201 in pingdotgg/t3code#35687
* fix(mobile): retain inline image zoom on iOS by @saphid in pingdotgg/t3code#33852
* fix(server): Claude's auto-compact setting now wins over CLAUDE_CODE_AUTO_COMPACT_WINDOW by @juliusmarminge in pingdotgg/t3code#35737
* fix(desktop): an SSH reconnect during a desktop update no longer locks the updated app out of its T3 home by @ajbarryiii in pingdotgg/t3code#26191
* fix(web): auto-balanced drafts recheck machine load before sending by @juliusmarminge in pingdotgg/t3code#35658
* fix(web): screen readers at Everything hear context compaction start by @juliusmarminge in pingdotgg/t3code#35739
* fix(server): t3_thread_read returns the newest items and fits in agent context by @darjss in pingdotgg/t3code#33428

## New Contributors
* @Mezzle made their first contribution in pingdotgg/t3code#27928
* @imluckii made their first contribution in pingdotgg/t3code#29288
* @dominikzabcik made their first contribution in pingdotgg/t3code#33923
* @fjooord made their first contribution in pingdotgg/t3code#33858
* @Ragulsundaram made their first contribution in pingdotgg/t3code#33405
* @EmanuelMedinaAraujo made their first contribution in pingdotgg/t3code#29441
* @Phicks-debug made their first contribution in pingdotgg/t3code#28683
* @AlexSandilands made their first contribution in pingdotgg/t3code#28584
* @Edge358 made their first contribution in pingdotgg/t3code#33953
* @Chinoman10 made their first contribution in pingdotgg/t3code#28798
* @NotReliable made their first contribution in pingdotgg/t3code#29036
* @grodriguez-fr made their first contribution in pingdotgg/t3code#29306
* @QRpeachKZ made their first contribution in pingdotgg/t3code#25879
* @fancyboi999 made their first contribution in pingdotgg/t3code#26888
* @ParasN made their first contribution in pingdotgg/t3code#28822
* @tao12345666333 made their first contribution in pingdotgg/t3code#28481
* @jokaperes made their first contribution in pingdotgg/t3code#29366
* @guryanov-a made their first contribution in pingdotgg/t3code#23627
* @vkpdeveloper made their first contribution in pingdotgg/t3code#23522
* @MasonNight made their first contribution in pingdotgg/t3code#35607
* @GuardedAirplane made their first contribution in pingdotgg/t3code#33687
* @meutex made their first contribution in pingdotgg/t3code#29511
* @KenDJohnson made their first contribution in pingdotgg/t3code#29461
* @zmievsa made their first contribution in pingdotgg/t3code#33585
* @johanjohnthomas made their first contribution in pingdotgg/t3code#33275
* @complynx made their first contribution in pingdotgg/t3code#33389
* @torztomasz made their first contribution in pingdotgg/t3code#25513
* @cabamarcos made their first contribution in pingdotgg/t3code#25862
* @ajbarryiii made their first contribution in pingdotgg/t3code#26191

**Full Changelog**: pingdotgg/t3code@v0.1.10-nightly.20261112.4606...v0.1.10-nightly.20261113.4616

</details>


Upstream release: https://github.com/pingdotgg/t3code/releases/tag/v0.1.10-nightly.20261113.4616
adampeterhiggins added a commit to adampeterhiggins/t3code that referenced this pull request Nov 14, 2026
* perf(ci): run server test files in parallel on each shard (pingdotgg#35618)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* chore(deps): bump the npm_and_yarn group across 2 directories with 1 update (pingdotgg#33945)

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): keep worktrees containing nested repositories (pingdotgg#25241)

Co-authored-by: Yukun Shan <92423096+nateEc@users.noreply.github.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(shared): Windows-style paths stay out of the slash menu (pingdotgg#27851)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(acp): let agents that advertise /compact compact their own context (pingdotgg#28822)

Co-authored-by: ParasN <25337333+ParasN@users.noreply.github.com>
Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* test(acp-registry): subscribe to snapshot changes before triggering them (pingdotgg#35627)

* fix(web): diff colors follow appearance contrast (pingdotgg#28947)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): keep the child thread open when its remote parent was deleted (pingdotgg#27436)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(desktop): poll for updates every 4 hours, not 4 minutes (pingdotgg#28481)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(usage): bill Codex requests whose cumulative counters rewind to an earlier snapshot (pingdotgg#29366)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): persist terminal output without rewriting all history (pingdotgg#24019)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* perf(server): stop serializing terminal chunks just to measure them (pingdotgg#24179)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): stop opening a menu when a terminal selection ends (pingdotgg#24112)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): agents keep named browser profiles after a server restart (pingdotgg#23484)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): reveal full composer command descriptions (pingdotgg#27414)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): restart the service after T3 home ownership clears (pingdotgg#25237)

Co-authored-by: shivam <91240327+shivamhwp@users.noreply.github.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): Enter on an empty prompt resumes an interrupted thread (pingdotgg#27079)

Co-authored-by: Toni Angerer <toni@MBP-von-Toni-5.fritz.box>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(shared): token counts just below a unit boundary use the next unit (pingdotgg#35634)

* fix(desktop): long annotation strokes stay cheap to draw (pingdotgg#24573)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(ssh): load remote profiles with the account login shell (pingdotgg#23934)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): terminals recreate a worktree removed by storage cleanup (pingdotgg#24156)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* test(web): Mod+E bold-word test no longer drops its first arrow key (pingdotgg#35646)

* fix(server): access list keeps changes made while its stream opens (pingdotgg#35640)

* fix(server): a failed Git version probe no longer advises installing Git (pingdotgg#35638)

* fix(server): PR worktrees keep local commits on an existing branch (pingdotgg#24674)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): branch labels keep the space before a middle-truncated name (pingdotgg#23627)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): stale provider update results clear once the provider is up to date (pingdotgg#24338)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* perf(server): merged-PR threads no longer re-read git state every minute (pingdotgg#24099)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): pending request reads keep question option previews (pingdotgg#35644)

* fix(opencode): preserve owned turns during server recovery (pingdotgg#25596)

Co-authored-by: Tin Zeljar <35663757+Zheruel@users.noreply.github.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): baseline existing pull request checks when watching (pingdotgg#23522)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): saved preview snapshot failures name the screenshot destination (pingdotgg#35649)

* fix(ssh): distinguish incompatible Node from a missing executable (pingdotgg#35607)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(pi): stop offering extension-only models for text generation (pingdotgg#24110)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): new worktrees show submodule downloads instead of a stuck checkout (pingdotgg#33687)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): restore composer focus when returning to a thread (pingdotgg#29179)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): linked calls fail fast when a peer stalls the MCP handshake (pingdotgg#28876)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web,mobile): keep line breaks in pending question text (pingdotgg#33353)

Co-authored-by: Grok 4.7 (Grok) <noreply@x.ai>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(mobile): opening a search result clears Home's search on iPhone (pingdotgg#28324)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(desktop): Escape during IME composition no longer cancels a preview annotation (pingdotgg#35645)

* fix(web): a click no longer lands on the sidebar row that just slid under it (pingdotgg#23517)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(mcp): background wait fails when none of its task IDs exist (pingdotgg#29773)

Co-authored-by: Taylor Dolezal <117500+onlydole@users.noreply.github.com>
Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): the collapsed mobile composer shows Stop while a turn runs (pingdotgg#29959)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* test(web): split the cloud voice cancel flow from the failed-then-saved setup (pingdotgg#35665)

* fix(desktop): nightly updates no longer depend on GitHub's tag order (pingdotgg#35599)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(web): PDF preview prerenders pages two viewports below the visible area (pingdotgg#35666)

* perf(server): uploading a large document with seven notebooks finishes 1.8x faster (median 823 ms -> 461 ms) (pingdotgg#33946)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* Typeset equations that fail only because of an unescaped percent (pingdotgg#29511)

Co-authored-by: meutex <342316034+meutex@users.noreply.github.com>
Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Elliot Drel <elliotdrel@gmail.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): wake PR watches when the merge queue drops a pull request (pingdotgg#24141)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): legacy collapsed project folders stay collapsed after a save (pingdotgg#35659)

* fix(server): explain how to close unclaimed browser tabs (pingdotgg#29205)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(editors): JetBrains IDEs open file links in the thread's project (pingdotgg#33558)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(mobile): select part of a source file's text on iOS (pingdotgg#33218)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(mobile): show loading until existing threads are known (pingdotgg#28663)

Co-authored-by: PR Batch Tester <agent@local.test>
Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): streamed device toolbar starts at the panel size (pingdotgg#35664)

* fix(acp): preserve OMP single-choice questions and custom answers (pingdotgg#24012)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): workspace search leaves out files Git ignores outside .gitignore (pingdotgg#35673)

* fix(server): worktree setup no longer hangs when the shell is Nushell (pingdotgg#29461)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(secrets): notify when a delegated agent asks for a secret (pingdotgg#33585)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): Cursor threads keep a stored model the catalog doesn't list (pingdotgg#24872)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): workspace search keeps backslashes in POSIX file names (pingdotgg#35672)

* fix(provider): make readiness probe timeouts configurable (pingdotgg#33275)

Co-authored-by: johan.thomas <johan.thomas@inceptionai.ai>
Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(cursor): mcp.json servers using ${workspaceFolder} now start (pingdotgg#24505)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): workspace search ignore filter follows the host path separator (pingdotgg#35697)

* fix(server): t3 uninstall names another home's service it leaves installed (pingdotgg#35684)

* fix(client-runtime): long MCP app download names no longer split an emoji (pingdotgg#35680)

* fix(web): collapsed reasoning previews no longer split an emoji (pingdotgg#35677)

* fix(web): a question answer restored with undo is no longer sent twice (pingdotgg#35574)

Co-authored-by: ixmxvii-hash <235766220+ixmxvii-hash@users.noreply.github.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): reject PR watches when hosting dependencies are unavailable (pingdotgg#33389)

Co-authored-by: Daniel D <1325654+complynx@users.noreply.github.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(chat): requeued steers no longer look delivered (pingdotgg#23420)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): reuse timeline rows while thinking and plans stream (pingdotgg#24121)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(mobile): Back from a PR opened in a chat returns to the chat on Android (pingdotgg#27653)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(mobile): tall iOS code blocks with long lines show their code (pingdotgg#33243)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(mobile): refreshing a review diff keeps unfolded files open (pingdotgg#35682)

* fix(server): snooze no longer waits two minutes after a steered queued message (pingdotgg#35661)

* test(server): count leaked PR cache caller data without WeakRefs (pingdotgg#35694)

* fix(cli): t3 pair's protocol mismatch hint keeps --base-dir (pingdotgg#35690)

* fix(web): a dismissed provider banner stays dismissed when you switch threads (pingdotgg#25513)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(cli): t3 status no longer suggests --no-lan under a T3CODE_HOST override (pingdotgg#35691)

* fix(desktop): wait for the server's database upgrade instead of reporting a failed start (pingdotgg#35056)

* fix(server): a C# dictionary term no longer rewrites standalone letters (pingdotgg#35700)

* fix(client-runtime): last user message sort ignores agent-posted messages (pingdotgg#33873)

* fix(web): long screen-reader announcements keep emoji whole at the cutoff (pingdotgg#35681)

* fix(server): Claude context meter honours CLAUDE_CODE_AUTO_COMPACT_WINDOW (pingdotgg#33806)

* feat(server): watched GitHub pull requests react when a check suite completes (pingdotgg#35709)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(web): triple-click selects the whole line around links and file chips (pingdotgg#32334)

* fix(projects): preserve grouped renames to repository names (pingdotgg#25244)

Co-authored-by: MacKinley Smith <1738501+mackinleysmith@users.noreply.github.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(acp): agents get T3's runtime notes once per session, not on every message (pingdotgg#26959)

Co-authored-by: Mariusz Mazur <mmazur@redhat.com>
Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* feat(server): keep the computer awake while agents work (pingdotgg#34128)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(web): find scheduled tasks in Settings search (pingdotgg#35715)

Co-authored-by: Brandon Julio Thenaro <brandon.julio.t@icloud.com>

* fix(preview): bound stalled screenshot captures (pingdotgg#25862)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* test(server): idle-resume replays wait for the session release before the next prompt (pingdotgg#35720)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(server): archive refuses a thread with running delegated tasks (pingdotgg#26078)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): minimap Next turn no longer sticks on adjacent user messages (pingdotgg#33036)

* fix(server): Claude background commands show how long they have run (pingdotgg#26668)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): honor worktree prefixes when automatic naming fails (pingdotgg#35718)

Co-authored-by: akash9676 <akashgundapuneni@gmail.com>

* fix(git): refuse remote switches that would silently detach HEAD (pingdotgg#35687)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(mobile): retain inline image zoom on iOS (pingdotgg#33852)

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): Claude's auto-compact setting now wins over CLAUDE_CODE_AUTO_COMPACT_WINDOW (pingdotgg#35737)

* fix(desktop): an SSH reconnect during a desktop update no longer locks the updated app out of its T3 home (pingdotgg#26191)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): auto-balanced drafts recheck machine load before sending (pingdotgg#35658)

Co-authored-by: Lakshmi Tanmay <lakshmi@voltcrash.com>

* fix(web): screen readers at Everything hear context compaction start (pingdotgg#35739)

* fix(server): t3_thread_read returns the newest items and fits in agent context (pingdotgg#33428)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): remote status keeps a local upstream named without a slash (pingdotgg#35766)

* fix(server): a watched PR stops holding its thread once its checks finish (pingdotgg#33214)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(server): a slow database write no longer freezes the server (pingdotgg#35728)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(mobile): Android sheets scroll without per-feature platform checks (pingdotgg#34207)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* docs: drop the GitCafe mirror runbook for a generic forge mirror (pingdotgg#35773)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(ssh): timed-out SSH commands no longer leave a proxy helper running (pingdotgg#35775)

* fix(dev): long-running dev servers no longer leak memory when other worktrees install (pingdotgg#35740)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(server): Changes totals no longer take index.lock (pingdotgg#35770)

* fix(server): keep repository identity for roots whose name ends in a space (pingdotgg#35767)

* fix(server): restored worktrees find pushed thread work on a non-origin remote (pingdotgg#35769)

* fix(ssh): an alias no longer hides known_hosts entries on other ports (pingdotgg#35779)

* fix(server): IPv4-mapped loopback SSH device targets count as this machine (pingdotgg#35774)

* fix(ssh): host discovery follows Include globs in directory components (pingdotgg#35771)

* fix(server): Commit finishes a merge resolved to the current branch's tree (pingdotgg#35778)

* fix(server): gh default remote lookup ignores an inherited GIT_DIR (pingdotgg#35764)

* docs: FAQ says the Claude 5-hour limit drain is fixed (pingdotgg#31473)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(server): fork PR worktrees no longer fail when a local branch owns the t3code/pr-<n> prefix (pingdotgg#35768)

* fix(web): a failed commit keeps the commit message you typed (pingdotgg#35781)

* fix(server): branches whose names start with a parenthesis are no longer reported as detached (pingdotgg#35780)

* test(ssh): launcher harness no longer races its background nohup (pingdotgg#35788)

* fix(web): Resume replaces Stop without a lingering Stopped button (pingdotgg#35755)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* test(web): typecheck the project file refresh test (pingdotgg#34061)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* docs(codex): explain native image generation setup (pingdotgg#31426)

Co-authored-by: David Preil <dzpreil@gmail.com>
Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* perf(shared): command output failure scan rejects clean output in one pass (pingdotgg#26935)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* test(web): sidebar drag test opens its folder past the shift guard (pingdotgg#35803)

* feat(mobile): secret field gets an eye toggle to show what you typed (pingdotgg#35783)

* feat(mobile): secret field gets an eye toggle to show what you typed

* fix(mobile): a cold launch shows the cached thread rows with the first frame (pingdotgg#35683)

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

* fix(server): ACP terminal MCP calls answer interleaved server requests (pingdotgg#35805)

* fix(web): "Try again" recovers from a failed page chunk load (pingdotgg#35806)

* fix(shared): read MCP event streams whatever the case of the content type (pingdotgg#35815)

* test(web): sidebar drag test ignores a fading row clone after folder collapse (pingdotgg#35817)

* fix(server): ACP streamed chunks compare message ids exactly (pingdotgg#35836)

* fix(install): name the archive when its download fails (pingdotgg#35854)

* test(server): drop the flaky Kiro provider_thread_resume replay (pingdotgg#35867)

* fix(web): arrow and page keys scroll the conversation again after a click or F6 (pingdotgg#27209)

* fix(kiro): compaction is refused when Kiro does not advertise _kiro/session/compact (pingdotgg#35866)

* fix(server): show self-resumed Claude subagents as running (pingdotgg#23831)

* fix(web): preserve single tildes in sent user messages (pingdotgg#35859)

* fix(desktop): clicking inside the in-app browser closes open host menus (pingdotgg#23141)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(desktop): update browser tab titles without navigation (pingdotgg#29452)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web,mobile): triage follows the draft when you switch machines (pingdotgg#33782)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): storage cleanup treats symlinked spellings of a worktree as one checkout (pingdotgg#25735)

* fix(server): migrate-dev-db refuses unimported legacy threads (pingdotgg#23191)

* fix(web): opening a thread on a touch tablet no longer focuses the composer (pingdotgg#29496)

* fix(acp): linked terminal output no longer reopens a completed tool call (pingdotgg#35844)

* fix(acp): root tool content replacements replace streamed output (pingdotgg#35862)

* fix(forgejo): fast-forward-only repositories can merge pull requests (pingdotgg#32738)

* fix(server): linking says why each address failed discovery (pingdotgg#30884)

* fix(terminal): show recovery guidance when a project folder is missing (pingdotgg#35849)

* fix(client-runtime): reopened files show host edits instead of the first read (pingdotgg#29409)

* fix(server): preview resize and status report the real size of desktop tabs (pingdotgg#32854)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(server): migration 64 skips thread payloads that are not valid JSON (pingdotgg#35884)

* fix(web): thread navigation shortcuts work from the terminal and macOS composer (pingdotgg#26433)

* fix(desktop): an unreadable settings file is no longer overwritten by a bounds save (pingdotgg#35890)

* fix(codex): accept hook event names newer than the pinned protocol (pingdotgg#35885)

* fix(opencode): /goal clear with no goal set finishes as a no-op (pingdotgg#35893)

* fix(ssh): a slow reconnect no longer replaces a healthy port forward (pingdotgg#35874)

* fix(desktop): one preview viewer or recording stopping no longer freezes the others (pingdotgg#31051)

* fix(server): Codex question answers keep their native question id (pingdotgg#35886)

* test(server): windowed history pages are checked against the full index at the window edges (pingdotgg#35877)

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

* fix(clients): answer history no longer shows provider context from review comments (pingdotgg#35897)

* fix(server): refresh sync status after agent merges (pingdotgg#35919)

* fix(web): copying an image with a spaced path keeps it an image (pingdotgg#35914)

* fix(server): Force stop no longer tears down a queued follow-up that started during cleanup (pingdotgg#31038)

* fix(azure-devops): TF400733 throttling fails diffs as rate-limited (pingdotgg#35929)

* fix(acp): v2 tool updates clear nulled fields and replace raw input (pingdotgg#35869)

* fix(server): interrupted SSH device bootstrap no longer leaks its forward (pingdotgg#35928)

* fix(mobile): Files search no longer carries into another workspace (pingdotgg#35902)

* fix(web): copying an alert callout keeps its alert Markdown (pingdotgg#35932)

* fix(server): Codex shadow home shares the instance environment CODEX_HOME (pingdotgg#35912)

* fix(preview): remote viewers recover from stalled first captures (pingdotgg#25381)

* fix(server): new turns wait for overlapping checkpoint file restores (pingdotgg#23671)

* fix(server): preview_navigate returns the committed page when load is slow (pingdotgg#26127)

Co-authored-by: Elliot Drel <elliotdrel@gmail.com>

* fix(server): unchanged-worktree cleanup waits for the thread to settle or archive (pingdotgg#35941)

* fix(web): zoom shortcuts work while a desktop browser page has focus (pingdotgg#35947)

* fix(server): ACP threads accept a model change after their session detaches (pingdotgg#35951)

* fix(server): service launcher keeps a T3 home that ends in a space (pingdotgg#35958)

* fix(server): launcher repoints the unit the systemd manager loads (pingdotgg#35966)

* fix(server): refuse a Linux service for a T3 home systemd cannot load (pingdotgg#35968)

* perf(server): sending, editing and steering queued messages no longer read a long thread's whole history (pingdotgg#29521)

* fix(server): Codex protocol records go to the log of the thread they belong to (pingdotgg#35975)

* fix(cli): service.env updates keep quoted multiline values intact (pingdotgg#35976)

* fix(web): open code blocks and diffs follow a changed word wrap default (pingdotgg#35922)

* fix(desktop): keep WSL distros whose names start or end with a dot or hyphen (pingdotgg#35971)

* test: run clock-free live tests on TestClock (pingdotgg#35868)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(server): SSH device test rejects Node below agent-device's 22.12 floor (pingdotgg#35991)

* fix(server): Simlock renewals no longer exceed the gateway's lease TTL cap (pingdotgg#35988)

* test: drop Effect.scoped wrappers that @effect/vitest already applies (pingdotgg#35993)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(device): Simlock viewer stops retrying an ended lease and keeps its reason (pingdotgg#35994)

* fix(server): Stop waits for the turn to settle instead of polling (pingdotgg#35948)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* ci: label issues from vouched and denounced authors, denounce coygeek (pingdotgg#36021)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* fix(scripts): dev runner ignores the parent T3 server's port and host (pingdotgg#36019)

* perf(server): reduce allocation while reading usage history (pingdotgg#36009)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: GPT 6.1 Sol <codex@openai.com>
Co-authored-by: Grok 4.7 <noreply@x.ai>

* fix(web): terminal paste keeps a leading U+FEFF without bracketed paste (pingdotgg#36018)

* fix(server): BEL inside a DCS no longer ends it in terminal history (pingdotgg#36020)

* fix(server): terminal history drops the ESC Z identity query (pingdotgg#36024)

* test(server): blocked pointer-file scan test waits for the unblocked read (pingdotgg#36031)

* fix(web): terminal OSC 8 file URIs open the file they name (pingdotgg#36025)

* fix(server): terminal history strips queries with embedded controls (pingdotgg#36027)

* fix(mobile): pasted images over the size limit attach instead of vanishing (pingdotgg#35845)

* fix(mobile): Android terminal resizes the engine when only the cell size changes (pingdotgg#36023)

* fix(server): t3_thread_read results fit MCP clients' output limit (pingdotgg#33455)

* fix(desktop): AppImage t3 shim no longer grows the CLI compile cache on every call (pingdotgg#33885)

* ci: agent context budget runs are no longer cancelled by label bursts (pingdotgg#36049)

* fix(server): a decoded U+009C no longer ends an OSC string in terminal history (pingdotgg#36022)

* fix(server): speech worker no longer segfaults when the backend dies mid-dictation (pingdotgg#33624)

* fix(codex): image items identified by fileId are no longer dropped (pingdotgg#35883)

* perf(server): linked-thread counts update when links change instead of polling every 10s (pingdotgg#35865)

* fix(scripts): scenarios reach the screen and say why they could not run (pingdotgg#36054)

* docs: AGENTS.md and the FAQ list Pi, Muse Code, Kiro, and fx (pingdotgg#33906)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): PR Code tab previews images the host sends no patch for (pingdotgg#36056)

* fix(web): pull request comment tooltips follow the Time format setting (pingdotgg#35396)

* fix(server): repository identity honors the gh default remote (pingdotgg#21574)

* fix(server): a rejected non-fast-forward push explains itself under a translated Git (pingdotgg#35386)

* fix(web): provider rows select from anywhere in the row (pingdotgg#32537)

* fix(client-runtime): imported threads that never ran no longer sit in Working as Waiting (pingdotgg#33042)

* fix(web,mobile): MCP app calls keep their tool row and result beside the app card (pingdotgg#33347)

* fix(forgejo): tea API failures keep Gitea's reason (pingdotgg#32244)

* fix(server): reap recorded Linux Claude survivors before recovery (pingdotgg#22834)

* fix(web): open documents no longer reload after every agent command (pingdotgg#32693)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Yash Singh <saiansh2525@gmail.com>

* fix(desktop): AppImage quit no longer leaves zygotes running on an unmounted image (pingdotgg#36057)

* fix(server): a message sent after a stopped worktree setup waits instead of running in the project checkout (pingdotgg#23618)

Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Julius Marminge <julius0216@outlook.com>

* test(server): terminal checkout test waits for the refresh's last event (pingdotgg#36071)

* fix(provider-core): Vite+ provider updates respect npm's release-age policy (pingdotgg#32340)

* fix(web): point pairing copy at Sharing & access (pingdotgg#27692)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): turn diffs cut off at the size limit say changes are incomplete (pingdotgg#36079)

* fix(web): narrow-window right panel dialog is named Workspace panel (pingdotgg#36085)

* fix(server): thread reads preview linked PRs and list_thread_pull_requests pages (pingdotgg#35858)

* fix(web): new project publish option tells screen readers whether it is on (pingdotgg#36090)

* fix(server): preview drag moves the mouse during the drag (pingdotgg#36096)

* fix(server): a fresh-session scheduled run no longer takes over the thread's conversation (pingdotgg#36103)

* fix(server): deferred-steer check and queued cancel no longer read a thread's whole history (pingdotgg#36100)

* fix(composer): keep reviews from different Git hosts in suggestions (pingdotgg#36104)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

* fix(web): terminal "Add to chat" now adds to a queued message being edited (pingdotgg#36117)

* fix(mobile): a final answer followed by more work in the same run stays visible (pingdotgg#36116)

* fix(web): a server updated with the app reconnects as soon as it is ready (pingdotgg#35327)

Co-authored-by: shivam <91240327+shivamhwp@users.noreply.github.com>

* fix(acp): Copilot runs end when its reply is done, not when its async shells exit (pingdotgg#36122)

* fix(server): a link forgotten mid-resolve says it is not linked (pingdotgg#36132)

* fix(server): t3_queue and handoff analytics no longer read a thread's whole history (pingdotgg#36130)

* perf(server): shell reads of a worktree or family no longer parse every thread payload (pingdotgg#36146)

* fix(desktop): Local environment off survives an unreadable value in desktop-settings.json (pingdotgg#36168)

* fix(mobile): a sub-agent opened from its parent on a narrow iPad keeps Back (pingdotgg#36175)

* fix(server): Codex usage answering after three seconds is no longer dropped (pingdotgg#36189)

* fix(mobile): iOS terminal font and theme changes no longer replay output (pingdotgg#36115)

Co-authored-by: Claude Haiku 5.5 <noreply@anthropic.com>

* fix(server): stop device daemon alongside slow hub shutdown (pingdotgg#33566)

Co-authored-by: maria-rcks <maria@kuuro.net>

---------

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: maria-rcks <maria@kuuro.net>
Co-authored-by: Yukun Shan <92423096+nateEc@users.noreply.github.com>
Co-authored-by: samkelly880 <hawk@invert.nz>
Co-authored-by: ParasN <prsnagpal@gmail.com>
Co-authored-by: ParasN <25337333+ParasN@users.noreply.github.com>
Co-authored-by: Lakshmi Tanmay <lakshmi@voltcrash.com>
Co-authored-by: Scott Norteman <snorteman@gmail.com>
Co-authored-by: Jintao Zhang <zhangjintao9020@gmail.com>
Co-authored-by: Joaquim Peres <79022517+jokaperes@users.noreply.github.com>
Co-authored-by: Aaron Queen <bompus@users.noreply.github.com>
Co-authored-by: fixfon <12512757+fixfon@users.noreply.github.com>
Co-authored-by: Victor <satsdev@only21mil.xyz>
Co-authored-by: shivam <91240327+shivamhwp@users.noreply.github.com>
Co-authored-by: toni92-beep <toniangerer92@gmail.com>
Co-authored-by: Toni Angerer <toni@MBP-von-Toni-5.fritz.box>
Co-authored-by: tatsuaki watanabe <tatsuaki.w@gmail.com>
Co-authored-by: Antony <97451137+TonybynMp4@users.noreply.github.com>
Co-authored-by: Artem Gurianov <gurianov.artem@proton.me>
Co-authored-by: Batuhan1999 <Batuhan.Ckr@outlook.de>
Co-authored-by: Tin Zeljar <35663757+Zheruel@users.noreply.github.com>
Co-authored-by: Vaibhav Pathak <45169573+vkpdeveloper@users.noreply.github.com>
Co-authored-by: Mason <43073153+MasonNight@users.noreply.github.com>
Co-authored-by: Godfrey Hendrix <samuelgodfreyhendrix@gmail.com>
Co-authored-by: Wout Stiens <71498452+StiensWout@users.noreply.github.com>
Co-authored-by: Grok 4.7 (Grok) <noreply@x.ai>
Co-authored-by: Mantas Kantautas <mantas.kantautas@gmail.com>
Co-authored-by: Taylor Dolezal <onlydole@users.noreply.github.com>
Co-authored-by: Taylor Dolezal <117500+onlydole@users.noreply.github.com>
Co-authored-by: TanJeeSchuan <89920999+TanJeeSchuan@users.noreply.github.com>
Co-authored-by: Theo Browne <me@t3.gg>
Co-authored-by: Andrei <108398860+Yuruzuu@users.noreply.github.com>
Co-authored-by: Meutex <admin@meutex.com>
Co-authored-by: meutex <342316034+meutex@users.noreply.github.com>
Co-authored-by: Elliot Drel <elliotdrel@gmail.com>
Co-authored-by: Bin Jamshid <gut.commune-0e@icloud.com>
Co-authored-by: PR Batch Tester <agent@local.test>
Co-authored-by: Ken Johnson <5758378+KenDJohnson@users.noreply.github.com>
Co-authored-by: Stanislav Zmiev <zmievsa@gmail.com>
Co-authored-by: Johan John Thomas <90568542+johanjohnthomas@users.noreply.github.com>
Co-authored-by: johan.thomas <johan.thomas@inceptionai.ai>
Co-authored-by: Logan <ixmxvii@gmail.com>
Co-authored-by: ixmxvii-hash <235766220+ixmxvii-hash@users.noreply.github.com>
Co-authored-by: Daniel <d.drizhuk@gmail.com>
Co-authored-by: Daniel D <1325654+complynx@users.noreply.github.com>
Co-authored-by: Alb11747 <albertli_042@hotmail.com>
Co-authored-by: vivek <mullangivivekananda@gmail.com>
Co-authored-by: Muzammil Hussain <58479501+Muzammil-Bit@users.noreply.github.com>
Co-authored-by: Tomasz Tórz <93620601+torztomasz@users.noreply.github.com>
Co-authored-by: MacKinley Smith <1738501+mackinleysmith@users.noreply.github.com>
Co-authored-by: Radu Stefan Popa <150839549+Radu028@users.noreply.github.com>
Co-authored-by: Mariusz Mazur <mmazur@redhat.com>
Co-authored-by: Brandon Julio Thenaro <brandon.julio.t@icloud.com>
Co-authored-by: Marcos Caballero <98906745+cabamarcos@users.noreply.github.com>
Co-authored-by: Fuksss <fuxik12345@gmail.com>
Co-authored-by: Dara Adedeji <76637177+SunkenInTime@users.noreply.github.com>
Co-authored-by: akash9676 <akashgundapuneni@gmail.com>
Co-authored-by: kirat02 <142911687+Kirat0201@users.noreply.github.com>
Co-authored-by: Alex Southwell <saphid@gmail.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: AJ <114173633+ajbarryiii@users.noreply.github.com>
Co-authored-by: Aviddaram <130821636+darjss@users.noreply.github.com>
Co-authored-by: sharath <29162020+tnfssc@users.noreply.github.com>
Co-authored-by: Linus Höhne <linus.hoehne@icloud.com>
Co-authored-by: Max Schwenk <maschwenk@gmail.com>
Co-authored-by: dpreil-ydlp <73763174+dpreil-ydlp@users.noreply.github.com>
Co-authored-by: David Preil <dzpreil@gmail.com>
Co-authored-by: Hieu Le <78847613+hieu-lee@users.noreply.github.com>
Co-authored-by: voida6 <kainth.ravdeep@gmail.com>
Co-authored-by: Ilkka Lappeteläinen <38039015+pnupu@users.noreply.github.com>
Co-authored-by: Entrpi <133157614+Entrpi@users.noreply.github.com>
Co-authored-by: Yash Singh <saiansh2525@gmail.com>
Co-authored-by: GPT 6.1 Sol <codex@openai.com>
Co-authored-by: Gourav Kumar Dhiman <gkrdhiman@gmail.com>
Co-authored-by: Pedro Navarrete <navhpedro@gmail.com>
Co-authored-by: Simon Maggini <simonmaggini@gmail.com>
Co-authored-by: Bilal Bakr <62337003+Bil0000@users.noreply.github.com>
lidge-jun added a commit to goat-si/goat that referenced this pull request Nov 16, 2026
…gh 93cbca3fcbd) (#3397)

* fix(server): a Codex skills scan reads its probe timeout from the environment it launches with

snapshotForCwd has read the host environment per scan since #3207, but its
timeout still resolved GOAT_CODEX_PROBE_TIMEOUT_MS from the environment captured
when the driver was created, so an override set later did not extend the scan.
The timeout now reads the same per-scan environment.

(cherry picked from commit 844f142d1c708a64b4d522f4cff2891ced6315be)

* refactor(server): move Claude tool frame helpers out of the adapter

Moves the Claude tool classification, tool input and output, tool result
frame and background task frame helpers (667 lines) from ClaudeAdapterV2.ts
to ClaudeToolFrames.ts so the next ports fit the adapter's line budget.
The moved lines are byte-identical; a script check found zero line
mismatches. The new module lists the names the adapter uses in one export
block at its end, and the adapter imports them from it.

* test(server): the stopped-child read test holds that the parent's wake is dispatched

The F2 test ("a summary read of a stopped task child whose queued
notice Stop cancelled leaves its completion delivery owed") held that
the read acknowledged nothing through the delivery's stored state.
e681c415b9 changed that expected state from "pending" to "claimed".
No behaviour change made "pending" wrong: it was a race in the test.
The orchestrator's run-update reactor, a forked fiber, finalizes the
stopped task and claims its delivery, and the test read the state
before that fiber had run. a909ea423c's window reads gave the fiber
time to run first and exposed it.

The test now also holds that the claimed delivery is a real wake: in
the fixture's layer, continuation requests queue instead of dropping
(layerTestWithContinuations), and the test takes the request that
offers the delivery's message, generation and parent run to the
parent's still-active run.

* refactor(server): move provider session manager errors to their own module

Moves the release reason schema and the provider session manager error
classes (79 lines) from ProviderSessionManager.ts to
ProviderSessionManagerErrors.ts so the provider retry stall watch fits
the manager's line budget. The moved lines are byte-identical (script
check: zero line mismatches). ProviderSessionManager.ts re-exports every
name, so callers keep their imports.

* test(server): the native-child settlement case shows that a working child settles with its parent

The drain case "drains idle provider-native children of a settled
parent and retries busy ones without PR lookups" gave its busy child a
running app run. A provider-native child at work has none: it works in
a runless root turn under its parent's run, and the parent records it
as an active subagent row. Only a person's own message on the child is
an app run.

The case, now "drains provider-native children of a settled parent
without PR lookups, one still working included", gives the working
child that active subagent row and no run. It holds what happens: the
child settles with its parent, as upstream, while its row on the
parent stays running. This is a known limit, by the coordinator's
decision of 2026-11-15. The options to change it are in goat-si/goat#3407.

* refactor(server): move the thread pull request link helpers out of the orchestrator

withPullRequestWatch and legacyPullRequestLink move byte-identically into threadPullRequestLink.ts so Orchestrator.ts stays within its 13908-line budget after the lane merge.

* test(server): move the Claude attachment tests to their own file

Moves the "ClaudeAdapterV2 attachments" describe block (599 lines) from
ClaudeAdapterV2.test.ts to ClaudeAdapterV2.attachments.test.ts so the
next ports fit the test file's line budget. The moved lines are
byte-identical (script check: zero line mismatches). The new file carries
its own copy of the Claude SDK resolveSettings mock, the adapter test
layer alias and the default settings constant. 12 tests pass there.

* test(provider-pi): move the Pi RPC framing test to its own file

Moves the "PiRpc framing" describe block (70 lines) from adapter.test.ts
to rpc.framing.test.ts so the retry recovery test change fits the file's
line budget. The moved lines are byte-identical (script check: zero line
mismatches).

* fix(server): fail any provider turn stuck in a retry

A provider can announce a retry and never resume, which left the turn
running forever. The provider session manager now watches each running
retry row: when the retry's thread stays silent for 15 minutes plus the
announced retry delay, it stops the turn and reports the stop as the
retry's failure (not retryable). Output on the retry's thread restarts the
window. Claude now settles a retry when the API starts a new message, and
Pi when it streams output, because both keep the retry row running while
a recovered reply streams. The watch lives in ProviderRetryStallWatch.ts
so ProviderSessionManager.ts stays inside its budget, and the manager test
registers ProviderRetryStall.testkit.ts. GOAT has no prompt suggestion
branch in the event pump, so only the command output branch reads the raw
event.

Upstream: 00766e2c027ac095e604da81b958113287f2f6b1
Decision: adapt

* fix(server): forget unanswered Claude ScheduleWakeup calls on reset and exit

The adapter records each ScheduleWakeup tool call until its result
arrives. A reset or a stop that drops the scheduled wakeup now also
forgets those calls, so a call whose result never came does not stay in
the set for the life of the session. The set holds tool use ids only, so
nothing user-visible changes; no test.

* test(server): a steer refused while its memory refresh waits is never offered

GOAT refreshes memory between the first turn check in steerTurn and the
steer's registration, so steerTurn checks steersClosed again before it
registers the steer. The new test holds the memory refresh on a Deferred,
stops the turn, then releases it: the steer fails, no pending status is
emitted and nothing is offered. With the second check removed the test
fails ("expected false to be true" on the steer's failure).

* fix(server): a steer that cuts off a streaming Claude reply marks it interrupted

When a steer aborts Claude mid-stream (terminal reason aborted_streaming),
the reply Claude was writing used to stay as a completed answer. The
adapter now remembers the root reply the latest assistant frame carried
and republishes it as interrupted when the steering abort result
arrives. The web timeline shows an "interrupted" badge on such a reply
even when it is not the turn's last message.

GOAT emits root replies from the text projector's block snapshot, so the
reply is recorded there (non-streaming blocks only) and cleared on every
root assistant frame; emitAssistantTextArtifacts gains an "interrupted"
option and returns its start time instead of upstream's separate publish
helper, and buildAssistantArtifacts (ClaudeAssistantArtifacts.ts) takes
an optional status next to GOAT's streaming flag. Before and after
captures are on #3397 (browser harness with the app stylesheet).

Upstream: c7618f6c10120bc7e684df6a0d7275679d8c260e
Decision: adapt

* test(server): move the Antigravity driver test harness into a testkit

AntigravityDriver.test.ts sat at the 1000-line limit and the managed CLI update port adds to its launch expectation. The host flags, request decoder, blocked credential keys, shellQuote and makeHarness move byte-identically into AntigravityDriver.testkit.ts; only unused imports are pruned.

* feat(antigravity): managed installs update to Google's verified native CLI

GOAT's managed Antigravity runtime stayed on its pinned 1.4.5 release. It now
looks up Google's CLI release manifest (at most hourly while provider update
checks are on and setup is watched, and on every Install or Update) and offers
the newer release in provider settings; nothing installs until the user picks
Update Antigravity.

The download is installed only after its SHA-512 matches the manifest, from an
allow-listed Google host, with redirects refused and download and extraction
capped. An unreachable or unusable manifest offers nothing and the pinned
release stays the fallback with its exact-size SHA-256 check; a release the
model manifest withdrew is neither offered nor installed, and an installed
native CLI newer than the offer is re-checked rather than downgraded. The
native CLI runs as agy acp with its own --gemini-dir and --app-data-dir, and
reports the system prompt append through its initialize metadata.

The installer is split into Records, Lookup, Validate and Archive modules plus
antigravityCliRelease.ts so each stays under 800 lines. Upstream's unhashed
registry path (315b06c5ec) is not ported.

Upstream: 4099e62767f725937aea6ab2f43ba44b8e4b651e Decision: adapt (managed CLI updates)

* refactor(server): move the filesystem browse failure mapping out of ws.ts

filesystemBrowseFailureContext and unexpectedCompatibilityError move unchanged (the formatter rewraps the exported signature) into workspace/filesystemBrowseFailure.ts, so ws.ts has room for the secret check handler under its budget.

* test(server): move the provider maintenance runner fixtures to a testkit

Moves the driver and instance ids, provider snapshots, mock spawner, mock
registry and runner builders (238 lines) from
providerMaintenanceRunner.test.ts to providerMaintenanceRunner.testkit.ts
so the installation gate tests fit the test file's budget and can live in
their own file. The moved lines are byte-identical (script check: zero
line mismatches); the testkit lists its exports in one block at the end.
31 tests pass.

* fix(server): worktree cleanup and removal by thread never take a checkout GOAT did not make

A thread can point at a worktree outside GOAT's worktrees folder, such as the
user's own linked checkout an imported session records. goat_worktree_remove
by threadId removed it (even with force) and goat_worktree_cleanup swept it
whenever a finished thread was bound. Both now refuse such paths.

* fix(server): provider launches and text generation wait for provider updates

A provider update could replace the CLI while a session was opening or a
text generation call was running it. A new ProviderInstallationGate lets
provider work run concurrently per instance and makes an update wait for
that work to drain while new work queues behind it. The update holds every
instance of the same driver whose update shares the lock key, so other
homes or accounts on the same installation wait too; another driver in
the same npm prefix does not. The update status now says "Waiting for
running provider work to finish." while it drains. The provider session
manager gates opening a session, the Claude adapter gates opening each
turn's query process, and text generation gates each call.

GOAT reads the gate with Effect.serviceOption (ProviderInstallationGate
.current), the same way it reads ProviderCliGate, so runtimes built
without one, such as most test layers and the hydration tests, run
ungated; the server, the runtime layer's text generation and the replay
harness provide the one shared gate, and the maintenance runner requires
it. GOAT's TextGeneration has one method per operation instead of
upstream's generate, so each method is gated by its model selection's
instance. The two maintenance gate tests are in
providerMaintenanceInstallationGate.test.ts and the Claude one in
ClaudeInstallationGate.test.ts. Upstream's ProviderRegistry.test.ts,
ProviderInstanceRegistry.test.ts, CodexTextGeneration.test.ts and
ClaudeAdapterV2.testkit.ts hunks only provided the gate and are not
needed with the optional read.

Upstream: 9ce9b1fd419538fa32cd251e9e490a360877259a
Decision: adapt

* fix(server): deleting a thread never removes a worktree GOAT did not make

Importing a session saves its linked worktree as the thread's worktreePath so
turns resume in that checkout. Deleting the thread on web or desktop then
offered "Delete the worktree too?" and ran a forced git worktree remove on the
user's own checkout, losing uncommitted work.

vcs.removeWorktree now refuses any folder outside GOAT's worktrees folder, so
older clients get a clear error instead of a removal. getWorktreeThreadIds
reports whether GOAT made the worktree, and web and desktop offer its removal
only when the server says it did.

* docs(user): say imported worktrees stay and that GitLab can open a revert

* feat(secrets): check a secret reference from Settings

A saved 1Password reference only showed it was wrong when a provider failed to start. Settings
now offer Check beside a saved reference. server.checkSecretSource reads it once through GOAT's
resolver, bypassing the cache, and answers Resolved or Failed with a reason; the value never
leaves the server. A readable secret clears the resolver cache and reloads the provider
instances that use the reference.

The method requires settings:write (CLIENT_GUARDED_RPC_SCOPES); servers advertise it with the
optional secretSourceCheck capability, and clients hide Check where it is absent, so a newer
client never calls an older server. The OTLP check RPC moves with it into
contracts/serverCheckRpc.ts to keep rpc.ts within its budget. Mobile has no secret reference
settings, so it is unchanged.

Upstream: ee9e36c8b20e739f107bdbacea82d12f912d97b1 Decision: adapt (GOAT secret resolver)

* fix(server): a cancelled queued Claude continuation no longer drops later wakes

When the continuation the adapter offered for a Claude background wake was
cancelled before its run started, the wake's latch stayed set, so every
later wake on that native thread was swallowed. The latch now remembers
the wake's last frame and whether the cancelled wake had opened. A
cancelled wake's buffered frames are retired (subagent starts that are
still running stay), its root frames are dropped until it ends, and the
latch is released so the next wake gets its own continuation; a wake
Claude never opens is released by a later task's notification. The
continuation service tells the adapter when a run ended without starting
(onRunEndedWithoutStarting).

GOAT keeps wake reports in claudeWakeNotice.ts, so it gains
peekWakeReports and dropWakeReports for the retirement path, and the wake
offer keeps GOAT's Wake.claudeWakeNotification. The adapter tests live in
ClaudeWakeRetirement.test.ts (with the wake harness queue capture mock);
upstream's awaitContinuation and until become awaitUntil on the
continuation request count and state. The integration test is in
ClaudeAutomaticDelivery.integration.test.ts with GOAT's Crypto service.

Upstream: ccc9aaf10b5f40382d0ca68b8523aa2f434c841f
Decision: adapt

* fix(server): archiving many linked threads sends each open list one refresh

Each pull request list subscriber refetches on a linked-thread signal, and the
signals queued without limit, so archiving N threads cost N refetches on every
open list. Each subscriber now keeps only the newest revision while it is busy.

* fix(contracts): an auto-merge method a newer host adds no longer breaks pull request reads

autoMergeMethod shares the merge-method literals, which just gained fast-forward;
a client that did not know a host value failed the whole detail or list. It now
decodes an unknown method as absent, as the merge-method lists already do. Also
lowers the GitHubPullRequestApi.test.ts budget to its length.

* refactor(server): move the orchestrator command thread id lookup to its own module

Moves commandThreadId (84 lines), the switch that names the thread a
server command targets, from Orchestrator.ts to
orchestratorCommandThreadId.ts so the delegated completion recovery port
fits the orchestrator's line budget. The moved lines are byte-identical
(script check: zero line mismatches).

* test(server): move the permission profile test fixtures into a testkit

outside, threadScope, fixture and seed move unchanged into PermissionProfiles.testkit.ts so a lane test can share them without growing PermissionProfiles.test.ts past its limit.

* test(server): type the worktree removal RPC test mocks

* test(server): give the Claude background task replay fixture its own module

Moves the claude_background_task_after_root entry of the replay fixture
list into claude_background_task_after_root/fixture.ts, the shape
acp_titled_multiselect already uses, so the delegated result port can add
a field without growing testkit/fixtures/index.ts past its budget. The
entry's fields and values are unchanged; only its indentation and the
transcript URL (now relative to the fixture folder) differ.

* test(server): move the host thread list merge tests out of HostThreadsMcpService.test.ts

The mergeHostThreadLists cases move unchanged into HostThreadsMcpService.merge.test.ts and their item builder into hostThreadsList.testkit.ts, so the forwarding tests have room under the file budget.

* fix(server): preserve delegated results across wake turns

A delegated Claude child whose background command woke it after its
first result could lose that result: the background task left the
roster as soon as the command exited, and a completion whose live run
update was missed (for example on a checkpoint capture retry) never
offered its wake or finalized its delivery. The adapter now keeps an
exited background command on the roster until its wake run starts. The
effect worker recovers the delegated task after each completion
checkpoint (recoverDelegatedTask with completedRunId), which re-offers
the completion delivery and finalizes it, and restart recovery passes
sourceRunId. recoverDelegatedTask reports projection failures instead of
logging them, and skips children with nothing pending
(ProjectionStore.hasPendingDelegatedCompletion).

GOAT's replay projection store keeps no completion delivery state, so its
hasPendingDelegatedCompletion answers true for any app-owned child and
never skips a recovery. The rewritten delivery tests are in
DelegatedCompletionDelivery.checkpointRetry.test.ts (the old single-case
test is replaced by its live, checkpoint-retry and checkpoint-offer-retry
cases), the shared delivery layer gains EffectOutbox.layer, GOAT's restart
test passes sourceRunId, and the failed-dispatch release check is in
ClaudeWakeDispatch.test.ts.

Upstream: 7462f234e89027943cf0d1a73d3ae1f36571d4e3
Decision: adapt

* feat(mcp): agents see a Stop the provider never confirmed

When a provider answers Stop but never ends the turn, clients show "Not responding to Stop" and
offer Force stop, but goat_thread_status, goat_thread_read and goat_task_status reported only a
running thread. They now carry an optional stop: { state: "unresponsive", since } built from the
run_interrupt_request's unresponsiveAt that clients already read, for the described run of any
provider (Claude, ACP, OpenCode and Codex reach this state). The descriptions say only a person
can Force stop today (goat#2338).

The field is optional both ways: a host that predates it omits it, including through the
thread-read fallback for hosts without goat_thread_status, and a caller that predates it ignores
it. goat_thread_interrupt and task_cancel keep returning interrupt_requested and cancel_requested.

task_cancel's description now says no wake follows a cancel and to read the child with
task_status; dropping the completion delivery is its existing contract on dev for every
provider. A lane test pins that an unanswered Stop keeps the child's lane until its run ends.

* fix(codex): a Stop Codex never confirms keeps the turn open and offers Force stop

When Codex answers turn/interrupt but never ends the turn, Stop now fails with
ProviderAdapterInterruptTimeoutError after 10 s instead of reporting success. The
turn stays open and marked interrupting, so clients show "Not responding to Stop"
and offer Force stop; whatever ends the turn later (Codex, Force stop's unload)
settles the rest of the Stop. A second Stop joins the first one's waiter instead
of stranding it, the survivor error mapping passes the timeout through, and GOAT's
failed-Stop path still leaves the subagents it reached running. GOAT's start guard
stays the fallback: after Force stop detaches the thread, the next run stops a turn
Codex still runs before it sends turn/start.

Upstream: b5b470542a5fb95aed16f3d3a23f8cd8de17e5d8 Decision: adapt

* docs(user): an imported worktree stays when its thread is deleted or worktrees are cleaned up

* test(antigravity): pin the native CLI manifest parser and its host allow-list

Upstream's antigravityCliRelease.test.ts (4099e62767): each platform's target and file, and the refused metadata (a bad digest, a pre-1.5.0 version, http, a lookalike host, another bucket or platform folder, query strings, a version mismatch). The manifest base stays private to the module.

* test(web): let the secret check fixture leave its capability unset

* docs(internals): regenerate the agent surface audit for the stop field and task_cancel wording

* test(server): align two GOAT-only tests with delegated wake recovery

The full gate after 7462f234e8 failed two tests that GOAT added before the
port and that encode the old behaviour:

- ClaudeBackgroundDetails.test.ts expected a background command to leave
  the roster as soon as a snapshot omitted it. Since 7462f234e8 an exited
  command stays listed until its wake run starts, as upstream's own
  roster tests now assert, so the test expects task-1 to remain.
- OrchestratorMcpToolkit.integration.test.ts expected exactly one
  delegated completion offer. Recovery after the completion checkpoint
  can offer the same delivery again (same parent run, generation and
  message); the continuation service dispatches it only while the
  delivery is open and undelivered. The test now requires every offer to name the same
  delivery message and keeps its single-delivery assertions, which
  pass.

Both pass with the old expectations on the integration tip bd69c4db0b,
which lacks 7462f234e8, and fail on this lane without this change.

* refactor(server): move the native fork recovery tests into ThreadFork.nativeRecovery.test.ts

ThreadFork.execution.test.ts was at 976 lines; the next recovery case would take it past the
1000-line cap. The helper and its two tests move byte-identically; only the imports differ.

* fix(server): native fork recovery leaves rewound turns out of the recovered history

The fallback read every item of the source's runs and of its fork ancestors' runs up to the
fork point, so the fresh session's handoff carried turns the user had rewound with Edit from
here, and inherited ones had no run status at all. Each segment is now filtered with its own
runs and attempts, as upstream's readOwnHistoryItems does and as the fork's timeline shows it.

Upstream: c3e5bfc027176c17c7d16e0d758eed120656e7b9

* fix(provider-cursor): settle a Cursor Cloud retry row once the retry is accepted

Cursor Cloud marked its automatic retry row running and settled it only
when the whole retried run ended. Polling a cloud run emits no thread
events, so the provider session manager's retry stall watch read a
retried run that worked longer than 15 minutes as a stalled retry,
stopped the turn and cancelled the remote run. The row now settles
("Provider recovered") right after the commit that marks the retried run
running; a retried run that fails later still reports its failure through
the terminal failure item, which shares the row's id.

Fail-first: adapter.retry.test.ts "settles the retry row once the retried
run is accepted, not when it ends" (retried run polled for 16 minutes):
expected [ 'running' ] to deeply equal [ 'running', 'completed' ].

* docs(internals): regenerate the agent surface audit after the task.create description change

c7e10789ac shortened the task.create description from 400 to 399
characters; vp run agent-surface:audit rewrites that row. Generated.

* test(server): require the repeated delegated offers to name the delivered message

The delegate_task question-wait test accepted repeated offers by checking
that every offer had the same message id, which also passed when every id
was missing. It now requires at least one offer and that the offers'
single message id is the delivery's message id the test reads below.

* fix(server): text generation takes the installation gate only for registered instances

The installation gate keeps one entry per provider instance id for the
server's lifetime. Text generation took the gate before resolving the
instance, so every call with an unknown instance id added a map entry.
It now resolves the instance first and runs an unregistered id (which
then fails, or is an API-configured generator) without the gate, so the
gate map is bounded by the configured instances.

Fail-first: TextGeneration.installationGate.test.ts "takes the
installation gate only for a registered provider instance":
expected [ 'never-registered' ] to deeply equal [].

* chore(licenses): list the Unicode licence for CLDR data in the mobile app

The FormatJS Intl polyfills the mobile app bundles carry tables generated
from Unicode CLDR, but their packages ship only FormatJS's MIT licence.
The mobile notice list now has a Unicode-3.0 entry naming those packages
and data; the SPDX template comes from licenses:sync and carries
Unicode's copyright line. Web and desktop lists are unchanged.

Model: anthropic/claude-opus-5-5 via Codex in GOAT

* test(mobile): translation catalogs match the strings the app renders

gt generate cannot rebuild the mobile catalogs (it empties them), so a
contract test checks that every gt() literal has an en.json entry under
GT's hash with the same text, and that each shipped locale has exactly
en.json's keys. An emptied catalog or upstream's "T3 Account" key fails it.
generaltranslation 9.5.9, already in the lockfile, becomes a mobile
devDependency for the hash.

Model: anthropic/claude-opus-5-5 via Codex in GOAT

* test(server): thread and task reads fetch the described run's Stop items

The stop field reads run_interrupt_request and run_interrupt_result items, so the activity and termination-cause mocks now answer getThreadRecords with an empty turn-item list. The stopped child read test asserted pending while the effect worker may already have claimed the delivery; it now asserts what its title says, that the delivery is still owed (pending or claimed), not acknowledged or disposed.

* test(server): move the 080-089 ledger fixtures into a testkit

The id-shift resume tests that follow seed the same old GOAT ledgers.
UPSTREAM_80S, isDatabaseNewerThanBuildError, goatNames, currentRow,
history, schema, seedOldGoatLedger, seedUserData and userData move from
reconcileGoatMigrationIds.upstream080.test.ts to
reconcileGoatMigrationIds.testkit.ts. They are exported and otherwise
unchanged, except that currentRow and seedOldGoatLedger narrow instead
of asserting non-null, since testkits count as source for that rule.
Titles and assertion calls in the test file: 13 and 59, before and after.

* test(server): run the Claude copy and cache upgrade cases on TestClock

They only touch fixture files, so it.live was never needed (upstream 43ce39595b).

* fix(server): a scheduled run the provider rejects shows why on its task

The terminal transition stored the bare status, so a run stopped by a usage limit showed only Failed. It now stores the root provider failure, or how the run ended, capped at 500 characters (upstream 5c0eb5124e).

* feat(mcp): scheduled task results carry the last run's reason

Optional lastRunError on goat_schedule_list/create/update and run_scheduled_task_now; older hosts omit it.

* fix(opencode): a hung model read of a lent OpenCode 2 server ends at 20 seconds

Since upstream a877d40b74 the model loader's first read had no cap of its own
and only the 2-minute plugin-activation deadline ended a read that never
answers, so a check (and the old profile's status it leaves shown, goat#3415)
could wait 2 minutes. The loader now hands the list a bound for reads of a
server already lent: 20 seconds, or GOAT_OPENCODE_PROBE_TIMEOUT_MS from the
check's environment when that is larger. A capped read lists nothing yet
(warning "did not list any models yet"); only the plugin-loading 503 retries
for up to 2 minutes. The server's own start stays outside the cap. 20 s sits
above the ~14 s OpenCode 2.2.7 holds a read on a busy cold host (upstream
#30251), whose regression still loads every model.

status.openCode2Profile.test.ts: the new profile's never-answering read now
ends at the 20 s cap, so the test advances 20 seconds.

Fail first (implementation stashed): the driver test was still waiting at
20 s (vitest 5 s timeout); the loader tests failed with "boundRead is not a
function".

* fix(contracts): task.create says which next to follow

The listing shows only the description and input signatures, so "Follow
next." had nothing to refer to before an agent sees a result. It now
reads "Follow the result's next.", matching the task guide. Twelve
characters come out elsewhere in the same description to stay within
its 400-character budget: "a separate GOAT-managed checkout" becomes
"a GOAT-managed checkout", "create worktrees" becomes "make worktrees",
and "uncommitted edits" becomes "uncommitted work". The agent surface
audit doc is regenerated (delegate_task 399 -> 400).

* fix(server): Tailscale Serve no longer takes a live server's route and repairs one left on a dead port

Before applying, read who holds the Serve port's root; another live server or an unreadable status leaves it alone and is never turned off. Every 30 seconds a route found on a dead loopback port is taken back. Ports upstream becc7d3510 with the serve-status parsing subset it needs from d897d3f3f0.

* fix(server): startup migrations commit one at a time and log progress

The Effect migrator runs every pending migration in one transaction, so
a server stopped during a long upgrade, such as an index build over
every turn item, rolled all of them back and started over on the next
launch. runMigrations now calls the migrator once per pending
migration, each in its own transaction, and logs "Upgrading database"
with the migration and its place in the run before each one on a
database that already has a ledger. A restart then resumes after the
last migration that finished.

Upstream's off-main-thread runner (NodeSqliteClient.layerWorker) is not
taken: GOAT's file databases already run opening and migration SQL on
the writer worker (Sqlite.test.ts "runs file database opening and
migration SQL away from the server thread"). Upstream's test makes 085
fail by creating its counts table first; GOAT's 085 checks for that
table, because the id shift may rerun it, so the adapted test makes
112_FleetHealth fail instead and expects 1-111 to stay recorded.

Upstream: 4ca5b9cdb8f5f8a0e463faba0d5c4b4f6bb322c0
Decision: adapt

* fix(server): GOAT's migration id shift resumes after a restart

A database from before cycle 261114 records GOAT's migrations at 80-147.
The first start of a build with upstream 080-089 moved those rows to
90-157 and ran 080-089 in one transaction, so a server stopped during
the long part (081's search-text backfill, 082-089's index builds, 085's
counts) rolled everything back and started over, and an SSH host whose
60 s ready limit stops it every time never finished (goat#3357).

reconcileGoatMigrationIds now commits in steps. The first transaction
checks the memory ledger, refuses before any write when a GOAT row would
land on an id another migration recorded, moves GOAT's rows to their
current ids, and runs the first displaced upstream migration (080, which
alters only scheduled_tasks). After that each displaced migration runs in
its own transaction: it reads the ledger again, runs the body and
records its row, so a failure inside one rolls back only that one, and
a restart runs whatever is next. displacedUpstreamMigrations names them
from the ledger alone (upstream entries below the highest recorded GOAT
row and not recorded under their id and name), so the resume does not
depend on the migrator's rule of running only ids above the highest
recorded. 080 commits with the move because a moved ledger holding only
GOAT names is one a build from before the shift would move back and
start on; with 080 recorded that build hits its collision check and
refuses as a newer database (DatabaseNewerThanBuildError, exit 79),
unchanged, in every state a restart can find. hasPendingMigrations
counts displaced migrations too, so a CLI command takes the server's
ownership lock in those states and refuses while a server owns the
home.

Two existing tests pinned the old all-or-nothing rollback and now pin
the new contract: "an upgrade interrupted in the last upstream step
keeps the steps before it" (was "... rolls back whole and resumes") and
"an interrupted relocation keeps the steps before the failure and
finishes" (was "retries an interrupted relocation without changing the
old schema or ledger"). reconcileGoatMigrationIds.ts no longer has a
non-null assertion, so its nonNull budget entry goes.

Upstream: 4ca5b9cdb8f5f8a0e463faba0d5c4b4f6bb322c0
Decision: adapt (goat#3357, coordinator option B)

* test(server): the usage-home fake Codex is a signed-in ChatGPT login

c9dfc404cc reads usage only for an OpenAI login, so the fake Codex answers account/read with a ChatGPT account and, like the real app-server, refreshes its model catalog at startup, which the probe waits for under a ChatGPT login. Assertions unchanged.

* test(server): the interrupted id-shift tests pin the whole stepped contract

The two tests rewritten for per-migration transactions held only that
the steps before a failure stay recorded. Each now holds all of it: a
failure in the first displaced migration (080, or 070 for the older
70-series ledger) leaves the old ledger and schema unchanged, since the
collision check, the ledger move and that migration commit together; a
later migration (089, or 076) that fails after doing its work rolls that
work back alone, every migration before it stays with its ledger row,
and a restart resumes there and ends equal to a fresh database. The
schema snapshot, the fresh-database snapshot and a helper that fails a
migration after its real body move to the testkit.

* test(server): the native-child drain case keeps a child a person's run holds busy

e66dce5c1e replaced the drain case's busy child with a runless working
child and so dropped upstream 20081ef95f's busy-child retry. The case now
has both: the working child (an active subagent row on its parent, no
app run) settles with its parent, and a child a person messaged, whose
app run is still running, is held until that run completes and is then
settled by the targeted read the run update triggers.

* test(server): the CLI lock tests take ownership the way a server does

The layerCliConfig cases held the home with holdServerOwnershipLock.
They now take it through acquireServerOwnership, the path the server
runs at start, with an empty home so no boot service on this machine is
read, as upstream 5da53f87d9's tests do.

* refactor(server): the steer follow-up queue position lives in steerFollowUpQueue.ts

Verbatim move of the declined steer follow-up queue position and the reorder of later queued runs; frees 34 lines of Orchestrator.ts (13903 to 13869, budget 13908) for L01 round 2.

* test(server): ProviderSessionManager.test.ts imports ThreadCommandExecutor for the L04 case the 1f34e35e42 merge brought

* feat(server): a parent can force stop its own task child once Stop went unanswered (goat#2338)

ThreadManagementService.forceStopDelegatedTask decides: the caller created the task, the task is local, and the child run's Stop is flagged unresponsive. It then runs the person's Force stop path on that run (superseded and cloud refusals unchanged) with a reason that names the parent thread. Other parents, grandchildren, launched threads, remote tasks and a first call that never timed out are refused and stop nothing.

* feat(mcp): task.cancel takes force: true for an unresponsive child (goat#2338)

The optional force reaches ThreadManagementService.forceStopDelegatedTask after the usual mode check on the child; refusals answer task_not_cancellable with not_owner, not_unresponsive, remote, superseded or cloud. The result carries the Force stop outcome and a note that the work may be incomplete. Descriptions and the stop field now name the path; the audit doc is regenerated.

* fix(server): only unrecorded upstream ids count as displaced by the id shift

f6f553c473 counted an upstream migration as displaced when the ledger did
not record it under its own id and name, below a recorded GOAT migration.
A database that recorded another migration at such an id (a site-local
migration at 51, or a v1 ledger's ThreadSummaryTimeline) then counted as
pending and was refused on every start, where the migrator used to skip
the id: the gate of 978975d31d found it in scripts/migrate-dev-db.test.ts
"seeds from a source that recorded another migration below V2" and
LegacyV1Cutover.integration.test.ts, both DatabaseNewerThanBuildError.

Only ids the ledger does not record at all are displaced now; an id
recorded under another name belongs to another history, as before the
stepped shift. The refusal of such an id during a move ("Migration N is
recorded as X, expected Y.") moves back into the move step, before any
write, with the range the one-transaction shift used (ids up to the
highest moved GOAT row).

* chore: ProviderSessionManager test budget follows the L04 merge

The lane's budget follow-up (e075571c48) set this entry to the file's length
then, 6096. chase/t3-261114-all at 3f9cb15cba (L04, 1f34e35e42) grew the file
to 6098 while keeping its own entry at 6326, so the merged lane head failed
goat(code-budget). The entry now records the merged length, still below the
integration branch's 6326.

* fix(server): Tailscale Serve waits for a status it cannot read yet and turns off only its own route

A status read that fails at start (tailscaled or the app not up yet) is retried on the apply schedule instead of leaving Serve off until restart; only a status that cannot be parsed leaves it off. At stop the route is turned off only while it still points at this server.

* fix(server): a cut run reason keeps whole emoji, and its test reaches the kept-reason path

* test(contracts): lastRunError rejects a wrong type, and the run-now guide names it

* fix(server): goat pair --tailscale no longer replaces another app's Serve route

Upstream 77f40aff37: the Serve config, not the MagicDNS probe, decides whether the port is free; the holder read moves into packages/tailscale and the startup layer uses it.

* fix(server): the id shift starts earlier ledgers with other names below V2 and keeps lost rows fail-closed

The first start on a database from an earlier GOAT build (GOAT's rows at
80-147) refused it with DatabaseNewerThanBuildError, exit 79, when its
ledger recorded another name below id 55: a v1 cutover keeps
41:ThreadSummaryTimeline, and site-local migrations share ids. The
move's "recorded as X, expected Y" check covered every upstream id up to
the highest moved GOAT row. It now skips ids below 55, as the migrator
and origin/dev do; ids 55 and up stay refused before any write. The
warning about the divergent row names only that id.

Displaced upstream migrations are now only the unrecorded tail above the
highest upstream id the ledger records below GOAT's block (coordinator
ruling 2026-11-15). A gap below that tail is a lost ledger row and keeps
the migrator's rules: at ids 55 and up it is refused as a skipped
migration before any write (inside the move too, so a ledger from
before the shift without 72 refuses before moving anything), below 55
it is ignored. So a shifted ledger that lost 61 refuses as before, one
that lost 7 starts as before, and the first transaction of a shift is
always the move plus the block's first id (080), also when the ledger
lost a row below 55. The skipped-migration check moves into
reconcileGoatMigrationIds.ts (skippedV2Migrations) so the move and
assertV2MigrationsCanRun share it.

The resume tests remove each file database's temporary directory once
its connection closes (they left 14 per run).

* test(server): the native-child drain case checks for PR lookups after the busy child's retry

As upstream 20081ef95f does, the case holds that no branch or summary
lookup ran after the final drain, so a lookup during the retried child's
targeted settle fails it.

* fix(contracts): the task.cancel force description is one text, and stopIsUnresponsive stays private

* chore: restore code budget entries of files this lane does not change

c70cbd140a, e075571c48 and 0377149d3f lowered the entries of
ProviderSessionManager.test.ts (6326 to 6098), ChatView.tsx (4239 to 4236) and
ModelPickerContent.tsx (1078 to 1072), files this lane does not change; another
lane's growth of ProviderSessionManager.test.ts would then fail
goat(code-budget) on merge. They are back at chase/t3-261114-all's values. The
removal of status.test.ts's entry stays: this lane changed that file (814
lines).

* feat(server): the host force stops a bound task thread for its origin (goat#2338)

originBoundForceStop checks the report binding's read key and thread, the caller's mode ceiling, and that the task's run is still live and ignored its Stop, then runs Force stop with a reason that names the origin. forceStopUnresponsiveRun is the shared step behind it and the parent's local force stop. New tests: task-level restart, cloud and shared-session cases, MCP-level task.cancel force, and the host refusals.

* feat(mcp): task.cancel force reaches a remote task child through its host (goat#2338)

A remote task's force now goes to the host's new goat_origin_force_stop, authorized by the task's report binding and read key. The origin first reads the host's orchestrator_capabilities and sends nothing unless features.taskForceStop is true, so an older host refuses with "remote:" and stops nothing; a cancel without force (all an older caller sends) never asks for it. The host refuses unless the task's run is still live and ignored its Stop, and keeps the session's mode ceiling.

The binding check moves to originBinding.ts so bound delete and force stop share it. The remote MCP fixture moves to OrchestratorMcpService.remote.testkit.ts; the audit doc is regenerated.

* docs(contracts): task.cancel force says where a remote task's stop state shows, and the user guide covers SSH hosts (goat#2338)

* test(server): port upstream Antigravity installer tests and reach the extraction cap and host allow-list

* fix(server): Antigravity never offers the pin as an update to a newer native CLI, and offers only supported channel releases

Review A P2-2 and P3-8. An installed release newer than the candidate and not withdrawn stays the offer at startup and in every refreshRelease path (nothing published, withdrawn or unsupported release), so a native 1.5.x install no longer shows "1.4.5 is available". A release from Google's channel is offered only once the model manifest marks it supported, as Codex requires; an installed release the manifest does not list keeps running. The ported withdrawn test lists 1.5.6 as supported in its override manifest, since an unlisted release is no longer offered.

* fix(server): Codex's managed-only start reads active.json, OpenCodex child setup never installs a newer release, and extraction is capped

Review A P3-7. startManagedOnly takes { allowUpdate } and reads the installed version from active.json, since the shared state names a Codex on PATH as source "local" and that made Install on an authless home point active.json back to the pin. OpenCodex child setup passes allowUpdate false: a failed managed resolve re-checks the installed release and never installs a cached newer one nobody pressed for. Install on an authless-home instance passes true.

Rulings A5. Extraction moves to CodexInstallationExtract.ts: the same entry checks, then each member is streamed out of tar and counted against a 1 GiB ceiling, so a package that inflates past it stops while it is written. CodexInstallation.ts stays under its 1000-line cap.

* fix(server): a verified OpenCodex route reads usage only for a non-API-key login

Review B P2-4. The OpenCodex usage exception now also requires that the login is not an API key, so an API-key login on that route reports usage as unsupported again, as on the tip. Two probe tests pin both sides: a ChatGPT login reads account/rateLimits, an API-key login never asks.

* fix(server): a secret Check applies the value it read, as upstream

Review B P3-1. A successful Check now hands the value from its one read to the reload: the providers that use that reference are rebuilt from it without a second read (so no second 1Password prompt), a store that fails after the Check no longer strips the secret it just reported "Read just now", and only that reference's cache entry is replaced instead of clearing every cached secret. reloadSecrets takes the checked value; without one it still re-reads everything as before.

* fix(web): the secret Check says the server could not be reached when its request fails

Review B P3-2. A Check whose RPC does not succeed showed nothing; it now shows "Could not reach the server", as upstream ee9e36c8b2 does. Browser test added (it failed with the message missing before the fix).

* test(server): restore the status-fallback test; the stop field gets its own tests

Review B P3-3. "status fallback keeps optional cloud=%s metadata and caller watch" is back to the tip's title, fixture and assertions. The stop field is covered in HostThreadsMcpService.stop.test.ts: from goat_thread_status (newer host and one that predates it, moved unchanged) and through the thread-read fallback (new). The shared harness moves unchanged into HostThreadsMcpService.testkit.ts, so the test file stays within its recorded length; its two non-null assertions become narrowing there.

* docs: Antigravity native goals, the one /goal form it accepts, Intel Macs on the pin, and supported-only offers

Rulings A6, A10 and the user-facing side of A2/A4: providers-antigravity.md gains upstream 4099e62767's "Run a native goal" section; composer.md's Goals names Antigravity and its single accepted form, /goal <objective>; Intel Macs install the 1.4.5 pin when Google's channel is down; install.md says only supported releases are offered and a newer installed release is kept.

* fix(server): the id shift refuses a lost row just below GOAT's old block before moving anything

A ledger from before the shift that lost the row directly below GOAT's
old block (79, or 78 and 79) took that row as part of the displaced
tail: the first transaction was the move plus 079, and once it had
committed, a pre-shift build moved GOAT's rows back and started. A
ledger with 1-54 and GOAT's rows at 80-147 failed every start at
"Migration 55_OrchestrationV2 failed" instead of being refused.

Before the shift every upstream id below GOAT's old block was recorded,
so the move now refuses missing rows up to the lowest GOAT id recorded
before the move, minus one, as a skipped migration and before any write,
as origin/dev does (55-79 today, 55-69 for the older 70-series layout).
The first transaction is then always the move plus 080. Its message
names the ledger's highest recorded id (147), where it said 79.

The lost-30 test also compares its end state with a fresh database
minus row 30.

* refactor(provider-cursor): move Cursor Cloud progress text helpers to their own module

Moves the Progress type, the tool progress labels, progressLine and
progressText (80 lines) from cloud/adapter.ts to cloud/progress.ts so the
retry timeout fits the adapter's 1000-line limit. The moved lines are
byte-identical (script check: zero line mismatches); the adapter
re-exports progressText for its tests.

* fix(provider-cursor): a Cursor Cloud retry that Cursor never takes ends the turn

The SDK's resume and send have no deadline. When Cursor never answered
the automatic retry, the task stayed "submitting", Stop refused it as not
attached, and the retry stall watch fired once and gave up, so the turn
never ended. retryOnce now bounds resume and send to two minutes; a hang
fails into the observer's existing failure path, which marks the task
unknown ("Cloud tracking is unavailable") and ends the turn failed.

Fail-first: adapter.retry.test.ts "ends the turn when Cursor never
answers the retry": expected [] to deeply equal [ ObjectContaining{…} ].

* fix(server): say when Tailscale Serve is left on because its routes could not be read at stop

* test(server): cover a settled provider retry and drop assertions the harness cannot back

- ProviderRetryStall.testkit.ts: a running retry row followed by its
  completed row, then 16 minutes on the TestClock, leaves the turn alone
  (interruptCount 0). With the watch arming on any non-interrupted row the
  case fails (expected 1 to equal +0).
- adapter.retry.test.ts: the Cursor Cloud harness has no stall watch, so
  its cancelRun assertion proved nothing; it is dropped.
- ClaudeBackgroundDetails.test.ts: the comment no longer names a commit.

* fix(server): worktree tools judge GOAT ownership by the same path rule as the delete prompt

goat_worktree_remove and goat_worktree_cleanup counted a worktree at any depth under
GOAT's worktrees folder as GOAT's, while clients, turn start and vcs.removeWorktree
accept only <folder>/<repository>/<name>, where GOAT creates them. Both now use one
helper. Adds a lifecycle test that getWorktreeThreadIds reports managed for a path
inside and outside the folder, and narrows the importer comment and welcome guide
to worktrees outside GOAT's folder.

* chore(server): keep run 7 testkit and extraction constants private (knip)

* fix(usage): count Codex request ledgers and replace archived cache entries

An identified Codex request (a token_usage_record with a response_id) is billed from the request ledger, which includes compaction spending that context snapshots can omit; once a ledger is active, snapshots only rebase the counters. A successfully read .jsonl.zst archive replaces its original's saved records and diagnostics when the original is gone. The usage scan cache moves to v14 (usage-scan-cache-v14.json); v13 files and older Codex entries are re-parsed.

GOAT adaptations: GOAT's bounded pending list (1024 unidentified requests, then unreportedOverflow) stays for requests without a response_id; the identified-request path runs first. The archive replacement lives in usageArchiveCache.ts, as UsageService.ts is at its 1000-line cap. Upstream's cache-upgrade edits land in usageCacheUpgrade.testkit.ts (versions 11 to 13 join the list; file name v14), and the compressed-rollout test with upstream's additions moves unchanged into usageCompressedRollout.testkit.ts, so UsageService.test.ts stays within its recorded length. Upstream's VcsStatusBroadcaster.test.ts edit has no GOAT counterpart.

Downgrade: an older GOAT reads only its own file name (usage-scan-cache-v13.json, usageScanCache.ts:21 at 17af62cba1), which the upgrade removed, so it rebuilds from transcripts; given a v14 document it rejects it as newer than it knows (usageScanCache.ts:196 at 17af62cba1).

Upstream: b42841808d47194de2713d8322e7222809c88166
Decision: adapt

* chore(lint): ScheduledTaskService.ts budget follows the merge of dev's #3398

The L09 lane lowered the entry to the file's length (2311). dev's #3398 adds two lines to the same file and git merged both without a conflict, so the merged file has 2313 lines. The entry is set to the merged length; dev's own entry is 2329.

* fix(desktop): wait for the server's database upgrade instead of reporting a failed start

Port of upstream 2e741fc089 (pingdotgg/t3code#35056) onto GOAT's startup watcher.

The server logs "Running database migrations" before pending migrations and
"Migrations ran successfully" after them. GOAT's stall watcher (60 s of silence,
10 min ceiling) restarts both budgets at a deadline while the backend's output
tail shows a running upgrade, so a long index build no longer opens startup
recovery. Once the finished line arrives, the usual budgets apply again.

Model: anthropic/claude-opus-5-5 via Codex harness in GOAT (lane L18).

* refactor(server): the in-memory projection store's record reads move to a GOAT module

ProjectionStore.ts had 37 lines left under its budget, and the window
and held-input ports that follow add about 200. The code that moves is
GOAT's own: layerMemory's record selection (getThreadRecords' filter
body, now selectThreadRecords), its settled-leaf helpers, and the replay
twin of the server-activity blockers query. It moves unchanged apart
from indentation and exports to goat/ProjectionStoreMemoryReads.ts,
which imports only types from ProjectionStore.ts. ProjectionStore.ts
goes from 8799 to 8582 lines; its budget entry is unchanged.

* perf(server): very long threads open without walking their whole newest turn

Port of upstream b015ffd7c2's code half; 082's index came with #3350.
The window's tail search walked every row of the newest turn to find the
rows sure to keep their own: on a long running turn, thousands of tool
calls for the newest few rows. It now reads conversation rows from the
partial conversation index and the running run's newest rows by its run
index, and when the newest turn holds more rows than a turn page, pages
from inside it with the row limit. A todo list stays active after its
run ends, so a long thread carried every old run's todo node and plan
into each window; those of settled runs no longer count as live work.
Messages of the window and of live runs are looked up by key and by run
instead of filtering the thread's messages, and the checkpoint of the
last finished turn stays in reach while a long run works.

GOAT's window code matched upstream's before this change apart from its
deferWork binding and the anchors fallback, which stay. The conversation
index's type list already existed (from a909ea423c's port). The test of
command reads now takes whether the store keeps only live work beside a
window: the SQL store does, the replay store returns every record.

Upstream: b015ffd7c2d360aed7debb52a3b64483ab4466ab
Decision: adapt

* test(server): two queue tests move to sibling files

SteeringCompletion.integration.test.ts and ProviderSwitch.integration.test.ts
sit at their line budgets, and the held-input port that follows adds
assertions inside their existing tests. "queues a plain send behind the
stopped queue and releases its hold" moves unchanged to
SteeringCompletion.stoppedQueue.integration.test.ts, and "starts a
scheduled fresh run while the previous run's session still has
background work" moves unchanged (one level less indented) into
ProviderSwitchBackground.integration.test.ts, whose own waitForIdle is
the same as the one the source imports. Titles and assertion calls are
unchanged; both pass in their new files.

* fix(chat): requeued steers no longer look delivered

Port of upstream 9fb4492329's code half; its index 086 came with #3350.
A steer that falls back to the queue keeps its queued-turn user item, and
that item read as delivered while its run waited. Queued input of a queued
run is now hidden like that of a cancelled run, on the server (window,
hidden-row read, shell item counts, full read) and in the client
projection and the shared visibility rule, until the run starts. Forks
keep rolled-back source history but still never inherit undelivered
input, also when a cancelled run is rolled back later. When a queued run
fails before it starts (an unsupported provider handoff), its retained
user item is republished as failed before the failure, so a client whose
bounded snapshot left it out shows the prompt beside the error.

GOAT: the republish helper lives in goat/RetainedQueuedInput.ts to keep
Orchestrator.ts under its budget; upstream's fork test sits in the new
ProjectionStore.heldInput.test.ts, after verbatim moves in 3d66f779423.
The client-runtime, shared, SteeringCompletion and ProviderSwitch edits
are upstream's text. The window-equivalence test gains upstream's held
queued turn kind, so windowed pages are still checked against the full
index with held input in the thread.

Upstream: 9fb4492329e5847efeabf2b4d4f5c40a58dc883e
Decision: adapt

* perf(server): shell item counts read each run's status from an index

Port of upstream 2d8a843e6b's run/status hunk, deferred until 9fb4492329's
held-input counts were in. The shell's per-run and per-thread item counts
looked each run up by its primary key, which loads the whole run row and
payload just to read its status; both lookups now name the run/status
index from migration 089 (in since #3350), which covers them. The count
plan test gains upstream's assertion that every run lookup of a count
query is a covering index search.

The rest of 2d8a843e6b (shell worktree and family filters, the archived
column, the presented run's failure read) was ported earlier in
13c7c8b621 and lives in ProjectionShellThreadRows.ts.

Upstream: 2d8a843e6be9ca170f3bce98cc92ad76617b00c4
Decision: adapt

* test(server): each permission profile allows one more write, goat_origin_force_stop (goat#2338)

* refactor(server): the memory store's settled-leaf check stays private to its module

e088e3743a exported isSettledLeafNode with the code it moved, but only
the module itself calls it; knip flags the unused export.

* fix(server): a held queued run no longer shortens a long thread's opening window

The window's tail search counts the rows of the oldest live run, read by
its run index past the window's hiding. Since 9fb4492329 a queued run has
no visible rows: its only row is held input. After a stop, a held queued
run was still that live run, so its hidden row counted toward the newest
turn: a turn of exactly a turn page (500 conversation rows) read as longer
and paged from inside itself (75 rows against the whole turn), and a
longer turn's window held one row fewer than the full read (76 against
77). Queued runs no longer count as the live run.

GOAT difference from upstream: upstream (b015ffd7c2 with 9fb4492329, and
upstream main) keeps 'queued' in that list; found by review A of the L03
chain (L03C-A-P3-1).

Test: ProjectionStore.liveWindow.test.ts "a held queued run after a stop
does not count toward the newest turn's length" (bounded window equals the
bounded full read at 500 rows; the window equals the full read's newest
rows at 600) fails on 49e609887c at both sizes and passes with this.

* test(server): wire budget covers stale todo lists in long threads

Upstream 06bd84623b (#34522), applied unchanged. Without b015ffd7c2's code half (673b3a5af2) the stale todo cold open fails at 739 KB with 82 nodes, 82 subagents, 41 plans, 41 provider turns and 40 checkpoints; with it, 74 KB and 2/2/1/1/1.

* test(server): a declined Cursor steer waiting in the queue does not look delivered

The declined-steer integration test (2248ca8db9's port) also checks that the queued follow-up is absent from visibleTurnItems while it waits, the state upstream 9fb4492329 (in L03's chain) guarantees for requeued steers. On this path it held before the chain as well; the assertion pins it.

* feat(web): screen reader window title leads with the thread's status and name

Port of upstream 1d218d1525 (#34697). The status helper lives beside resolveThreadListStatus in client-runtime because Sidebar.logic.ts is at its code budget; the switch joins the existing Activity section in Settings > General.

* test(server): wire budget measures what clients fetch today

Upstream 476fc9be66 (#35225), adapted. The client opt-ins are typed Required against GOAT's contracts, so GOAT's acceptTextAppends is sent too. GOAT_TRANSFER_BUDGET_SCENARIOS_PATH joins the kept env.

Not ported: the stale shell resume row (snapshot-required with httpSnapshotFallback) measures 9557ab46c5, deferred in cycle 261109 and absent from GOAT; .github/scripts/thread-transfer-report.cjs, its test and ci.yml do not exist in GOAT (goat-ci.yml runs no transfer report), so the PR comment part is left out and the doc says how to write the report locally.

* test(mobile): the Android workspace Back test mocks the Bots home mode hook

dev's #3459 (a21e558635) made WorkspaceRootScreen.android.tsx import
useMobileGoatBotHomeMode. This test came with the catch-up (664ae37c3e) and
mocks every module that screen imports, so the new import loaded the real
hook and Expo's runtime, and the file failed to load with "__DEV__ is not
defined" (goat-test-changed on e943ef0c5f; 757 other files passed). The mock
returns the Threads mode; the two Back expectations are unchanged.

Model: anthropic/claude-fable-5-2 via Codex harness in GOAT (coordinator).

* test(desktop,server): an upgrade ends on readiness, and an announced upgrade that runs nothing still closes

Review L18R-P3-3 of the 2e741fc089 port. The desktop's upgrade test also covers readiness arriving
during the upgrade (upstream's test has it): the backend becomes ready and no failure is reported.
The server test pins Migrations.ts:478: a run that announces pending work and executes nothing still
prints the finished line after the started line.

Model: anthropic/claude-opus-5-5 via Codex harness in GOAT (lane L18).

* fix(server): goat_thread_read results fit MCP clients' output limit

Port of upstream 6237824a2c (#33455). The read budget counts UTF-8 bytes of the whole result instead of characters, and a single item too large for the budget has its text, then its title, cut to fit. The helpers live in threadReadItems.ts beside timelineItem. Left out: upstream readPastEnd guard (GOAT acknowledges a child result only on a textOffset 0 read already) and the toolkit integration paginated case (absent in GOAT).

* fix(server): report failed preview navigation URLs

Port of upstream c58f472a35 (#34580). Status, snapshot and the tab list name the page that failed instead of chrome-error://, and a navigate or open that ends on Chromium's error page fails with the page (credentials, query and fragment cut) and the reason. GOAT adaptation: a new tab whose first page failed stays open and is named (openedTabId/unloadedTab), matching unloadedOpen.ts, where upstream closes it. The checks live in serverBrowserLoadError.ts and the scenario bodies in LoadErrorAssertions.testkit.ts to keep ServerBrowser.ts and its test at their code budgets.

* fix(mobile): a reply a steer cut off shows an interrupted badge

Follow-up to upstream c7618f6c10 (GOAT 28c4865c8a, server half and web badge). assistantReplyStatusBadge in client-runtime thread-workflows holds the rule both clients use: an interrupted reply shows its status wherever it sits in the turn, and any other unfinished reply only in the meta row that closes the turn. Web behaviour is unchanged; mobile ThreadFeed now shows the same badge.

* refactor(server): ServerBrowser reads load errors through one namespace import

Keeps ServerBrowser.ts within its code budget for the next fix.

* fix(server): closing an unclaimed tab is reported as never sent

Remainder of upstream c60aa1aaf9. The refusal to close a tab with no agent owner now carries beforeInput, as other pre-input refusals do, so the broker marks it not dispatched and preview_close tells the agent the close was rejected before it reached the tab instead of that it may have applied.

* fix(server): an Antigravity install recorded by the 1.1.1 pin is offered updates again

showOffer compared versions with compareSemverVersions, which falls back to
localeCompare for "agy_acp_server_1.1.1" and ranked it above every release, so
Settings showed Reinstall and never offered 1.4.5 or a channel release. It now
uses managedRelease.ts's isNewer, which treats an unparsable installed version
as older than any release. Codex already offers through selectManagedRelease.

Review: L02R7-A-P2-2

* fix(server): Codex extraction refuses option and pattern member names and has a deadline

Each archive member name is passed to tar, so a name starting with "-" could be
read as an option (GNU tar's --to-command runs a command) and *, ? or [ as a
pattern that writes other members into that file. Such names are now refused
with the other unsafe entries before anything is unpacked, and the member is
passed after "--". Every member re-reads the whole archive, so the extraction
as a whole now ends after 10 minutes; a real package unpacks in seconds.

Review: L02R7-A-P3-3

* test(server): an OpenCodex Child's Codex setup never installs an update

The child-ready mock now records the allowUpdate each install asks for, and a
new test pins that OpenCodex child setup passes false, so a cached newer
release is never installed there without someone pressing Update.

Review: L02R7-A-P3-4

* fix(server): musl Linux is never offered Antigravity's glibc-only pinned runtime

The 1.4.5 pin was chosen by platform and CPU only, so with Google's channel
down a musl host downloaded a glibc runtime. The pin now shares the CLI
lookup's musl detection and is null there; setup reports that no release is
available, and the Antigravity guide says so.

Review: L02R7-A-P3-5

* test(server): a parent cannot force stop a promoted or provider-native task (goat#2338)

The ownership checks that refuse a child the person promoted to a top-level
thread and a p…
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

size:M 30-99 changed lines (additions + deletions). vouch:trusted Author is trusted by repo permissions or the VOUCHED list.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug]: T3 Code Desktop Windows cant open anymore 0.1.6

1 participant