[Fryuni]

Problem

The fork is 719 commits behind upstream/main. This sync keeps the features recorded in docs/adr and uses upstream's implementation wherever it can.

Change

Merged pingdotgg/t3code main at 13f89ae0a. 27 files had content conflicts. git show --remerge-diff 3963297f82 shows the whole resolution.

Workflows (ADR 0011). The fork keeps its own ci.yml layout and required check names. Upstream's other workflows, which it modified in this range, are dropped again, along with its new issue-vouch.yml. ci.yml and server-linux-release.yml now pin their actions to the same commit SHAs as upstream (pingdotgg#33696).

Implemented upstream

• ADR 0028 (removed .git file). Upstream fix(server): turns recreate a worktree folder that lost its checkout pingdotgg/t3code#34894 keeps a worktree whose .git file is gone as "not a linked worktree", which the fork already did. Cleanup uses upstream's check, and the fork's duplicate test is dropped.

No upstream commit in this range supersedes a whole ADR.

Re-homed on upstream's structure

• ADRs 0002 and 0020 (PR targets). Upstream moved PR target resolution from the MCP handlers into pullRequest/pullRequestTarget.ts, so its new watch service can share it (fix(server): reject PR watches when hosting dependencies are unavailable pingdotgg/t3code#33389). The fork's mappings moved with it:

  • the clone-host mapping;
  • the Forgejo repository fold;
  • the webUrl origin.

  The watch service passes the thread's project, as link_pull_requests does.

• ADR 0008 (Check out in worktree)

  • Upstream's generatedTemporaryBranch (fix(server): honor worktree prefixes when automatic naming fails pingdotgg/t3code#35718) excludes existing-branch checkouts, so their branch is never flattened or renamed.
  • The rule that a failed launch deletes only a branch it created now sits beside upstream's new submodule stage.

• ADR 0009 (default base branch). The project's base branch still comes first, ahead of upstream's ref-based resolveDefaultWorktreeBaseBranch (fix(web): a new worktree thread opened during a clone bases on the real default branch pingdotgg/t3code#36379). Its write goes through upstream's re-planning settings writer.

• ADR 0023 (setup commands). Upstream now drops a failed send's optimistic row earlier (fix(web): a failed send leaves the timeline when a new draft was typed pingdotgg/t3code#35102), and the leading commands' rows are dropped with it. The mobile outbox keeps both the split and upstream's launchedCreations reuse (fix(mobile): a new thread's first message no longer returns to the composer after a lost launch reply pingdotgg/t3code#32330).

• ADR 0025 (OhMyPi presets). Pi status keeps modelInventory alongside upstream's marking of text-generation models (fix(pi): stop offering extension-only models for text generation pingdotgg/t3code#24110). The test harnesses for status and text generation now cover both the omp config reads and upstream's version probe.

• ADR 0028 (Never)

  • Upstream's new cleanup of No project scratch folders (fix(server): storage cleanup prunes idle No project folders under ~/.t3/scratch pingdotgg/t3code#36369) now removes folders that hold files under Never, as Edited tracked files does. This change has a test, and docs/user/project-settings.md says so.
  • Upstream's pre-removal walk (fix(server): keep worktrees containing nested repositories pingdotgg/t3code#25241) keeps a worktree it cannot finish walking. Cleanup now checks that the worktree still exists, so one removed during the walk is skipped, as before.

• Smaller resolutions

  • Merging both sides:
    • the mobile add-project screen keeps projectTitle (ADR 0022);
    • the mobile project overview keeps checkout targets;
    • _chat.settings keeps the provider fan-out rule (ADR 0024);
    • the SettingsPanels imports and tests from both sides are kept.
  • agentContextBudget.json: mcpToolListJsonBytes is 50,811, which is upstream's 49,806 plus the fork's unchanged 1,005.

Fixes for new upstream interactions

• Upstream's new ThreadDetailsPanel and useComposerSelection browser tests now pass the fork's props.
• A mobile triage test mock now includes environment, which the fork's capability read needs.
• In PullRequestSyncReactor.test, the shared-PR test keeps the fork's lowercase repository expectation.
• Upstream's build-desktop-artifact test for the macOS preview (fix(ci): macOS PR previews no longer run the PR's bundle next to signing secrets pingdotgg/t3code#36407) reads desktop-macos-preview-publish.yml, which the fork drops, so the test is dropped too (ADR 0011).

ADRs

All ADRs now compare against 13f89ae0a. Notes were added in 0002, 0008, 0009, 0011, 0020, 0023, 0025 and 0028.

Verification

• vp run typecheck: 41/41 packages pass.
• vp run check:changed origin/main:
  • Format, lint, knip, GraphQL, typecheck, node --test and the React compiler pass.
  • The 819-file test stage hit the local 900 s limit, so CI's full suites cover it.
  • unused patches failed. That check resolves dependencies fresh, without the lockfile, and a newly published unenv trips an infra/relay peer warning. The manifests are identical to upstream's, and no unused patch was reported.
• Targeted tests on every file both sides changed, all passing:
  • Server: 26 files / 1,525 tests, plus storageCleanup (223) and AgentContextBudget.
  • Packages: provider-pi, contracts, shared, client-runtime, source-control-forgejo.
  • Web: 19 unit files / 637 tests, plus 13 browser files / 91 tests in Chromium.
  • Mobile: 8 files / 145 tests.
  • Desktop: 3 files / 34 tests.
  • scripts/dev-runner.
• Three GitManager tests fail locally because this machine's global git config sets push.default=upstream. They pass with GIT_CONFIG_GLOBAL=/dev/null.
• An independent review checked every conflict resolution and all 145 files both sides changed against the ADRs. It found no defects introduced by the merge.

Not run locally: a manual pass in a real client, the fork release workflow end to end, and the full test suites, which CI runs.

Model: Claude Opus 5.5; harness: Claude Code in T3 Code.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-14T10:58:44.291415Z	0b830c7	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3963297f82

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[fryuni-proval]

@Fryuni, thanks for the work on desktop startup and remote-connection reliability.

Overview

The reviewed paths add faster backend readiness signaling, safer update handoffs, preview-browser controls and certificate approvals, and Linux process-spawning safeguards. My targeted review focused on desktop lifecycle, SSH ownership, and the new spawning wrapper; it was not exhaustive across all 1,837 changed files. Tests were inspected but not executed.

Main Issues

Recommendation: request changes for the two failure cases detailed inline.

⚫️ apps/server/src/process/NonBlockingSpawner.ts:27–31 — Two hung filesystem checks exhaust the global slots and indefinitely block unrelated local process spawns.

⚫️ packages/ssh/src/tunnel.ts:673–675 — Reusing a remote server overwrites its launcher owner, enabling another desktop to claim that process after an alias change.

Both would benefit from focused regression tests: an unrelated spawn while two checks remain hung, and two desktop launcher IDs sharing the same SSH state key before a rename.

[fryuni-proval]

@Fryuni, thanks for aligning the packaging tests with the fork’s workflow policy.

Overview

This follow-up removes the assertion that reads the intentionally absent macOS preview workflow and documents that exclusion in the ADR. The runtime regression test that checks an untrusted macOS bundle never executes remains in scripts/build-desktop-artifact.test.ts:1433. Tests were inspected, not executed.

Good points

🟢 docs/adr/0011-fork-release-source.md:36 records the specific workflow-dependent test exclusion and links its upstream change, helping future syncs preserve the fork boundary without dropping shared packaging coverage.

Main Issues

No new actionable issues found in this two-file push. No additional changes requested for this follow-up; the earlier PR-wide recommendation is unchanged.