[juliusmarminge]

Problem

After a branch's PR is squash- or rebase-merged, the Git action button still offers to ship it. With the head branch deleted and a clean worktree identical to origin/main, it shows Push & create PR, and clicking it fails with head invalid. With the head branch kept, it shows Create PR. Squash and rebase merges land new commits, so origin/main..HEAD still counts the branch's original commit, and the remote status reports aheadCount/aheadOfDefaultCount of 1.

Reproduced on main 8b74a69019 in the web app (fresh database) with a local squash-merged fixture: head branch deleted showed an enabled "Push & create PR", head branch kept showed "Create PR".

Change

Remote status already looks up the branch's latest PR in any state, and the forge returns its head SHA, but status dropped it. readRemoteStatus now keeps that head SHA next to the status PR. When the PR is merged into the default branch (resolved from the primary remote when origin/HEAD is missing) and its head SHA equals HEAD, status reports aheadOfDefaultCount as 0, and aheadCount as 0 when the branch has no upstream (where that count is only the fallback delta against the base). A real upstream delta is left alone, so a branch that tracks its base still shows its divergence. HEAD is read only for merged PRs, so open-PR and no-PR status reads cost nothing extra. The last-known-PR fallback keeps the head SHA too, so a failed lookup does not bring the action back.

Web, desktop and mobile already disable push and create PR when these counts are 0, so this is server-only with no contract change. A commit added after the merge, a dirty worktree, a PR merged into another base, or a PR without a head SHA keep today's behaviour.

Scope and approval

Fixes #28467

Verification

vp test run apps/server/src/git/GitManager.test.ts: 113 passed. New tests, all against a real bare remote with a real squash commit:

Test	main 8b74a69019	this branch
status reports nothing to ship after a squash merge deleted the remote branch	fails: aheadCount 1, aheadOfDefaultCount 1	passes
status reports nothing to ship after a squash merge kept the remote branch	fails: aheadOfDefaultCount 1	passes
status keeps a squash-merged branch shipped when a later PR lookup fails	fails: aheadOfDefaultCount 1	passes
status still reports work to ship for a commit added after a squash merge	passes	passes
status still reports work against main after a merge into another base	passes	passes
status keeps real divergence for a squash-merged branch that tracks its base	passes	passes
status reports nothing to ship after a squash merge into a non-origin default	fails: aheadOfDefaultCount 1	passes

vp run check:changed: format, lint, knip, typecheck (3 packages) and tests (2 files, 114) pass.

Not checked: before/after captures of the Git action control with a merged PR are not attached. The status counts that drive it are covered by the tests above, and the client logic that maps zero counts to a disabled action is unchanged. The historical "Changes +N −M" total is unchanged by this PR.

Model and harness: Claude Opus 5.5 (Claude Code) with GPT-6.1 Sol (Codex) reviews, via T3 Code.

🤖 Generated with Claude Code

Takeover

No code changes; evidence and verification added below. The captures that the Verification section above lists as not attached are now in Evidence.
One correction to the Change section: with a dirty worktree the primary action is unchanged (Commit, push & PR), but Create PR becomes disabled at an already-merged HEAD in the web menu and the mobile overview. Web's Push entry also becomes disabled when the remote branch is gone (A3). In A4, and on mobile, Push was already disabled for a dirty tree.

Evidence

Before 71646e0 (merge-base), after 0d27d85 (head). Each side ran from its own checkout with a fresh isolated database (T3CODE_DEV_SEED=0), against the same fixture clones. The fixtures are real squash-merged GitHub PRs in a synthetic private fixture repository, read through the GitHub provider's GraphQL path. Baseline Push and Create PR actions were never run. Each row has its own sheet (linked from the State column) with the full step list, playback timestamps, RPC payloads and every still.

Environment: web Chromium 148 headless, DPR 1, font 100%, no device emulation, reduced motion, JS clock frozen. 1440×1000 dark and 390×844 light. Desktop: Electron 44.10 on Xvfb, 1440×1000 dark, DPR 1. Native: one dev client paired to both servers. iPhone 17 Pro (iOS 27.2, 402×874 pt, DPR 3) and iPad Pro 13 M5 (1032×1376 pt, DPR 2) dark; Pixel 9 (API 36, 411×923 dp, DPR 2.625) light; font scale 1. Native OS clocks could not be frozen, so each before/after pair was captured in the same minute. Every still passed two byte-identical screenshots taken 500 ms apart. Diff0 amplifies any channel delta ×255; DiffT marks deltas >16/255.

State	Before	After	Diff0	DiffT >16	Recordings	What changed
A1 deleted branch, clean · web 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	Push & create PR → disabled Push; menu Push, Create PR disabled
A2 kept branch, clean · web 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	Create PR → disabled Commit; menu Create PR disabled
A1 · web 390 light (details popover)	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	as A1; nothing outside the Git control at zero threshold
A3 deleted, dirty · web 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	primary stays Commit, push & PR; menu Push, Create PR disabled
A4 kept, dirty · web 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	primary unchanged; Create PR disabled
A5 tracks main, diverged · web 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	Sync ref stays; Create PR hint: behind upstream → no local commits
A7 behind feature upstream · web 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	Pull stays; Create PR hint: behind upstream → no local commits
A6 real upstream ahead delta · web 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	Push & create PR and Push stay enabled; Create PR disabled
A9 lookup fails after merged lookup · web 1440 dark, harness¹	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	before stays Push & create PR; after stays disabled Push
A8 zero-turn, zero totals · web 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	Git action and Changes rows hide (existing quiet-draft gate)
A1 · iPhone 17 Pro dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	“1 ahead” → Clean; Push & create PR → disabled Push; overview Push, Create PR disabled
A2 · iPhone 17 Pro dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	overview Create PR disabled with reason; rows shift down
A3 · iPhone 17 Pro dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	subtitle loses “1 ahead”; Create PR disabled
A4 · iPhone 17 Pro dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	Create PR disabled
A5 · iPhone 17 Pro dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	divergence kept; Create PR reason changes
A1 · Pixel 9 light	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	overview loses “1 ahead”; Push, Create PR disabled
A2 · Pixel 9 light	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	Create PR disabled
A3 · Pixel 9 light	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	subtitle loses “1 ahead”; Create PR disabled
A2 · iPad Pro 13 inspector dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	inspector Create PR disabled
A1 · Electron 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	as web A1; nothing else at zero threshold
A2 · Electron 1440 dark	71646e0	0d27d85	71646e0→0d27d85	71646e0→0d27d85	before · after	as web A2; nothing else at zero threshold

¹ A9 is labelled Integration harness: real git, controlled forge response. A real merged lookup runs first, then a fixture-scoped GraphQL failure and a normal refresh in the same server process.

What else changed:

• Web and Electron: nothing outside the Git control and its hints at zero threshold.
• Native stills: faint sheet-material antialiasing outside the changed rows, removed at threshold 16 and listed per row in the sheets.
• Layout: the added disabled-reason descriptions move later overview rows down.
• iOS A5: popover placement differs; noted in its sheet.
• Existing gates, intentionally unchanged: Changes +N/−M totals, divergence (Sync ref, Pull) and a real upstream ahead delta (A6) stay as they were. A8's row hiding is the existing quiet-draft gate.

Guards (RPC/git/forge transcripts, no new display class):

• G1, post-merge follow-up commit: keeps A1/D2. transcript
• G2, merge into release: keeps D1. Same transcript.
• G3, default resolved without origin/HEAD and via upstream/HEAD → develop: gives D1→0. transcript, develop default

Scripts (uploaded, not committed):

• Seeding: seed-fixtures.py, establish-history.py
• Web drivers: browser-driver.py, capture-row.py
• A9 harness: capture-fallback.py, forge-failure-harness.mjs
• Native: capture-native-r2.py, native-driver.py
• Desktop: capture-desktop.mjs
• Everything else: all scripts, recapture scripts, fixture matrix, full evidence index

Verification log

• Code: this round changed no code. Head stays 0d27d85; tests and reviews are as recorded in the Verification section above (113 tests in GitManager.test.ts, 7 new).
• CI on 0d27d85: Build and Typecheck, Changes, Check and the rest are green. Deploy and EAS previews were skipped.
• Proof audit: pass 1 derived the 15-row fixture matrix. Pass 2 took three rounds:
  • Round 1 rejected the native rows (live status-bar clock, clipped rows), stale Electron hint overlays and Diff0 noise on web 390 and A9.
  • Round 2 rejected iOS A2 (unchanged glyphs surviving the threshold), two truncated recording uploads, two caption timestamps and an extra iOS A5 comparison.
  • The final round accepted every row and asked for two body wording fixes, applied above.
  • An independent rebuild of web A2 at the head matched the delivered capture exactly (0 / 1,440,000 px).
• Unverified: the iOS reduced-motion setting could not be overridden on the pool simulators; stable pixel-identical pairs were used as the settle gate. Native OS clocks could not be frozen; the pairs were captured in the same minute instead. Optional matrix guard G4 (PR without head SHA, resolver failure) was not exercised visually; it is covered by the server tests.
• Fixture: synthetic private GitHub repository, not deleted because the token lacks delete_repo scope.
• Model and harness: Claude Opus 5.5 (Claude Code) lead; GPT-6.1 Sol (Codex) capture and proof-audit steps; via T3 Code.

[macroscopeapp]

Approvability

Verdict: Approved at 1c3d7b3

Macroscope's review found this PR approvable — Corrects status so already-merged branches no longer offer redundant push or PR actions. Exact commit matching limits the change, and real upstream divergence remains visible; no repository data is modified.

No code changes detected at 0d27d85. Prior analysis still applies.

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

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	−29 B (−0.6%)	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	−29 B (−2.3%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.6 KiB	21.5 KiB	−41 B (−0.2%)	24.4 KiB	✅
Codex	Live turn messages	2	1	−1 (−50.0%)	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	1.3 KiB	1.3 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	2	2	0 (0.0%)	8	✅

Baseline: 71646e0 · PR result: 0d27d85 · Source CI: success

Scenario and decoded snapshot size

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

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

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

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

GitManager now retains pull request head SHAs through status lookup and fallback. When a merged PR’s head matches HEAD and its base matches the resolved default branch, remote status adjusts ahead counts. Tests cover squash-merge scenarios and related status cases.

Changes: Merged PR status

Layer / File(s)	Summary
Retain PR head SHA apps/server/src/git/GitManager.ts	Status lookup and last-known PR fallback results now carry the associated head SHA.
Calculate merged PR status apps/server/src/git/GitManager.ts, apps/server/src/git/GitManager.test.ts	Remote status zeroes eligible ahead counts when a merged PR’s head SHA matches HEAD. Tests cover retained or deleted branches, later commits, base branches, lookup failure, and a non-origin default branch.

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

Change: Bug fix · Severity of issue fixed: Medium

Suggested labels: bugfix

Suggested reviewers: t3dotgg, maria-rcks

Merge Risk: Minimal · up to 1c3d7

The merged-PR status change is ready to merge based on the reviewed behavior; dirty worktrees retain the combined action.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title clearly and concisely describes the main server-side fix: squash-merged branches no longer offer push or create-PR actions.
Linked Issues check		[#28467] The reported symptom is an enabled "Push & create PR" after a squash or rebase merge. The diff in apps/server/src/git/GitManager.ts keeps the PR head SHA through lookupStatusPr and the last-k…
Out of Scope Changes check		The whole-PR diff against merge base 2a4d128 touches apps/server/src/git/GitManager.ts and GitManager.test.ts. The GitManager.ts changes all serve the squash-merge status fix. They add headSha to Las…
Description check		The pull request description includes all required sections: Problem, Change, Scope and approval, and Verification. It explains the reproduction, implementation, scope, linked issue, test results, edg…

Full details: Linked Issues check

Explanation

[#28467] The reported symptom is an enabled "Push & create PR" after a squash or rebase merge. The diff in apps/server/src/git/GitManager.ts keeps the PR head SHA through lookupStatusPr and the last-known-PR fallback. readRemoteStatus sets aheadOfDefaultCount to 0 when the PR is merged into the default branch and its headSha equals HEAD. It sets aheadCount to 0 only when the branch has no upstream. Client logic in GitActionsControl.logic.ts then yields no create action. With counts at 0 and no upstream, resolveQuickAction returns a disabled "Push" hint, and buildMenuItems disables "Create PR". The issue's suggested fix (pass the head SHA through status, treat a matching merged PR as shipped) is implemented on the server. Commits added after the merge keep HEAD different from headSha, so counts stay non-zero. Dirty worktrees still reach the hasChanges branch of resolveQuickAction, which offers commit-based actions. Test coverage is taken from the PR description and was not re-run. Not addressed, and excluded by the PR objectives: the local porcelain reader and remote reader handling of a gone upstream (issue section 'Separately'), a guard in runStackedAction, and the historical change total.

Full details: Out of Scope Changes check

Explanation

The whole-PR diff against merge base 2a4d128 touches apps/server/src/git/GitManager.ts and GitManager.test.ts. The GitManager.ts changes all serve the squash-merge status fix. They add headSha to LastKnownPr and lookupStatusPr, change resolveLastKnownPr to return the full record, and compute shippedByMergedPr for the count adjustment. toStatusPr is unchanged, so the client contract does not change. No unrelated behavior change appears in the reviewed hunks. One description inaccuracy: the PR text says a dirty worktree keeps today's behavior. The server counts do change for a dirty worktree at a merged HEAD, and the menu item 'Create PR' becomes disabled until the user commits. The quick action still offers commit-based actions. This is a consequence within scope, not an unrelated change.

Full details: Description check

Explanation

The pull request description includes all required sections: Problem, Change, Scope and approval, and Verification. It explains the reproduction, implementation, scope, linked issue, test results, edge cases, evidence, and remaining limitations.

Finishing Touches

Generate docstrings

• Commit to this branch
• Create a new PR

Generate unit tests (beta)

• Commit to this branch
• Create a new PR

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

CodeRabbit Verify

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

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: fix/issue-28467
• File: .coderabbit.config.ts
• Commit: 1c3d7b3
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Team
• Run ID: 0e531dfb-923f-4305-8ae5-17a7d6c8b659

Commits

Reviewing files that changed from the base of the PR and between 5ff1482 and 1c3d7b3.

Files selected for processing: 2

• apps/server/src/git/GitManager.test.ts
• apps/server/src/git/GitManager.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 7 remain after this review.

[juliusmarminge]

Taking over this one