[don-petry]

Summary

Fixes #470 — checkout_pr_in_worktree aborts with fatal: '<branch>' is already used by worktree at '<stale path>' (exit status 128) when a crashed prior run on the same runner left the PR's head branch checked out in a now-defunct worktree. This broke review-changes on #383 (run 27107230249), which then reported ci-failing and could never be reviewed by the cascade.

Fix

Before gh pr checkout, checkout_pr_in_worktree now clears stale state:

1. git worktree prune — drops registry entries whose directories are gone.
2. Looks up the PR head branch and releases whatever holds it:
   • the main agent checkout → git checkout --detach in place (it can't be removed);
   • any other worktree → git worktree remove --force;
   • then git worktree prune again.

Tests

tests/dev-lead/unit/test_pr_worktree.bats — 3 new cases for every collision shape:

• PR branch held by a stale sibling worktree still on disk;
• stale registry entry whose directory was deleted;
• the agent checkout itself on the PR branch (detached in place, never removed).

Local: full dev-lead unit suite 372/372, shellcheck --severity=warning -x clean.

Note

Prior autonomous dev-lead runs reported this issue "Implementation Complete" but committed to the already-merged branch fix/dev-lead-pr-worktree-checkout (PR #449, closed) and never opened a PR — so the fix never reached main, and retriggers dedup-skipped on the stale dev-lead/issue-470-* branches. Implemented manually on a fresh branch off main.

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request introduces recovery logic to clear stale Git worktree states left by crashed prior runs, preventing checkout collisions when checking out a PR branch. It also adds comprehensive unit tests to verify recovery under various scenarios. A review comment suggests normalizing the worktree and agent directory paths using git rev-parse --show-toplevel before comparing them, ensuring robust matching on systems with symlinks.

[Copilot]

Pull request overview

This PR addresses dev-lead failures caused by leftover/stale git worktrees on the same runner, where gh pr checkout can abort with “branch is already used by worktree …” (issue #470), preventing automated review/fix runs from proceeding.

Changes:

• Add pre-checkout cleanup logic to prune stale worktrees and release the PR head branch from any existing worktree holder (including detaching the agent checkout when necessary).
• Extend unit coverage with new Bats cases for the main collision shapes (stale sibling worktree, stale registry entry, agent checkout holding the branch).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
scripts/lib/pr-worktree.sh	Adds collision-recovery logic before creating/checking out the PR worktree.
tests/dev-lead/unit/test_pr_worktree.bats	Adds regression tests covering worktree collision recovery scenarios.

[don-petry]

Dev-Lead — fix-bot-comment (no-changes)

Agent reasoning

Issues addressed: 0
Files changed: none
Skipped (informational): 0
```
The SonarCloud comment is a clean passing report — Quality Gate passed with 0 new issues, 0 security hotspots, and 0 duplication. There are no actionable findings to fix.
There are also no Tier 1 CI blockers (no failed/cancelled/timed-out checks) and no `CHANGES_REQUESTED` reviews. The PR is unblocked pending the in-progress checks (CodeQL, bats, SonarCloud scan, etc.) that are still running.

[don-petry]

Dev-Lead — fix-bot-comment (no-changes)

Agent reasoning

Issues addressed: 0
Files changed: none
Skipped (informational): 0
Notes:
  - Quality Gate passed: 0 new issues, 0 security hotspots, 0 accepted issues
  - All CI checks green (success or skipped — no failures)
  - No CHANGES_REQUESTED reviews — zero Tier 1 blockers
  - No open review threads from sonarqubecloud[bot] to resolve
```
The SonarCloud analysis is fully clean — no actionable findings and no stale threads to resolve. The PR is unblocked from this bot's perspective.

[donpetry-bot]

Automated review — APPROVED ✓

Risk: MEDIUM
Reviewed commit: 4af1196c3982623a8fea18d528e475bb7ef98483
Review mode: triage-approved (single reviewer)

Summary

Targeted dev-lead fix for issue #470: clears stale worktree state before gh pr checkout so a crashed prior run on the same runner can't permanently break review-changes for a PR. Logic is well-scoped (prune → release branch holder → prune again), distinguishes the agent's own checkout (detach in place) from sibling worktrees (force-remove), and ships with focused regression tests for all four collision shapes (sibling, missing dir, agent self-hold, agent self-hold from subdir cwd).

Linked issue analysis

Closes #470 (checkout_pr_in_worktree aborts with fatal: '<branch>' is already used by worktree at '<stale path>'). The diff directly fixes the failure mode described in the issue — prior runs leaving the PR head branch claimed by a defunct worktree — and the new bats cases reproduce each crash shape before exercising the fix. PR-body context about the previously-merged-but-orphaned #449 attempt is consistent with what's on main (fix/dev-lead-pr-worktree-checkout not present in current source).

Findings

Prior bot findings — all resolved in head commit 4af1196:

• Gemini (high): direct string-equality against PR_WORKTREE_AGENT_DIR would misroute on symlinked runners → fixed: both sides now canonicalized via pwd -P before comparison.
• Copilot: PR_WORKTREE_AGENT_DIR from pwd may be a subdir, breaking the equality check → fixed: git rev-parse --show-toplevel resolves to the repo root first; a dedicated subdirectory regression test was added.
• Copilot (×2): two new tests leaked mktemp -d parent dirs → fixed: both now rm -rf the captured parent at end.

Independent observations (non-blocking):

• gh pr view ... 2>/dev/null || true silently falls through to a no-op if the API call fails (rate limit, transient network); on collision the next gh pr checkout will then surface the original exit 128 — acceptable degradation, matches today's behavior.
• The cleanup runs under the EXIT trap chain established two lines earlier, so failures inside the loop won't leak the PR worktree.
• No secrets, no migrations, no GHA workflow changes, no new dependencies.

CI status

All required checks green: shellcheck, bats, unit-tests, CodeQL (actions+python), Agent Security Scan, AgentShield, Secret scan (gitleaks), SonarCloud (Quality Gate passed, 0 new issues), validate-agent-profiles, gh-aw-compile. Dependency-audit ecosystem matrices skipped as expected (no ecosystem files changed). PR is BEHIND main but MERGEABLE — base ref update only.

---

Reviewed automatically by the PR-review agent (single-reviewer mode: opus 4.7). Reply if you need a human review.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[don-petry]

Dev-Lead — review-changes (no-changes)

No changes were needed for this PR.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a699132666

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".

[chatgpt-codex-connector]

Avoid deleting worktrees by branch name alone

On a persistent runner, if this repo has any other registered worktree checked out to a local branch whose name matches the PR's headRefName (common for fork PRs using names like main or patch-1), this removes that worktree even though it may be unrelated to a stale dev-lead temp checkout. I checked git worktree remove -h; --force forces removal even when the worktree is dirty or locked, so this can discard active/uncommitted work instead of only releasing stale /tmp/dev-lead-wt-* worktrees.

Useful? React with 👍 / 👎.