Session Handoff [agent-cleanup]: Cleanup Script Landed in #3186 and Hub Clone Swept, Skill Wiring Next #3187

Next steps, in priority order

1. PR 2 for Script the Local Half of Post-Merge Cleanup in pr_review.py cleanup #3184: point every skill that creates or finishes a worktree at the new commands. cleanup <n> --repo O/R --checkout <primary> replaces the raw local git steps. cleanup --sweep --repo O/R --checkout <primary> [--apply] covers what crashed or abandoned sessions leave behind. The surfaces:

   • repo-worktree "Listing and Cleanup";
   • drive-pr "The Drive Loop" step 4;
   • merge-and-release step 8;
   • backlog-burndown "Cleanup Is the Orchestrator's";
   • pr-review-conduct's post-merge line;
   • unattended-handoff, whose worker has no cleanup step at all;
   • resync-a-repo and audit-a-repo, whose hub-reach worktrees are never removed.

   Done means every surface names the command instead of raw git worktree remove / git branch -D / git push --delete, the generated skill trees are rebuilt with scripts/build_dist.py, and the PR is merged to develop. Edit under skill-lifecycle, and read scripts/README.md's cleanup paragraph for the exact contract.

2. Re-measure the related issues against the tree once PR 2 lands, and close or narrow each with evidence:

   • The Prescribed Post-Merge Remote-Branch Delete Cannot Run Where the Procedure Puts It #1682, merge-and-release's Cleanup Prescribes a git push the Write Guard Denies #1693 and Requirement 6 Denies the Remote Branch Delete That merge-and-release Step 8 Prescribes #1642: the remote delete has been scripted since before this round;
   • Decide whether merge-and-release's branch cleanup should use compare-and-delete #946: the local half now re-reads the tip before git branch -D, while the remote is still verify-then-delete;
   • backlog-burndown: nothing removes a standalone clone after its group merges #1301, backlog-burndown: a merged-but-undeleted remote branch claims its files forever #1302, backlog-burndown: cleanup leaves an abandoned group's claim and its pushed remote branch standing #1274, backlog-burndown: unconditional worktree retire, and a stopped clone group with no end state #1303 and backlog-burndown: the worktree-only disposition has no actor or timing, and its use sites disagree with it #1272: backlog-burndown cleanup gaps.

3. Promote develop to main with both PRs, where the "Closes Script the Local Half of Post-Merge Cleanup in pr_review.py cleanup #3184" line belongs, since closing refs register only on default-branch PRs.

4. Optionally sweep the other fleet primaries (for example Financial-Modeling) with cleanup --sweep --repo <owner/name> --checkout <its primary>: a dry run first, --apply on the maintainer's go-ahead.

External blockers

None.

Internal dependencies

• Step 2 follows step 1, since several of those issues are closed by the skill text, not the script.
• Step 3 follows step 1, so one promotion carries both PRs.

State

• #3186 merged to develop as 2f047fa. No release was dispatched.
• Its worktree, local branch and remote branch were removed by the new cleanup 3186 --checkout, the first real use of the local half.
• The hub primary is clean on develop at origin/develop.
• cleanup --sweep --apply ran on the hub primary:
  • removed 39 worktrees, 19 local branches with their 18 merged remote branches, and 4 stale entries;
  • left 4 detached Financial-Modeling hub copies that live sessions were using, kept by the 24h rule;
  • re-derive the current state with git worktree list and git branch.
• Script the Local Half of Post-Merge Cleanup in pr_review.py cleanup #3184 is open, with a comment correcting its premise.

The parked decision queue

0 issues carry decision in this repository.

What the last round did

• Landed #3186: pr_review.py cleanup --checkout, the local half, and cleanup --sweep, with the scripts/README.md contract.
• Corrected Script the Local Half of Post-Merge Cleanup in pr_review.py cleanup #3184's premise in a comment. gh-write-guard allows git branch -D and an unforced git worktree remove against a primary. The real blocker is that only python3 scripts/pr_review.py:* is on the permission allowlist.

What not to repeat

• Script the Local Half of Post-Merge Cleanup in pr_review.py cleanup #3184's helper-worktree route is unnecessary. The script runs git itself, so the guard never sees those calls.
• The local strict-review passes on this new code took nine passes to reach zero findings. Each round's fixes produced the next round's edge cases.
  • The per-push edit budget, with remaining findings listed in the PR body, is what let the PR keep moving.
  • Copilot's single round raised the same findings the body already listed.

New learnings

• git worktree prune --expire ages a prunable entry by its index file's mtime, not its gitdir file. This is recorded in CleanupSweep.entry_hours.
• scripts/pr_review.py already fails to import on Python 3.11, because of PEP 695 syntax that predates this round. A 3.11-only finding against it is moot.