Land and clean up worktrees

What you get: an inventory of your worktrees with a class for each, and the merged clean ones removed after your GO.

Inputs

• repos (optional) — repo paths to scan; default is the current repo's main worktree (git rev-parse --path-format=absolute --git-common-dir, parent dir)
• <owner/repo> for each repo comes from git -C <repo> remote get-url origin

Steps

1. Ask — repos (default as above) → run log in the start directory — repo list confirmed; stops for approval
2. using-git-worktrees — per repo: git -C <repo> worktree list --porcelain → inventory table (path, branch, HEAD) in the run log — the main worktree, the worktree running this playbook, and any entry marked locked or prunable are keep
3. github — gh auth status (fails → log it, skip the squash predicate, only ancestry merges count); git -C <repo> fetch origin --prune; default branch: default=$(git -C <repo> symbolic-ref --short refs/remotes/origin/HEAD); default=${default#origin/} (the command prints origin/main) → both in the run log — fetch fails → log it, classify against the local refs, mark the table "stale"
4. Classify, first match wins:
   • dirty: git -C <wt> status --porcelain --ignored shows anything other than !! entries; !! (ignored files, deleted by removal) are listed per row in the GO line
   • detached HEAD (no <branch>): always local-only, never removed
   • merged: git -C <repo> merge-base --is-ancestor <branch> origin/<default>, or (squash merge) gh pr list -R <owner/repo> --head <branch> --state merged --json number,headRefOid has a PR whose headRefOid equals the worktree HEAD (git -C <wt> rev-parse HEAD); a reused branch name with a different headRefOid does not count
   • remote-only: the upstream exists and git -C <wt> rev-list --count @{u}..HEAD is 0 (safe on the remote, unmerged)
   • local-only: everything else; commits exist only here
   • check: every row has exactly one class plus the output of its evidence command
5. pr-recap — per non-merged worktree: recap file from pr-recap; if it cannot be built log git -C <wt> log --oneline origin/<default>..HEAD and git -C <wt> diff --stat origin/<default>...HEAD instead — one recap or fallback per row
6. [GO] removal list = rows classified merged and not keep, shown in full with these exact commands per row: git -C <repo> worktree remove <path> (label the row merged (empty) when git -C <repo> rev-parse <branch> equals git -C <repo> rev-parse origin/<default>; never --force; git refuses dirty trees), then git -C <repo> branch -d <branch> (-d refuses squash-merged branches → log "branch kept, needs -D, human decides"; never -D). The run stops here until the human types GO. GO covers exactly that list, once. The hook does not cover these commands, so this GO is their only guard. Never remove dirty, remote-only, local-only or keep rows, and never use rm -r.
7. Verify — git -C <repo> worktree list → removed paths are gone, kept rows unchanged
8. Hand-off — remote-only and local-only rows → next action per row in the run log ("land via /pb-ship", or "push needs GO in that worktree") — nothing runs

Run log: production_artifacts/pb-worktrees-land-<date>.md in the start directory, never committed

Rules

• Anything other than the literal GO (case-insensitive) is not a GO; a GO covers only that one step, one time.
• GO is by contract in every harness. The go-gate hook is only a backstop on Claude Code, OpenCode and agy (Codex: unverified; Cursor, Kimi: none). A missing hook is never permission to proceed.
• A failed check stops the run: write the failure into the run log and report. No silent retries.
• Write one run-log line per step as it completes (N. done|skipped|failed — artifact — check result) and WAITING FOR GO: <step> at each gate.