OPS-07 · SEC. 03 Operations & DevOps

Parallel Worktree Risk Audit

Sweep every worktree before touching any of them, and label each safe or risky.

FORMAT: loop
DIFFICULTY: advanced
TIME: 15 min
TOOLS: claude-code · codex-cli
MODELS: any
COPIES: 0 so far

When to use this

You've got several worktrees or branches open at once (Conductor-style, one per feature) and you've lost track of which ones have uncommitted work, mid-rebase state, or stashes you can't afford to lose. Run this before acting on any of them.

The pattern

Pastes as plain text

I have multiple git worktrees/branches in flight and need a risk audit
before I touch any of them. Audit every worktree this repo knows about
(start from `git worktree list` here; if my worktrees live under a
different root, I'll name that directory in my next message):

1. List every worktree with `git worktree list`, and for each one:
   - Current branch name and whether it's tracking a remote.
   - `git status --short` output (uncommitted/untracked files).
   - How far ahead/behind its upstream it is
     (`git rev-list --left-right --count <branch>...origin/<branch>`).
   - Whether it has any stashes (`git stash list` run inside that worktree).
2. Classify each worktree as one of:
   - SAFE TO TOUCH — clean or only local commits, nothing shared depends on it.
   - CAUTION — uncommitted or stashed work exists; local edits are fine,
     do not push/merge/delete without asking.
   - DO NOT TOUCH — mid-operation (unresolved merge/rebase, detached HEAD,
     or a rebase/cherry-pick in progress).
3. Output a table: worktree path | branch | status | classification | reason.
4. If any worktree is DO NOT TOUCH, name it first and stop there — don't
   move on to suggesting actions on the others until I've seen that one.
5. Do not run any push, merge, branch delete, or `git worktree remove` as
   part of this audit. This is read-only reconnaissance.

Real example output

git worktree list: 4 worktrees under ~/projects/masjidos-worktrees/

| Path        | Branch                   | Status                           | Classification | Reason |
|-------------|--------------------------|-----------------------------------|-----------------|--------|
| rabat/      | feature/prayer-times-api | clean, 2 ahead of origin          | SAFE TO TOUCH   | Local commits only, nothing pushed yet |
| melbourne/  | feature/admin-dashboard  | 3 modified, 1 untracked           | CAUTION         | Uncommitted UI work, no stash |
| nagoya/     | fix/timezone-bug         | 1 stash ("wip-dst-edge-case")     | CAUTION         | Stashed experiment, don't discard |
| des-moines/ | main                     | rebase in progress (REBASE_HEAD)  | DO NOT TOUCH    | Interrupted rebase, needs manual resolve first |

DO NOT TOUCH: des-moines/ has an interrupted rebase (REBASE_HEAD present).
Resolve that before I take any action there. The other three are safe for
local work; melbourne/ and nagoya/ need your OK before anything leaves them.

Why it works

Auditing all worktrees in one read-only pass before acting on any of them catches the one mid-rebase or half-merged worktree before an unrelated command in a different worktree accidentally collides with it. Classifying instead of just listing forces a safe/risky decision up front, not mid-action.