arbeitsplan-preflightlisted

# Preflight Know what is on the bench before measuring it. A run that discovers mid-swarm that the repo cannot make a worktree has already spent N dispatches on a result it cannot use. ## Steps 1. **Git and worktrees.** Is this a git repository? Can it create a worktree here — is the candidate directory ignored? A repo that cannot make worktrees can still use the matrix backend, so report the alternative rather than just failing. 2. **Acceptance checks.** What test command does this repo actually use? A run whose criteria carry no runnable check cannot be refereed, and that is better known now. 3. **Other plugins' live guards.** Report every guard that would see this run's edits — takt, cupertino, lehre, andon, confab, self-assess, nacharbeit — and its marker. A fix pass runs under them, and a denial from one is that plugin doing its job. **Never suggest another plugin's escape hatch.** Naming it as an inventory fact is reporting; recommending it is helping the user disable someone else's guard. 4. **Existing state.** Is `analysis/arbeitsplan/run_scope.json` open, and how old? Is `.claude/takt.local.md` present, and does it carry this plugin's provenance key? 5. **The matrix backend.** Is `claude` on PATH? Is `timeout` or `gtimeout` available (cells run untimed without one)? Is this session nested — because if so, the matrix runner will refuse, correctly, and the user must run it from a terminal. 6. **Say what the run can and cannot meas