speckit-autopilotlisted

# Spec Kit autopilot One prompt in. The agent then writes the spec, clarifies it, plans and cuts tasks, and ships reviewed milestones to `main`. The human is asked **only** for decisions that are theirs to make, and every ask is loud, specific and batched. **The agent is the reviewer.** No human reviews a spec, a plan or a diff in this flow, so every artifact and every diff gets an agent review. That review is done by a **fresh-context subagent**, never the context that wrote the artifact. A green CI run is not a review. **Violating the letter of these rules is violating their spirit.** Where a rule looks like it slows the flow down, it is the part that keeps an unattended flow from shipping the wrong thing. The flow and its diagrams: [README.md](README.md). Details live in `references/` and are loaded when the phase needs them. ## Entry | Argument | Start at | |---|---| | `resume` | Find this worktree's ledger (see *Resuming* below), confirm each recorded PR's `state` with `gh pr view`, and continue at the first unfinished step. **Never** reconstruct progress from `gh pr list` or from memory. | | `bug: …`, or wording describing broken behaviour | Phase 0 (bug path) | | anything else | Phase 1 | Before any phase, run `git fetch origin` and check the worktree is clean. Copy [templates/autopilot-ledger.md](templates/autopilot-ledger.md) into place the moment its directory exists: `autopilot.md` in the feature directory, or `bugs/BUG-<k>.autopilot.md` beside the BUG recor