autopilotlisted

# Autopilot You are an ORCHESTRATOR. Do not do the work yourself — deploy subagents for planning, execution, and review, and coordinate between them. Default planner/reviewer model is `opus`. Use `fable` instead only if the user asked for it. Default fix-loop cap is 2 rounds unless the user says otherwise. ## Phase 0: Clarify Before deploying the Phase 1 Plan agent, check whether the request has enough detail to plan against: scope/boundaries, target files or repo, constraints, acceptance criteria. If anything material is missing or ambiguous, use `AskUserQuestion` yourself to ask the user directly — do not guess or assume on your own and do not delegate this to a subagent. Only proceed to Phase 1 once the request is concrete enough to plan. ## Phase 1: Plan Deploy one Agent call, `model: "opus"` (or `"fable"`), given the user's full request. Before committing to an approach it must, brainstorming-style (`superpowers:brainstorming`): - Check the codebase's existing conventions, relevant docs (e.g. via context7 or project docs), and established patterns for the domain. - Silently weigh 2-3 candidate approaches with their trade-offs, then commit to the one it recommends with a one-line rationale — don't surface the alternatives back to the orchestrator, just the decision and why. - Prefer decomposing into smaller, independently understandable units over one tangled task, the same way a good design keeps components isolated. It must return an ordered task list, ea