sprint-autopilot-onlisted

Engage autonomous story execution for BMad Method with git workflow integration. Implements stories end-to-end with automatic branching (git worktrees), commits, linting, and PR creation. Uses standard git worktree commands for story isolation — works with any coding agent. Falls back to stock BMad behavior when git is disabled. Use when user says "/sprint-autopilot-on" or "start autopilot".
ikunin/sprintpilot · ★ 3 · Code & Development · score 76

Install: claude install-skill ikunin/sprintpilot

## STOP — read this entire file before doing anything Sprintpilot is driven by a deterministic Node.js state machine at `_Sprintpilot/bin/autopilot.js`. The LLM owns in-skill execution, diagnosis, triage, and small-judgment decisions — not the flow. Follow **`./workflow.orchestrator.md`** verbatim. Flow control lives in `_Sprintpilot/bin/autopilot.js` (a Node CLI you call via `autopilot next` / `autopilot record`). The orchestrator emits actions; you execute them. ### Never improvise - Never decide which BMad skill runs next yourself — the state machine emits an `invoke_skill` action telling you. - Never skip the `autopilot next` → `autopilot record` cycle. Even when a step feels "obvious," route through the CLI so the ledger, verify, and bookkeeping enforcement run. - Do not search for `workflow.md` or reconstruct it from memory; do not read cached BMad legacy patterns and apply them ahead of the orchestrator's state machine. - **Never pause on your own initiative.** `user_input { kind: 'pause' }` is a HUMAN command. The autopilot's purpose is to drive without stopping until `session_story_limit`, a TRUE BLOCKER, retry-budget exhaustion, or `sprint_is_complete`. Heuristics like "PR opened, time for review" / "natural breakpoint" / "let CI catch up" / "context budget" / "clean checkpoint" / "merge cadence" / any meta-reasoning about session length, story count, or your own resource usage are NOT valid reasons to pause. Phrases like "User-initiate