7 config files that turn Claude Code into an autonomous execution engine. Scans your project, prioritizes tasks, executes with self-validation, commits after each one. You review the PR.
AcePilot doesn't replace Claude Code. It configures it. When you type /acepilot, Claude Code absorbs your project and shifts into a different operating mode.
Reads your codebase, README, build output, TODOs, and specs. Returns a prioritized task list scored by impact, urgency, and confidence.
AUTO (exact fix, known pattern) → execute. PLAN (3-5 files) → show plan, proceed. ASK (multiple approaches) → stop, present options. Never over-commits.
Build, lint, test after every change. Then a separate qualify step: did the code solve the right problem? Two independent checks, not one.
8 state files track tasks, decisions, knowledge, and mode. Survives compaction, model switches, and session crashes. Resumes where it left off.
3 consecutive failures → stop. Same error 3 times → stop. Same file edited 3 times in one task → stop. Enters recovery mode, breaks the problem down, retries.
Commits after each verified task. Every checkpoint is a buildable state. Stashes dirty trees on start, restores on finish. Feature branches enforced.
From planning-only to fully autonomous with auto-push. Each mode inherits the safety systems — confidence gate, circuit breaker, qualify step.
Scan and plan. Don't execute. Review TASKS.md, adjust, then switch to an execution mode.
Execute tasks. Stop on issues. You stay in the loop on anything flagged.
Self-fixing. Auto-resolves review findings. Stops before push for your confirmation.
Same as auto + creates the PR automatically. ASK-level decisions still pause.
Zero interruptions. Scan to PR. Makes best-call decisions autonomously and logs them.
Outcome-driven. "Reduce load time" or "increase test coverage to 80%." You name the goal, AcePilot figures out the work.
Full test suite, code review, squash commits, prepare PR. The final gate before pushing.
Review all acepilot commits. Grouped findings. No auto-fixes. Your code review assistant.
Type one command. AcePilot scans, plans, executes, verifies, commits, reviews. You come back to a PR.
AcePilot installs into Claude Code's native config structure. No runtime, no daemon, no API keys. It's the same Claude Code you already use — configured to operate differently.
CLAUDE.md (trigger), settings.json (permissions + status line), commands/acepilot.md (the brain), agents/researcher.md + reviewer.md (Haiku scanner, Sonnet reviewer).
settings.json (hooks for code formatting and compaction survival), rules/browser.md (UI verification). Applied per-project.
8 flat files. TASKS.md, DECISIONS.md, KNOWLEDGE.md, CONTEXT.md, MODE, CIRCUIT, STASH_REF, USER-KNOWLEDGE.md. All plain text. All git-trackable.
Start with interactive modes. Pro unlocks full autonomy — the modes that let you walk away and come back to a PR.
Join the waitlist. We'll notify you when Pro launches.
No spam. Just a launch email when Pro is ready.
Or start with the free version now:
View on GitHub Read the docs