ruvnet/ruflo/tree/main/plugins/ruflo-autopilot
ruflo-autopilot
학습, 예측, 진행률 추적, autopilot_* MCP 도구 10개를 제공하는 Claude Code용 자율 /loop 기반 작업 완료 플러그인입니다.
이 mod 소개
ruflo-autopilot은 10개의 autopilot_* MCP 도구(status, enable, disable, config, reset, log, progress, learn, history, predict)를 감싸고 Claude Code의 기본 /loop와 ScheduleWakeup으로 지속적이고 캐시를 인식하는 작업 완료 루프를 구동하는 Claude Code 플러그인입니다. team-tasks, swarm-tasks, 파일 체크리스트 전체에서 진행 상황을 추적하고 성공 패턴을 autopilot-patterns AgentDB 네임스페이스에 학습하며 다음 최적 작업을 예측합니다. 플러그인은 function-hook mod로도 제공되며 hooks가 플러그인과 함께 로드됩니다. 기본 활성화된 가드는 autopilot_log/autopilot_learn 쓰기에 포함된 비밀이나 토큰과 한도를 넘은 maxIterations를 차단하고, /autopilot-mod(status, scan, limits)는 로컬에서 응답하며 상태 파일은 .claude-flow/autopilot-mod/status.json에 기록됩니다. 옵션은 기본 활성화된 guard와 iterationCap(1000)입니다. /plugin marketplace add ruvnet/ruflo와 /plugin install ruflo-autopilot@ruflo로 설치하고 smoke.sh(예상 10 passed, 0 failed) 또는 claude plugin validate/test로 검증합니다.
설치
먼저 작성자의 README에서 marketplace와 플러그인 이름을 확인하세요. 저장소 구조에 따라 명령어가 달라질 수 있습니다.
claude plugin marketplace add ruvnet/ruflo claude plugin install ruflo-autopilot
원문 / README
ruflo-autopilot
Autonomous /loop-driven task completion with learning and prediction.
Combines Ruflo's 10 autopilot MCP tools with Claude Code's native /loop + ScheduleWakeup for persistent, cache-aware task completion loops.
Install
/plugin marketplace add ruvnet/ruflo
/plugin install ruflo-autopilot@ruflo
Features
- Autonomous loops: Enable autopilot, then
/loopdrives iterative task completion - Progress tracking: Monitors team-tasks, swarm-tasks, and file checklists
- Learning: Discovers success patterns from completed tasks via AgentDB
- Prediction: Predicts optimal next action based on state and learned patterns
- Cache-aware: ScheduleWakeup at 270s keeps prompt cache warm between iterations
Commands
/autopilot-- Enable, configure, or disable autopilot/autopilot-status-- Quick progress summary
Skills
autopilot-loop-- How to run an autopilot /loop iterationautopilot-predict-- Use learned patterns to pick the next task
MCP surface (10 tools)
| Tool | Purpose |
|------|---------|
| autopilot_status | Current autopilot state + learning stats |
| autopilot_enable | Turn autopilot on for the project |
| autopilot_disable | Turn autopilot off |
| autopilot_config | Read/update configuration |
| autopilot_reset | Clear learned patterns and progress (testing) |
| autopilot_log | Append a structured log entry |
| autopilot_progress | Progress summary across team/swarm/file checklists |
| autopilot_learn | Train on a completed task; writes to autopilot-patterns |
| autopilot_history | Browse past iterations |
| autopilot_predict | Predict the optimal next action from learned patterns |
All 10 are wired in v3/@claude-flow/cli/src/mcp-tools/autopilot-tools.ts.
Compatibility
- CLI: pinned to
@claude-flow/cliv3.6 major+minor. - MCP surface: the 10 tools above.
- Verification:
bash plugins/ruflo-autopilot/scripts/smoke.shis the contract.
Cache-aware /loop integration
Autopilot pairs with Claude Code's native /loop + ScheduleWakeup skills. The recommended fallback heartbeat is 270 seconds — under the 5-minute prompt-cache TTL so the next wake-up reads conversation context cached. Going past 300s pays a cache-miss; rounding to 5 minutes is the worst-of-both case.
For event-driven loops, arm a Monitor and let the 270s wake be the safety net.
Namespace coordination
This plugin owns the autopilot-patterns AgentDB namespace (kebab-case, follows the convention from ruflo-agentdb ADR-0001 §"Namespace convention"). Reserved namespaces (pattern, claude-memories, default) MUST NOT be shadowed.
autopilot_learn writes to this namespace via agentdb_pattern-store semantics — see ruflo-intelligence ADR-0001 for the 4-step pipeline this feeds (RETRIEVE → JUDGE → DISTILL → CONSOLIDATE).
Verification
bash plugins/ruflo-autopilot/scripts/smoke.sh
# Expected: "10 passed, 0 failed"
Architecture Decisions
As a mod
Autopilot also ships as a function-hook mod (ADR-445 pattern; hooks in hooks/, loaded with the plugin). No network, no process, no model call.
- Guard (default on): refuses an
autopilot_logorautopilot_learnwrite that holds a key, token or password, and anautopilot_configmaxIterationsaboveiterationCap. It only tightens: it never allows anything the session would deny, and the refusal never repeats the secret. Turn it off with theguardoption. /autopilot-mod: answered locally./autopilot-mod status,/autopilot-mod scan <text>,/autopilot-mod limits.- Status file:
.claude-flow/autopilot-mod/status.json(version,updatedMs, counters), written at session start and whenever a call is refused; the console reads it. - Options (
userConfig):guard(onby default),iterationCap(1000).
Test: claude plugin validate plugins/ruflo-autopilot, claude plugin test plugins/ruflo-autopilot, and bash plugins/ruflo-autopilot/scripts/smoke.sh.