---
name: ruvnet/autopilot-loop
source: https://app.decimal.ai/s/ruvnet-autopilot-loop@1/SKILL.md
source_sha256: b92b44c766b3
---

Run one autopilot iteration using Claude Code's native /loop:

1. Check status: `mcp__plugin_ruflo-core_ruflo__autopilot_status`
2. If all tasks complete or max iterations reached, call `mcp__plugin_ruflo-core_ruflo__autopilot_disable` and stop
3. Get prediction: `mcp__plugin_ruflo-core_ruflo__autopilot_predict` for the optimal next action
4. Execute the predicted task (spawn agent, edit code, run tests, etc.)
5. Log via `mcp__plugin_ruflo-core_ruflo__autopilot_log`
6. Schedule next: `ScheduleWakeup({ delaySeconds: 270, reason: "next autopilot iteration" })`

### Cache-Aware Scheduling

Always use delay 270s (under 300s cache TTL) to keep the prompt cache warm between iterations.

### Task Sources

Autopilot discovers tasks from:
- **team-tasks**: Claude Code TaskList entries
- **swarm-tasks**: MCP task_list entries
- **file-checklist**: Markdown checkbox items in tracked files

Configure: `mcp__plugin_ruflo-core_ruflo__autopilot_config({ taskSources: ["team-tasks", "swarm-tasks"] })`