自动巡环智能体工作流

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"] })