This is the orchestration layer for autonomous work sessions in Claude Code. It runs a continuous loop that checks task status, predicts the next best action, executes it (whether that's spawning an agent or editing code), then schedules itself to wake up in 270 seconds to do it again. The timing is deliberate: staying under the 300 second cache TTL keeps your prompt context warm between iterations. It pulls tasks from multiple sources including Claude's native TaskList, MCP task lists, and even markdown checkboxes in your project files. Useful when you want to kick off a development session and let Claude work through a backlog without manual intervention between each step.
npx -y skills add ruvnet/ruflo --skill autopilot-loop --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
juliusbrussee/caveman
mattpocock/skills
shadcn/improve
obra/superpowers
forrestchang/andrej-karpathy-skills
vercel-labs/skills