What problem does it solve?
Coordinates and automates the nightly workflow-improvement loop so you no longer have to manually schedule research runs, update run state, or manage rejected plans and blocklists. It centralizes schedule control, focus selection, and rejection handling while delegating actual research work to a dedicated sub-skill.
Core Features & Use Cases
- Schedule Management: create and remove nightly cron triggers and store the trigger ID in state.
- State Ownership: read and write only designated fields (schedule_id, focus) in ~/.claude/recursion/state.md.
- Focus & Filtering: set or clear a thematic focus that downstream research runs consume.
- Reject & Blocklist: mark plans as rejected by updating plan frontmatter and appending reasons to blocklist.md.
- Delegation: dispatch synchronous research runs via the Skill tool or schedule background runs via RemoteTrigger without performing research itself.
- Use Case: run a nightly automated discovery session at 01:03, review proposed plans the next morning, accept or reject them, and keep a persistent blocklist of rejected ideas.
Quick Start
Invoke the orchestrator now by issuing the slash command /recursion now to run a synchronous research dispatch.