This is the top-level autonomous improvement loop that runs post-mortem cycles until you tell it to stop. It wraps the RPI skill in a fitness-driven outer loop: measure what's broken via goals and DORA metrics, pick the worst thing from harvested follow-ups or failing directives, fix it with a full research-plan-implement run, then measure again and repeat. The work selection ladder is concrete, starting with freshly harvested next-work items, then open beads, then failing goals, then coverage gaps, then complexity mining. Quality mode prioritizes post-mortem findings over new features. Dream owns overnight knowledge compounding, evolve owns daytime code compounding. This is the operator's "make everything better until I interrupt" command, not a one-shot task runner.
npx skills add https://github.com/boshu2/agentops --skill evolve