autoresearch plugins

by Surah Li

Iteratively improve any skill using eval-grounded autoresearch and Hamel's Three Gulfs methodology.

by XRenSiu

Ratchet — Goal-driven multi-agent persistent optimization system. Combines Goal-Driven master/subagent separation with AutoResearch signal design methodology. Implements an 'independent evaluation + kill-and-restart + ratchet progress' autonomous loop: master only judges, subagent only executes; if a subagent stalls or claims success without meeting acceptance criteria, it is killed and replaced. Suitable for long-running autonomous coding tasks that require verifiable deliverables and explicit termination conditions.

by XRenSiu

Skill Evolution Engine — Darwin-style autonomous SKILL.md optimizer. Applies a 9-dimension rubric (60 structural + 40 effectiveness; structure now elevates the three SkillLens-validated high-signal dimensions — failure-mechanism encoding / executable specificity / high-risk-action blacklist — that lift pairwise judge accuracy 46.4%→73.8%), default multi-judge independent scoring, mandatory with-skill-vs-no-skill baseline comparison (negative-transfer is a hard non-ship gate), and git-backed ratchet hill-climbing (keep-or-revert) to evolve any Claude Code skill from initial draft toward 90+. Ports SkillOpt's three stability controls: rejected-edit buffer (dead-ends.md) + slow-update memory (learnings.md) + text learning-rate (≤30-line edit budget). Inspired by Karpathy's autoresearch, alchaincyf/darwin-skill, and Microsoft SkillLens/SkillOpt.