The Self-Evolution skill empowers AI agents to transition from simple retries to permanent learning by identifying and repairing systemic failures in real-time. By classifying errors into Gaps, Failures, or Regressions, the skill guides Claude through evidence collection—including DOM snapshots and screenshots—gated code edits, and rigorous verification steps. Crucially, it maintains 'The Map,' a persistent domain playbook that ensures once a fix is applied, the agent retains that knowledge for all future interactions, effectively preventing the recurrence of stale selectors, broken automation scripts, or missing API helpers.
주요 기능
01Automated evidence collection including DOM snapshots and screenshots
02Three-tier failure classification for Gaps, Failures, and Regressions
03Permission-gated code editing within defined repository boundaries
04Persistent domain playbooks to ensure long-term learning and memory
05Detailed evolution logging for transparent audit trails of autonomous changes
063 GitHub stars
사용 사례
01Automatically building missing API helper functions when encountering an unimplemented capability
02Repairing broken web automation scripts when a website's DOM structure changes
03Maintaining a persistent knowledge base of UI selectors and timing requirements across multiple sessions