This is a meta-skill that attempts to modify itself based on a "residual pyramid" approach to detect cognitive gaps in Claude's reasoning. It uses predictive coding concepts to trigger adaptive reflection, caches learned patterns to avoid redundant processing, and only accepts mutations that improve long term value. The documentation is mostly in Chinese and leans heavily on AI research terminology like value gating and experience replay. It failed the Gen Agent Trust Hub audit, which makes sense given it's literally trying to rewrite skills at runtime. Interesting experiment in autonomous skill evolution, but the practical benefits over just iterating on prompts manually aren't clear from the docs.
npx -y skills add jackjin1997/clawforge --skill "Self-Evolving Skill" --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