SkillForge

Why we included this project

SkillForge is for teams that build and maintain skills for Claude Code or Codex and want to treat those skills as engineered artifacts rather than loose documents. It routes any skill request to the right workflow, and its core idea is an evidence loop: a fresh agent runs a task without the skill, then with it, and the skill only ships if the with-skill run clears the recorded baseline failures. Per-skill regression evals catch degraded behavior as the ecosystem changes, and cross-runtime discovery indexes personal, Codex, and Claude Code plugin caches so nothing gets skipped. If you've been writing skills by hand and guessing whether they actually help, this gives you a concrete, repeatable test loop instead.