Back to directory

Skills Constitution

A cross-platform meta-rule set that forces an AI agent to check its installed skills before acting, load a matching skill when one exists, and search when none does, installable across Claude Code, ChatGPT, Cursor, and Gemini.

Skills Constitution

Skills Constitution is a set of governing rules, distributed as an installable skill or rule file, meant to sit above every other skill or tool an agent has. It defines a pre-check step that forces the agent to look at its capability registry before doing specialized work, a mandatory-use rule that requires loading a matching skill instead of falling back to a generic response, and a search-first rule for when no skill matches, with a deterministic classifier deciding whether a task counts as simple or specialized.

An install script detects the host platform and adapts: skill-directory platforms such as Claude Code get the skill copied in and their skill tree automatically rebuilt, including skills bundled inside plugins; rule-file platforms such as Cursor, Windsurf, or Cline get the rules written into their rule directory; and prompt-only platforms such as ChatGPT or Gemini get an extracted injection template to paste into custom instructions. An optional flag registers host hooks that enforce the rules automatically, with backup, rollback, and idempotent reinstallation.

The project is maintained primarily in Chinese, with an English README provided separately.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.