dao-skill

Invoke the Skill with your rough idea and it first reduces the request to a root problem, core tension, and system structure. When a writable target exists, it then generates the SKILL.md, references, examples, and validation files directly rather than only returning a plan.

Ask for an evaluation and the Skill declares an evidence level, runs a five-item Trust Gate covering permissions, sensitive data, input safety, dependencies, and environment, then applies a 100-point rubric. A Trust failure blocks publication regardless of the numeric score.

It requires only Python 3.9 or later with the standard library. All validation scripts, including the installer, quality checks, and behavior-contract checks, are dependency-free and run on Linux, macOS, and Windows.

No. Structural review is evidence level E1 and caps reliability and effectiveness scores. Verified publishable claims require E3 executed evidence such as regression replays, and benchmark claims require E3 or E4 independent evaluation.

It runs an evolution protocol: builds an evidence packet, retrieves the nearest existing rule, decides create, merge, or discard, patches conservatively, retests the old failure, and records deployment status with a rollback condition.

Generated child Skills go to an explicit user path or the project-local .dao/skills directory, never inside the dao-skill source repository. SkillBank state, traces, and ledgers stay under DAO_SKILL_HOME or the project .dao directory.