Skill Evaluator

**Quick judgement** Strong, well-structured evaluator skill that appears publish-ready and highly useful as a standard QA tool for OpenClaw/Clawdbot skills. It offers a solid mix of automated checks and a comprehensive manual rubric grounded in reputable frameworks. **Strengths** - Multi-framework rubric (ISO 25010, OpenSSF, HCI + agent-specific heuristics) covering 25 criteria across 8 categories. - Concrete scoring guidance (0–4 per criterion, 0–100 total) and clear interpretation bands (Excellent/Good/Acceptable/etc.). - Automated checks for file structure, frontmatter, script syntax, dependency/credential issues, and env-var documentation. - Well-documented workflow (quick start, stepwise evaluation process, EVAL.md template) that supports consistent use across reviewers. **Key risks / limitations** - Relies on a local Python environment (3.6+ with PyYAML), so automation depends on correct tooling setup and filesystem access. - Manual scoring still depends on reviewer judgment; consistency across reviewers will vary unless teams train on the rubric. - Security coverage is intentionally “baseline”; could create a false sense of completeness if users don’t also run deeper tools (e.g., SkillLens). - Quality of outcomes depends on how thoroughly reviewers read SKILL.md and the scripts; the tool guides but does not enforce depth of review. **Recommended scenarios** - Use as the default gatekeeper before publishing community or internal skills to Clawhub. - Integrate the automated eval-skill.py checks into CI to block structurally invalid or obviously unsafe submissions. - Standardize team code reviews around its rubric so multiple reviewers can produce comparable evaluations. - Pair with a dedicated security scanner (like SkillLens) for high-risk or sensitive skills where security is critical.