Skill Rubric Reviewer
When to Use
- User says review skill or task matches this capability
- User says skill rubric or task matches this capability
- User says audit SKILL.md or task matches this capability
- User says score skill quality or task matches this capability
Pre-publish validation loop
Run before publishing any skill:
python3 scripts/check_no_private_data.py
python3 scripts/publish_safety_check.py
python3 scripts/validate_collection.py
-
validate_collection.pycompletes successfully - Description includes what, when, and Triggers
- No private emails, paths, or LAN IPs
-
hitl-approverreferenced for destructive ops
Workflow
- Read SKILL.md and any references/, scripts/, assets/.
- Score 10 dimensions (frontmatter, description, conciseness, structure, clarity, freedom, errors, disclosure, scripts, completeness) 1-5 each.
- List issues for dimensions scoring 3 or below.
- Provide before/after rewrite suggestions; apply only if user requests.
- Flag publication blockers: secrets, harmful instructions, undeclared telemetry, PII in logs.
Integrations
skill-evolverhyper-skill-testerprivacy-redactor
Error Handling
| Failure | Response |
|---|---|
| Missing SKILL.md | Report path; do not score directory without file. |
Gotchas
- Rubric inspired by Agent Skills spec; independent implementation.
Safety & Ethics (Publication-Ready)
This skill is designed for public distribution. Constraints:
- Reviews content for safety before publish; blocks skills with harmful instructions.
- No automatic submission of reviews to external services without user consent.
- Does not exfiltrate skill contents to third parties.
Prohibited actions
- No unauthorized access, malware, or harmful automation
- No silent exfiltration of data, credentials, or telemetry
- No destructive system changes without hitl-approver
- No publication of user PII or environment secrets in outputs
Attribution
- Rubric aligned with Agent Skills spec best practices; original skill text.
Example
Input: User request matching triggers above. Output: Structured result per workflow; local artifacts only unless user opts in.