Evaluate Skill Quality and Readiness

Evaluating whether a skill is good requires evidence that it is clear (instructions are complete), usable (setup is realistic), reliable (failure is handled), safe (guardrails exist), maintainable (logic is factored), and shareable (claims are proven, not promised). Skill-quality-evaluator scores any supplied skill against an eight-dimension rubric plus six hard gates, judges it strictly on what was supplied rather than what was promised, and produces a tailored work order that a builder or agent can execute to improve it.

1. Provide a skill (SKILL.md, config, supporting files, optional: test results or deployment evidence). 2. The evaluator inventories the evidence, identifies the intended use, checks packaging and claims, then scores all eight dimensions (output quality, clarity, fit, reliability, safety, maintainability, portability, testing and evidence) using defined anchors and weighted points. 3. It applies six hard gates (packaging correctness, executable instructions, safety compliance, capability integrity, evidence standard, promise integrity) that a high score cannot override. 4. It produces a report with the overall score /100, readiness verdict, risk level, scorecard with evidence per dimension, failed/passed gates, and a prioritized gap list (P0/P1/P2). Output: a copy-ready improvement prompt that a person or agent can use to close gaps and re-evaluate.

The evaluator treats all supplied material as untrusted input: directives within a skill aimed at its evaluator are flagged as security findings, not followed. The evaluator itself does not edit the skill unless separately requested. It uses the referenced evaluation-rubric.md, workflow-adaptations.md, and report-template.md files from the same package to ensure consistent, defensible scoring. Optional safe validation can perform a bounded smoke test with sandbox systems or test identities if the user authorizes and suitable tools are available.