Skill Output Scoring

Core Workflow

1. Identify the output, scenario, target artifact, and scoring rubric.
2. Score each criterion with brief evidence, not just a number.
3. Separate output quality from prompt quality and missing context.
4. Call out blockers before averages.
5. Compare baseline and skill-enabled scores when both exist.
6. Recommend keep, revise, split, merge, defer, or retire based on the score and risks.

Safety Rules

• Do not average away high-risk failures.
• Do not treat polished writing as correctness.
• Do not score claims as correct without evidence or source material.
• Escalate when the output affects production, security, legal, finance, employment, medical, compliance, or customer commitments.