Skill Audit Scorecard

Use this scorecard after Tessl so the audit stays evidence-based and repo-aware.

Blockers

Treat these as must-fix before calling the skill ready:

• invalid frontmatter or missing name or description
• description fails discovery because it does not say what the skill does and when to use it
• commands, paths, or links are stale or broken
• workflow has no clear start, no evidence loop, or no completion criteria
• the skill conflicts with the target repo's guidance or conventions

Major Findings

These usually lower trust or activation even if the skill technically works:

• name is vague, generic, or forgettable
• SKILL.md is bloated with detail that belongs in references/
• boundaries and handoffs are missing or muddy
• the skill asks the model to invent deterministic steps that should be scripted
• examples are abstract instead of practical
• optimizer use is suggested without explicit approval

Minor Findings

These are worth tightening after the blockers and majors:

• wording is repetitive or over-explains obvious concepts
• output format is implied instead of stated
• references exist but are not linked from SKILL.md
• agents/openai.yaml lags behind the skill's current wording

Audit Dimensions

Score each dimension qualitatively as strong, mixed, or weak:

• Discovery: can the right request trigger the skill from metadata alone
• Workflow: does the body tell the agent how to proceed and what evidence to gather
• Progressive disclosure: is detail placed in the right file
• Repo fit: does it match local conventions and links
• Verification: does it use Tessl or another concrete loop instead of taste-only review
• Boundaries: does it say when not to use the skill and where to hand off

Compact Review Template

scope audited:
Tessl:
strengths:
findings:
recommended changes:
rerun: