Paste any AI skill, prompt pack, or custom instruction set. The Auditor grades it on five outcome-honest dimensions — will it deliver what it claims, and how does it fail when it's wrong — and flags the snake oil with receipts.
The rubric comes from the same research base we build client systems on: skills can inject knowledge and scope behavior, but they can't install competence that isn't there. The question is never "does it sound impressive" — it's "what happens when it's wrong."
Claims vs. mechanics — will it actually do what it promises?
Fail-safe skills degrade to mediocre. Fail-dangerous skills are confidently wrong.
Baked-in "facts" that aren't your current truth are anti-knowledge.
"You are a world-class CFO" is a costume. Procedure + context are substance.
A good skill knows when it applies — and when it doesn't.