What it does
An assay tests what something is made of, and how good it is. CraftAssay is an installable review skill for AI agents. It does that for a tool, site, skill, specification, or related portfolio.
An agent reads the skill and writes a prioritized report with findings, scores, and evidence. After you revise the work, compare it with the earlier run to see which findings are resolved, which remain, and why scores changed.
Get started: install the skill in your agent, then use it.
It scores usefulness, clarity, quality, and presentation. It is not a certification. It is not release approval.
What it reads, writes, and changes
| Reads | A project, site, file, URL, or paste |
| Writes | A dated run folder with report.md, a scorecard, findings, and coverage |
| Changes | Nothing. You decide what to fix |
| Dimension | Asks |
|---|---|
| Utility | Is the bounded result worth the effort for this user? |
| Accuracy | Are the inspected claims free of consequential errors? |
| Uniqueness | Is the approach distinct in a way that matters? |
| Quality | Is the inspected work complete and thoughtful? |
| Clarity | Can a reader see the job, the result, and the next step? |
| Aesthetics | Does the inspected presentation help or hinder? |
| Overall appeal | Would the stated audience want to try it? |
Each dimension gets a whole number from 1 to 10, where higher is better. NR means not rated, because the evidence was too thin to score. An interface nobody inspected gets NR for Aesthetics, not a guess.
CraftAssay and Cold-eye
Cold-eye checks readiness for a claimed first-use path. CraftAssay still reports first-use friction when it affects its scores. It does not issue a release-readiness verdict. A clean Cold-eye review does not mean CraftAssay will score the work highly.
A sample report. A sample comparison. The report shape.