Score a skill against a rubric: review run, review view, review list, review retry. Three scored areas — validation (structure and format), description (how well the description triggers the skill at the right time), and content (how well the instructions guide the agent once loaded).
It is deliberately the cheap instrument of a pair, and tessl is careful about the distinction: a review tells you whether a skill is well written; an eval tells you whether it works. A review is fast, structural, runs automatically on publish, and is explicitly compared to “static analysis for code”. Reach for it for quick structural feedback; reach for the eval when you need evidence of effect.
The description score is the one with no code analogue. In a skill, the description is not documentation — it is the trigger condition the agent matches a task against, so a badly written description means a correct skill that never loads. Scoring it separately from the content is a recognition that a capability has two independent failure modes: never firing, and firing badly.
Two mechanisms make it governance rather than advice. review run --threshold fails the build below a score, so substandard skills never merge (pattern-deterministic-gates); and the default rubric is forkable, so an organization encodes its own bar as a versioned artifact and applies it to every review → artifact-standards. “When a team shares skills, quality stops being a personal matter. A weak skill that one person tolerates becomes everyone’s problem once it is in the shared repository, shaping how every agent on the team behaves.”
See Also
- tessl-eval — the expensive sibling that measures effect rather than form.
- tessl-review-security — the safety scan run alongside the quality score.
- tessl-review-fix — iterates a skill until it clears the threshold.
- artifact-review-report — the output.