Content
71%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a tight, well-structured instruction sheet with an exact output contract and sensible rules, and it makes the right call keeping the lens out of SKILL.md. Its critical flaw is that the referenced lens.md — which the body says defines slop, what to look for, and how to choose the rating — is missing from the bundle, leaving the skill's core evaluation criteria unavailable.
Suggestions
Ship lens.md alongside SKILL.md (or inline a condensed version of its slop definitions and the Low/Medium/High rating bands) so the core detection criteria and rating-selection rules are actually reachable.
Add a brief fallback in the body for when lens.md cannot be found, e.g. a one-line summary of each rating band, so the review can still produce a defensible Rating line.
Number the workflow steps (gather inputs → read the diff → apply the lens → write the report) so the sequence and the rating checkpoint are explicit rather than implied by section order.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~60-line body is lean throughout: concrete inputs, a one-line git command, terse rules ("Do not charge the change for problems in code it does not touch"), and an output template with no concept re-teaching or padding. Every token earns its place and it assumes Claude's competence, matching anchor 5; the only mild redundancy (read-only stated in the intro and Rules) is not enough to drop to 4. | 5 / 5 |
Actionability | Concrete guidance is present — "git -C <checkout> diff <base> <head>" with the "-- <path>" variant and an exact machine-readable report template ("The last line must be `Rating: <value>` with nothing after it") — but the core detection criteria and how to choose Low/Medium/High are delegated to lens.md ("Read lens.md in this skill's directory and apply it. It defines slop, what to look for, and how to choose the rating"), and that file does not exist in the bundle. Key details are missing, fitting anchor 3; not 4 because the central evaluation guidance is unresolvable, not 2 because the commands and output contract are fully executable. | 3 / 5 |
Workflow Clarity | The flow — gather inputs (with a recovery path: "If the caller leaves one out, work it out and say what you used"), read the diff, apply the lens, emit the fixed report — is clear and the read-only rules remove risky operations. Minor gaps keep it at anchor 4: the steps are implied by section order rather than sequenced, and the rating-decision step is underspecified because the lens that defines it is absent; not 5 since a checkpoint on how findings map to the rating is missing. | 4 / 5 |
Progressive Disclosure | The architecture is right — a lean SKILL.md delegating definitions, detection criteria, and rating method to a single clearly-signaled, one-level-deep file — but the referenced lens.md is not present in the skill's directory (no references/, scripts/, or assets/ exist, and no lens.md anywhere in the bundle). The promised detail layer is absent and navigation is broken, which fits anchor 3; not 4/5 because the disclosure structure does not function as written, not 2 because nothing is inlined or buried and the reference is well signaled. | 3 / 5 |
Total | 15 / 20 Passed |