Content
78%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.
A tightly written, highly actionable review procedure with an exact command set, rating rubric, and report template, weakened by one structural flaw: its central 'What to look for' section points to a `lens.md` that is absent from the bundle, leaving the code-target criteria unreachable. The skill is well-organized and concise, but the broken reference costs it both navigability and a fraction of its executability.
Suggestions
Ship `lens.md` with the skill (e.g., in `references/`) so the instruction "Read `lens.md` in this skill's directory and apply it" resolves — the current bundle contains only SKILL.md, making the core analysis criteria for code targets unreachable.
Inline a condensed version of the lens criteria for code targets, as the body already does for design targets under 'What to look for', so the skill remains operable even if the reference file is missing.
Add a verification checkpoint before writing the report, such as confirming each finding's `Where: <file>:<line>` location against the diff (or the spec file and heading for designs), to close the workflow's only validation gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence throughout: it gives the exact command ("Read the change with `git -C <checkout> diff <base> <head>`. Add `-- <path>` to read one file.") with no git tutorial, defines the rating tiers in one tight paragraph, and never explains concepts Claude already knows. Every section instructs rather than describes; there is no padding to trim, which fits anchor 5 rather than anchor 4's "minor instances of over-explanation". | 5 / 5 |
Actionability | Mostly executable guidance: exact diff command, explicit input contract with a recovery rule ("If the caller leaves one out, work it out and say what you used"), an inline list of design-target look-fors, and a copy-paste output template with field-by-field semantics. The gap is that the core analysis criteria for code targets are delegated to `lens.md` ("Read `lens.md` in this skill's directory and apply it"), and that file is not present in the bundle, so the single most load-bearing instruction cannot be executed from what ships here — slightly more than the "minor gaps" of anchor 5 but well above the pseudocode-level incompleteness of anchor 3. | 4 / 5 |
Workflow Clarity | The sequence is clear and complete: identify target type, gather inputs, read the target via the given git command, apply the lens, assign a rating from defined thresholds, and emit the fixed report shape, with the output template acting as a checklist (including the explicit empty case "write 'No findings.'"). Being read-only, the destructive/batch validation cap does not apply. It falls short of anchor 5 only because there is no explicit checkpoint verifying findings before reporting (e.g., confirming each "Where: <file>:<line>" against the diff), leaving a minor validation gap at anchor 4. | 4 / 5 |
Progressive Disclosure | The in-file structure is good — well-labeled sections (Targets, Inputs, Rules, Rating, Output format) and a single, clearly signaled reference ("Read `lens.md` in this skill's directory ... It explains how to judge scope, edge cases, coordination, and the shape of the code") that is one level deep. However, scored against the actual bundle structure, the referenced `lens.md` does not exist (no `lens.md`, no `references/` directory alongside SKILL.md), so navigation dead-ends at the skill's core judgment content; the body is also ~100 lines with the detailed criteria for code targets neither inlined nor resolvable, which is the "could be better organized" territory of anchor 3 rather than the clean split of anchor 4. | 3 / 5 |
Total | 16 / 20 Passed |