Content
60%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 well-organized, appropriately cautious about scope creep, and token-efficient, but it delegates its entire analytical substance to `pr-review-guide.md`, which is absent from the bundle. As shipped, the skill tells Claude to follow checks and named risks it cannot actually read.
Suggestions
Bundle `pr-review-guide.md` under `references/` (or inline its checks and named risks into SKILL.md) so the repeatedly referenced analysis content actually resolves — currently every mention of 'the guide' is a dangling pointer.
Replace the 'When to Use' section's verbatim copy of the frontmatter description with a short, distinct usage note; the duplication carries the description's mid-sentence truncation ('pastes code asking...') and a 'pR' typo into the body.
Spell out at least one concrete example check per category (change propagation, duplication, complexity) in the body so the skill remains actionable even before the guide is loaded.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence ("A large diff or long function is a review lead, not by itself a demonstrated defect"), with no re-teaching of known concepts. The only waste is the "When to Use" section repeating the frontmatter description nearly verbatim, including its "pastes code asking..." truncation and the "pR" typo, so it is not a 5. | 4 / 5 |
Actionability | Concrete elements exist — the finding record format ("Symptom (source location), Source (verified contract or principle), Consequence (reachable effect), and Remedy (small actionable fix)") and the report format ("mode `PR Review`, scope, findings by severity, checks actually run, limits and next action") — but the core analysis instructions are delegated to `pr-review-guide.md` ("the guide's change propagation, cognitive overload, duplication, complexity, dependency and domain checks"), and that file is not present in the bundle, so the actual checks cannot be executed as written. This is 'some concrete guidance but incomplete; missing key details', not the minimal-hints level of 2. | 3 / 5 |
Workflow Clarity | Five clearly numbered steps with an epistemic checkpoint ("Separate confirmed findings from questions and unavailable evidence") and an explicit terminal condition ("No findings is a valid result"). Not 5 because there is no explicit verify/re-validate loop, and step 2's checks ("when applicable") depend on the missing guide, leaving a gap in what is actually run. | 4 / 5 |
Progressive Disclosure | The body is cleanly sectioned and the one reference is clearly signaled ("Read `pr-review-guide.md` for the retained analysis process"), but the bundle contains no `references/` directory and no `pr-review-guide.md` — the reference is dangling, and the checks, named risks, and evidence the body repeatedly points to (four separate mentions of "the guide") have no home in the skill. The disclosure structure collapses to a broken pointer rather than a working overview-to-detail split. | 2 / 5 |
Total | 13 / 20 Passed |