Content
92%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.
An exceptionally lean, executable workflow body with explicit validation checkpoints and error recovery. Its one real defect is bundle integrity: the Rules section and one in-step link point at rules/ files that are not present in the skill bundle, so progressive disclosure fails at navigation time despite good structure.
Suggestions
Add the missing rules/ files (dispatch.md, output-contract.md, lenses.md) to the bundle — all three links in the Rules section, plus the rules/dispatch.md link in step 1, currently resolve to nonexistent paths.
If the rules are meant to live in the full reviewer's tree rather than this skill's, replace the relative ./rules/ links with an explicit pointer to that location (e.g. 'in agents/pr-reviewer/rules/') so the paths are unambiguous.
Verify every relative link resolves as part of release checks; the ✓/✗ reporting contract shows the same discipline applied to output, and broken reference links deserve the same guard.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every line is operational — the transport-selection table, "each call also pays opencode's own fixed context (about 194k tokens, reported in `usage.json`)", exit-code semantics, and the two-line ✓/✗ reporting contract all carry information Claude does not already have. Lean throughout with no concept explanations to trim, matching the anchor-5 'every token earns its place' example; anchor 4 would require at least minor over-explanation, and none is present. | 5 / 5 |
Actionability | Two copy-paste-ready commands with all flags shown (`node ${CLAUDE_SKILL_DIR}/scripts/lite-review.mjs --pr <url|#n> --out <dir> --repo-dir <checkout>` and the lite-compare invocation), plus concrete failure handling ("Exit code 3 means the first model call got a 401 or 403") and a concrete output contract ("lite found 2 findings in 9 of 11 files; 2 files not reviewed"). Fully executable with the common cases covered; anchor 4 would require minor gaps, and none are evident. | 5 / 5 |
Workflow Clarity | Four clearly sequenced steps with explicit validation and recovery: step 2's checkpoint ("Read `<dir>/skipped.json`. Every file lite did not review is listed with its reason; report that list, never 'reviewed everything'"), the exit-3 feedback loop ("Report 'the review did not run', name the rejected key, and skip step 3"), and step 3's precondition ("after the full reviewer has posted on the same head"). The skill is shadow/read-only (writes artifacts only), so no destructive-cap applies; not 4 because checkpoints are explicit rather than implicit. | 5 / 5 |
Progressive Disclosure | Signaling is good in principle — each reference carries a one-line label ("what is reviewed, grouped, and skipped") and stays one level deep (references/design-rationale.md, scripts/lite-review.mjs, scripts/lite-compare.mjs all exist) — but `rules/dispatch.md` (linked twice), `rules/output-contract.md`, and `rules/lenses.md` resolve to nothing: no rules/ directory exists in the bundle, so the entire Rules section is dead navigation. Not 4 because half the referenced paths are broken; not 2 because the body's own split (main flow inline, rationale and scripts correctly external) is appropriate and clearly organized. | 3 / 5 |
Total | 18 / 20 Passed |