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.
An exceptionally disciplined instruction skill: dense, unpadded prose with concrete executable commands and explicit validation loops. Its main weaknesses are bundle integrity — four referenced files absent and one orphaned — which leaves the end-to-end workflow and the artifact schema unreachable from this file alone.
Suggestions
Ship the missing referenced files (workflows/review.md, workflows/codebase-review.md, workflows/qa-review.md, schemas/review-finding.md) in the bundle, or inline the minimal artifact-schema skeleton so the output contract is self-contained.
Link references/perf-qa.md from SKILL.md (e.g., from the workflows table or QA-review section) so the bundle file is discoverable.
Move the heavily procedural detail of the mutation-stability wait loop (the verbatim sh -c poll snippet, timeout handling, self-matching pgrep caveat) into the mutation-stability script's --help or a dedicated reference, keeping the pairing rationale in the body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and dense with zero padding: no concept explanations Claude already knows, detail delegated to references ("Read references/finding-disposition.md before classifying findings. It owns verification, dispositions, recurrence and the filing bar"), and sibling-skill rules cited rather than restated. Every clause encodes a rule the reviewer cannot derive on its own. | 5 / 5 |
Actionability | Concrete, copy-paste-ready commands throughout: the two review-artifact-check invocations with the fix-until-ok loop, the full mutation-stability flag list, the verbatim sh -c foreground-poll snippet, and the git archive/mktemp/df extraction recipe. It stops short of 5 because the central output format (the findings JSON) is delegated to ./schemas/review-finding.md, which is missing from this bundle, so the artifact shape is not actionable from what is here. | 4 / 5 |
Workflow Clarity | Sequences are explicit with real validation checkpoints and feedback loops: capture starting fields, write the artifact, then "Fix until this prints \"ok\": true"; re-review rounds define suppression and re-report rules; mutation-stability bounds runs by budget and timeout. Not 5 because the top-level diff-to-findings-to-verdict flow lives in workflows/review.md, which is absent from this bundle, leaving the end-to-end sequence only partially visible. | 4 / 5 |
Progressive Disclosure | References are one level deep and clearly signaled, and finding-disposition.md, review-depth, and mutation-stability all exist in the bundle — but four referenced paths are missing (./schemas/review-finding.md and all three workflows/*.md in the workflow table), and references/perf-qa.md is an orphan no path in SKILL.md reaches. Navigation is partly broken, which is more than a minor organization gap. | 3 / 5 |
Total | 16 / 20 Passed |