Content
77%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 dense, expert-oriented audit methodology with excellent sequencing, evidence requirements, and validation feedback loops for its destructive operations. Its weaknesses are structural: the whole method lives monolithically in SKILL.md with no progressive disclosure, the scripts its proofs rely on are named rather than supplied, and the LICENSE link is dangling.
Suggestions
Move the Junk patterns catalog and Evidence-tier proof details into references/ files, keeping SKILL.md to the bar, the workflow, and clearly signaled one-level pointers.
Ship the Shape/Duplicate member-listing and matching scripts in scripts/ (or inline minimal runnable versions) so the tier proofs are executable as written rather than described.
Fix or remove the dangling [LICENSE](LICENSE) link: include the LICENSE file in the bundle or state the license attribution in plain text.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense expert prose with no padding and no explanations of concepts Claude already knows ("A dead-code tool in production mode, such as `knip --production`, lists them"), but keep/cut rules recur across "The bar", "Junk patterns", "Retention bar" and "Evidence tiers" — internal entries are defined in both "The bar" and the Internal tier — and several multi-clause sentences could be tightened. Not 5: minor redundancy across sections; not 3: almost every token earns its place. | 4 / 5 |
Actionability | Concrete procedures and named tools throughout ("Prove the first edit by mutating a scratch copy of the owner so the behavior breaks, and run the test", `knip --production`, a `git grep` of each name, the project's delete-by-title codemod), and validation steps name what to run. Not 5: the Shape and Duplicate proofs depend on listing/matching scripts that are described but never provided or sketched; not 3: the guidance is specific and executable, not high-level hints. | 4 / 5 |
Workflow Clarity | A numbered discovery sequence with explicit validation gates ("Report the evidence before editing", prove both edits on scratch copies), a Validate section with feedback loops ("Run each rewritten test against its named mutant and see it fail, and against its named refactor and see it pass"), and a per-candidate checklist with a readiness rule ("a missing field means it is not ready"). Not 4: checkpoints are explicit for every destructive or batch edit, including full-suite mode's ledger-resume protocol. | 5 / 5 |
Progressive Disclosure | Internally well-structured with clear sections and anchor links, but the entire methodology — a 16-item junk-pattern catalog, tier proof rules, and the Full-mode ledger protocol — is inlined in a single ~115-line SKILL.md with no references/ split, and the one external link, [LICENSE](LICENSE), does not resolve to a file in the bundle. Not 4: reference-grade catalogs clearly belong in separate files; not 2: the structure is good and nothing is buried. | 3 / 5 |
Total | 16 / 20 Passed |