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.
A well-conceived thin-index skill body with clear phasing, gates, and an output contract, undermined by a broken bundle: the three rule files that carry the actual detection and self-heal procedures are missing, so the skill cannot be executed as written. Redundant restatement of the confidence gate is the main token-efficiency drag.
Suggestions
Add the missing rules/static-check.md, rules/mutation-check.md, and rules/self-heal.md files (or inline their procedures) — the Required Reading by Phase table and the Phase-gate workflow currently point to nonexistent files.
State the confidence ≥ 90 % gate once in the Inputs section and reference it elsewhere ('see Inputs: --fix') instead of restating the threshold in the workflow table, decision flow, output contract, and Definition of Done.
Replace or supplement the pseudocode Quick Decision Flow with the concrete detection heuristic for at least the static check (e.g., what counts as a shadowed export) so the cheapest check is executable even without the rule files.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is largely lean and assumes competence ('No test files? Exit 0.'), but the confidence gate (≥ 90 %) is restated in the Inputs table, the paragraph below it, the Workflow table, the Quick Decision Flow, the Output Contract, and the Definition of Done — roughly six repetitions of the same fact. Anchor 5 requires every token to earn its place; this repetition could be consolidated into one canonical statement. | 4 / 5 |
Actionability | Some concrete guidance exists (argument table with defaults, `git diff --name-only <base>...HEAD`, `git restore <sut-file>`, structured report template), but the actual executable procedures — how static_check detects shadowed exports, how the mutation is performed and restored, how the extraction is planned — all live in `rules/static-check.md`, `rules/mutation-check.md`, and `rules/self-heal.md`, none of which exist in the bundle. The Quick Decision Flow is explicitly pseudocode. Key execution details are therefore missing, matching the 'incomplete / missing key details' anchor rather than the 'minor gaps' one. | 3 / 5 |
Workflow Clarity | The three phases are clearly sequenced with explicit gates ('Phase 2 runs only when Phase 1 passes', mutation check re-run after heal, '<TEST_CMD> re-run is green') and a Definition of Done checklist — feedback loops are present for this mutating operation. Not a 5: the per-phase operational detail is delegated to rule files that are absent from the bundle, and the pseudocode flow's `Skill("confidence", ...)` call lacks a concrete invocation contract. | 4 / 5 |
Progressive Disclosure | The thin-index design is correct in principle — 'This SKILL.md is a thin index... Detailed procedures live in rules/*.md and load on demand', one level deep, clearly signaled — but scored against the actual bundle structure, all three referenced rule files (static-check.md, mutation-check.md, self-heal.md) do not exist; only the two references/*.md files are present. The 'Required Reading by Phase' table sends the reader to dead links for every phase, leaving the skill's core procedure unreachable. Structure is more than 'minimal' but the broken reference graph pulls it below the midpoint. | 2 / 5 |
Total | 13 / 20 Passed |