Content
56%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 sophisticated, carefully hedged skill whose description and mode contracts are unusually rigorous, with real executable commands, verification checklists, and feedback loops. Its weaknesses are token efficiency — the same pass-budget, protected-content, and fidelity rules are restated three or more times — and bundle hygiene: several referenced paths (examples/, detector/validate.js) do not resolve while one bundled script is orphaned.
Suggestions
Consolidate the editing-pass budget, protected-content, and source-fidelity rules into a single authoritative section; they are currently restated across "Iterate to convergence", "Automatic marks pass", the Editing contract, and the output-format sections, which costs hundreds of lines without adding rules.
Fix the dangling references: either bundle examples/README.md, examples/<name>.json, and detector/validate.js, or mark them as components of the full distribution; likewise reference or remove the orphaned scripts/markdown-prose.js.
Move the Severity tiers and House style sections into references/patterns.md alongside the catalog they summarize, keeping SKILL.md as a lean operating manual that points to the detail file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 317-line body restates the same policies repeatedly: editing-pass budget semantics appear in "Iterate to convergence", "Automatic marks pass", and the rewrite-mode output section; protected-content and source-fidelity rules each recur three or more times across the Editing contract, mode sections, and output format. Dense interlocking prose like "A corrective edit and a preservation repair share the requested budget: once its limit is reached, report any residual or verification failure instead of changing the text again" forces multiple readings, matching the "noticeably verbose; several padded sections" anchor. | 2 / 5 |
Actionability | Concrete executable commands are given ("node scripts/normalize-quotes.js <rewritten-prose> --reference <original> --write", "node scripts/check-style.js <file> --config <path>" with exit codes 0/1/2, "node detector/validate.js --residual-policy warn <original> <rewritten>") plus explicit mode-trigger phrases and fully specified output formats. The gap from 5 is the mix of abstract contract prose and the missing referenced paths (examples/README.md, detector/). | 4 / 5 |
Workflow Clarity | Each mode has a numbered job list ("1. Read the file… 2. Edit in place… 3. Verify"), validation checkpoints (re-read the file, compare final text with source), feedback loops (apply missing edits within the remaining budget, re-validate), and an explicit four-item stop-reason checklist. The 5 anchor is missed because the workflow is scattered across interleaved sections (contract, modes, marks pass, output format) rather than presented as one coherent sequence. | 4 / 5 |
Progressive Disclosure | The main reference is well handled — a clearly signaled reference-loading block directs reading references/patterns.md (which exists, 870 lines, one level deep) — but the body also references "examples/README.md", "examples/<name>.json", and "detector/validate.js", none of which exist in the bundle, while bundled scripts/markdown-prose.js is never referenced at all. Navigation is partially broken and a large amount of policy detail is inlined, matching the "some structure but could be better organized" anchor. | 3 / 5 |
Total | 13 / 20 Passed |