Content
66%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.
The body is highly actionable with excellent mode separation, validation checkpoints, and a properly externalized pattern catalog, but it is noticeably padded — the same scope, protection, and pass-accounting rules are restated across many sections. Tightening the duplicated policy language would substantially cut tokens without losing guidance.
Suggestions
Consolidate the pass-accounting and verification rules: state them once in the Editing Contract or Output format section and reference them from the mode sections instead of restating 'editing passes'/'no-op'/'budget' rules in at least four places.
Merge the three per-mode output-format descriptions into a single table or shared template with mode-specific deltas, keeping the four-item Verification checklist as the one canonical definition.
Either bundle the referenced examples/ files (examples/README.md, example configs like examples/technical.json) or remove/qualify those path references the way detector/validate.js is qualified ('If the repo ships the detector engine').
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | At ~5,000 words the body restates its own policy repeatedly: scope/protection/pass-budget rules appear in the Editing Contract, again in 'Iterate to convergence,' again across three paragraphs of the rewrite-mode output format ('0 editing passes' accounting is explained in at least four places), and again in detect and edit mode sections. This matches anchor 2 ('noticeably verbose; several unnecessary explanations or padded sections') — the content is novel, not things Claude already knows, so it is above anchor 1, but the duplication is well beyond the 'some unnecessary explanation' of anchor 3. | 2 / 5 |
Actionability | Guidance is largely executable: exact commands (`node scripts/normalize-quotes.js <rewritten-prose> --reference <original> --write`, `node scripts/check-style.js <file> --config <path>`), explicit flag syntax, quoted trigger phrases per mode, and defined output templates with required sections. It falls short of anchor 5 because there is no worked before/after audit example in SKILL.md itself, the `<rewritten-prose>` scratch-file workflow is intricate to execute, and two referenced paths (`examples/README.md`, `examples/<name>.json`) are not in the bundle. | 4 / 5 |
Workflow Clarity | Each mode has a clearly sequenced job list (edit mode: Read → Edit in place → Verify), validation is explicit and multi-layered (re-read the file, compare final text against source span-by-span, run the preservation validator with documented exit codes), and there are genuine feedback loops (corrective pass within a stated budget, residual reporting, the four-item Verification checklist: Editing passes / Checks / Residuals / Stop reason). This matches anchor 5's 'explicit validation steps; feedback loops for error recovery; checklists' despite the in-place file editing being potentially destructive. | 5 / 5 |
Progressive Disclosure | The heavy material (125KB pattern catalog) is correctly one level deep in references/patterns.md and clearly signaled ('Before auditing or rewriting any text, read references/patterns.md in full'), and referenced scripts (normalize-quotes.js, check-style.js) exist in the bundle. It stays at anchor 4 rather than 5 because the body references files absent from this bundle (examples/README.md, examples/<name>.json) and inlines substantial policy detail (editing contract, tone calibration) that makes the overview itself long. | 4 / 5 |
Total | 15 / 20 Passed |