Content
73%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 a strong, dense behavioral contract: modes are unambiguous, commands are executable, and the verification framework (pass accounting, residuals, stop reasons) is exceptionally rigorous. Its weaknesses are repetition — pass-counting, residuals, and protection rules are each explained multiple times — and two dangling examples/ references. Tightening the restated rules and fixing or removing the missing-path citations would raise conciseness and progressive disclosure.
Suggestions
Consolidate the editing-pass, residual, and protected-content rules into one authoritative section and reference it from Output format instead of restating each rule 3-4 times across the Iterate and Output format sections.
Fix or remove the dangling references to examples/README.md and examples/technical.json (either ship the examples directory or describe the config schema inline), since the body presents them as bundled paths.
Add a compact worked before/after example (a short flagged sentence, its rewrite, and the Verification report) to ground the audit-to-rewrite workflow without relying on the external patterns.md catalog.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body mostly adds non-obvious policy (editing contract, provenance constraints) rather than explaining known concepts, but restates the same rules repeatedly: editing-pass counting appears in "Iterate to convergence" and four more times in Output format, residuals are explained in three places, and protected-content rules appear in both the Editing contract and Output format. Mostly efficient but could be meaningfully tightened — not 2 (no padded concept explanations), not 4 (the repetition is substantial, not minor). | 3 / 5 |
Actionability | Provides copy-paste-ready commands ("node scripts/normalize-quotes.js <rewritten-prose> --reference <original> --write", "node scripts/check-style.js <file> --config <path>", "node detector/validate.js --residual-policy warn <original> <rewritten>"), explicit mode triggers ("detect," "flag only," "scan"), and option syntax. Not 5: no worked before/after audit example, and the referenced "examples/README.md" / "examples/technical.json" paths do not exist in the bundle. | 4 / 5 |
Workflow Clarity | Each mode has numbered steps (Read → Edit in place → Verify; Audit → Rewrite → Summarize), explicit validation (re-read the file, span-by-span comparison of final text against source), feedback loops (repair within remaining editing budget, re-validate), and a literal four-item Verification checklist (Editing passes / Checks / Residuals / Stop reason). This matches the 5 anchor's sequenced steps with validation and error-recovery loops. | 5 / 5 |
Progressive Disclosure | The 125KB pattern catalog is properly externalized to references/patterns.md (one level deep, clearly signaled with reference-loading markers and a read-first instruction), and referenced scripts scripts/normalize-quotes.js and scripts/check-style.js exist. Not 5: examples/README.md and examples/technical.json are cited as factual paths but absent from the bundle, and ~300 lines of dense behavioral policy sit inline in SKILL.md; not 3, because the heavy content is well split and clearly signaled. | 4 / 5 |
Total | 16 / 20 Passed |