Content
67%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, operationally serious policy document: concrete commands, strict mode/option semantics, and unusually thorough verification and pass-budget enforcement. Its costs are repetition of the same gating rules across four sections, a muddled editing-pass accounting subsection, and references to bundle paths (examples/, detector/) that do not ship.
Suggestions
Consolidate the gating rules (protected content, source fidelity, pass budgets, residual reporting) that are restated across 'Editing contract', 'Modes', 'Output format', and 'Mechanical check' into one canonical section referenced from the others — roughly a third of the body is currently redundant restatement.
Rewrite the editing-pass accounting prose in the 'Iterate to convergence' section as a short rule table or two bullets (default ceiling 2 passes; --iterate 1 caps at 1; rechecks and audits consume none); the current paragraph form is the hardest-to-follow part of the document.
Fix or hedge the dangling bundle paths: 'examples/README.md' and 'examples/<name>.json' are stated as if they ship, but the bundle contains only references/patterns.md and scripts/ — either bundle them or mark them as user-supplied/optional like the 'if the repo ships the detector engine' hedge.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | No filler explaining concepts Claude already knows — everything is skill-specific policy — but the same gating rules (protected content, source fidelity, pass budgets, residual reporting) are restated across the 'Editing contract', 'Modes', 'Output format', and 'Mechanical check' sections, and the editing-pass accounting prose ('One editing pass is one stage that changes the returned text...') is convoluted. Mostly efficient content that could be tightened considerably. Not 2: none of it is generic padding of known concepts. | 3 / 5 |
Actionability | Gives concrete commands with flags and exit codes ('node scripts/normalize-quotes.js <rewritten-prose> --reference <original> --write', 'node scripts/check-style.js <file> --config <path>' exit 0/1/2), an explicit option mapping ([--mode...], [--voice...], [--iterate 1|2]), and per-mode numbered job lists. Not 5: 'examples/README.md', 'examples/<name>.json', and 'detector/validate.js' reference paths that do not exist in the bundle (only references/patterns.md and three scripts ship), and the '<rewritten-prose>' scratch-file convention is placeholder-level. | 4 / 5 |
Workflow Clarity | Each mode has a numbered sequence (Audit/Rewrite/Summarize; Read/Edit in place/Verify), verification is mandatory and checklist-structured ('Editing passes', 'Checks', 'Residuals', 'Stop reason'), and feedback loops exist (re-read the file after editing, re-validate within budget, no-op detection before drafting). Not 5: the editing-pass budget rules ('--iterate 2... use the same two-pass ceiling as the default') take several reads to follow, and the actual end-to-end flow must be reassembled from four interleaved sections rather than one linear sequence. | 4 / 5 |
Progressive Disclosure | One-level-deep reference clearly signaled ('Before auditing or rewriting any text, read references/patterns.md in full' in a marked reference-loading block; verified the file exists), and the three scripts are invoked with concrete commands. Not 5: the ~43-line inline severity-tier list duplicates catalog entries already in references/patterns.md ('Cutoff disclaimers', 'Chatbot artifacts'), and the examples/ and detector/ paths dangle against the actual bundle structure. | 4 / 5 |
Total | 15 / 20 Passed |