Content
92%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 tightly engineered verification skill: the deterministic path is executable and matches the real bundle, result handling and repair routing are unambiguous, and the bounded repair loop with explicit stop conditions is exemplary workflow design. The only weakness is repetition of the model-only/don't-overclaim rule across several sections, which could be consolidated.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and procedural with no concept explanations or fluff, but the model-only/don't-overclaim rule is stated three times ("Never claim it ran unless the current host executed it", "Report this review as model-only rather than claiming the deterministic validator performed it", "Report that portion separately as model-only semantic review") and the repair-loop bound is repeated across three sections, so some trimming is possible. Not 3 because there is no padded explanation of things Claude already knows; not 5 because the duplicated disclaimers and run-on paragraphs (e.g. the 'Use the explicit warn residual policy...' paragraph) exceed 'every token earns its place'. | 4 / 5 |
Actionability | Gives fully executable, verified guidance: `node scripts/validate.js --residual-policy warn before.md after.md` matches the actual CLI, and `const { validate } = require("./scripts/validate.js")` matches the actual module export; both were confirmed against the bundled script. Result statuses (PASS/REVIEW/FAIL), envelope fields, and the repair-owner routing table ('returned text -> voice-preserving-rewriter', 'named file -> file-edit-in-place') are concrete and unambiguous, matching the copy-paste-ready anchor. | 5 / 5 |
Workflow Clarity | The full verification loop is explicitly sequenced with validation checkpoints and a hard feedback-loop bound: require both versions (return to router if missing), run the deterministic validator, apply a separate semantic review, emit a status, route REPAIR only 'while the shared editing budget has room', verify once more, then 'Stop and report on a second verification failure. Do not start another repair loop.' This matches the anchor for clear sequence with explicit validation steps, feedback loops for error recovery, and checklists (PASS/REVIEW/FAIL criteria are each enumerated). | 5 / 5 |
Progressive Disclosure | SKILL.md is a ~106-line procedural overview while 194KB of implementation (scripts/validate.js at 574 lines, scripts/patterns.js at 3,762 lines) is correctly externalized and referenced with executable usage; the heavy patterns data never leaks into the overview. Cross-skill references (`../avoid-ai-writing/SKILL.md`, `../avoid-ai-writing-router/references/handoff-contract.md`, `agency-role-lenses.md`, `skill-graph.json`) are clearly signaled and one level deep, matching the well-signaled one-level-deep-references anchor. | 5 / 5 |
Total | 19 / 20 Passed |