Conciseness

The body is dense operational prose with zero padding about concepts Claude already knows — every section states contract rules, statuses, or commands. However, the honesty rule about not claiming execution ('Never claim it ran unless the current host executed it', 'Do not claim the deterministic validator checked semantic representation details that it does not implement', 'label the result as model_only') is repeated across three sections and could be consolidated. This matches anchor 4 — efficient with minor trimming possible — not the every-token-earns-its-place anchor 5.

Actionability

Both code paths are copy-paste ready and verified against the actual bundle: the CLI invocation 'node scripts/validate.js --residual-policy warn before.md after.md' matches the script's real argument parser, and 'const { validate } = require("./scripts/validate.js")' with 'validate(original, rewritten, { residualPolicy: "warn" })' matches the real export signature. Concrete statuses (PASS/REVIEW/FAIL), envelope fields (execution_evidence.verifier, verification_summary.status, pass.index/pass.max), and named routing targets cover the common cases, meeting the score-5 anchor.

Workflow Clarity

The sequence is clear — accept inputs (require both versions, else return to router), run the deterministic validator or fall back to model-only, apply the semantic and representation lenses, then route PASS/REVIEW/FAIL outcomes — with an explicit bounded feedback loop: 'At most one repair re-entry is allowed... After repair, verify once more. If that check still fails, stop and report.' Validation is the core of the skill and every branch has explicit checkpoints, matching the score-5 anchor with feedback loops for error recovery.

Progressive Disclosure

Structure is good: the body stays a concise overview, the referenced scripts/validate.js (534 lines) and the cross-skill router references are one level deep and clearly signaled with concrete commands. The gap is that the bundled scripts/patterns.js (3443 lines, a runtime dependency auto-required by validate.js for residual scoring) is never mentioned anywhere in SKILL.md, leaving a large bundle file without navigation. That minor organization gap fits anchor 4 rather than the fully well-signaled anchor 5.

Specificity

The description enumerates the exact checked content classes — 'code, frontmatter, quotes, tables, links, paths, numbers, headings, and residual AI-pattern regressions' — and names the verification action ('deterministic check', 'rewrite preserved protected content'). Coverage of check targets is comprehensive with no meaningful gaps, matching the score-5 anchor; it exceeds the score-4 anchor, which requires minor gaps in coverage.

Completeness

Both halves are explicit: the 'what' is 'a deterministic check for [enumerated content classes]' and the 'when' is a three-part 'Use when' clause covering direct before/after supply, explicit preservation questions, and deterministic-check requests. The trigger guidance is as explicit and specific as the score-5 anchor example, not the score-4 anchor where 'when' could be more specific.

Trigger Term Quality

Natural user phrasings like 'provides an original and rewritten version', 'asks whether a rewrite preserved protected content', and 'deterministic check' are present, but common synonyms such as 'before/after', 'verify', 'unchanged', or 'diff' are missing. This sits between anchor 4 ('good keyword coverage; a few natural terms missing') and anchor 5 (comprehensive synonyms/extensions), fitting 4 best.

Distinctiveness Conflict Risk

The primary triggers (original+rewritten pair, preservation question) carve a clear verification niche distinct from rewriter or detector skills, but the tertiary trigger 'residual AI-pattern regressions' overlaps with a sibling AI-writing-detector skill's territory. That is exactly the score-4 anchor — mostly distinct with minor overlap risk against a closely related skill — rather than the minimal-conflict score-5 anchor.