Conciseness

The body is lean and assumes competence ('Prompts persuade; hooks decide.') with no explanations of concepts Claude already knows; every section carries rule content. The 'Canonical response anchors' section lightly restates terms already used as headers, but it is a deliberate terminology lock, not padding — matching the 5 anchor rather than 4.

Actionability

Concrete, near-executable guidance: exact decision semantics ('Exit 2: block. Print the rule violated and the sanctioned alternative on stderr') and a class/trigger/decision table, with a copyable policy template one reference away. Not 5 because the body itself contains no literal config or hook snippet — a populated example of one guardrail class in .claude/settings.json syntax would make it copy-paste ready. Not 3 because the guidance is specific and executable in substance, not pseudocode or high-level hints.

Workflow Clarity

The test-lock workflow is sequenced with validation checkpoints ('Write the failing test first', 'Unlock only when the fix is verified'), and fail-closed behavior covers the destructive production classes, so the destructive-cap is not triggered. Not 5 because the skill's overall authoring flow — pick classes, fill template, map to host adapters, verify enforcement — is implicit across sections rather than presented as one ordered procedure with an explicit verify step in the body (verification lives in the runtime-adapters reference).

Progressive Disclosure

The ~75-line body keeps the standard inline and delegates correctly: 'Guardrail Policy Template' -> references/guardrails-template.yaml and 'Runtime Adapters' -> references/runtime-adapters.md, both real files, clearly signaled in a References section, one level deep, with no nested references. This matches the 5 anchor's clear overview with well-signaled one-level-deep references.

Specificity

The description lists multiple concrete actions — 'protected paths, test-file locks during bug fixes, post-edit formatters, production approval gates, secret deny rules' — covering the domain comprehensively, matching the 5 anchor rather than 4 because there are no notable coverage gaps.

Completeness

It explicitly answers both: 'what' — define deterministic guardrails with five enumerated classes — and 'when' — 'Use when writing or reviewing a hook policy, or moving an always-do-X rule out of prose into enforcement.' The when-clause contains concrete trigger phrases, matching the 5 anchor exactly.

Trigger Term Quality

Good natural terms ('hook policy', 'guardrails', 'approval gates', 'protected paths') that users would plausibly say, but common variations and concrete artifacts like 'PreToolUse', 'hooks.json', or '.claude/settings.json' appear only in metadata triggers, not the description itself. Not 3 because keyword coverage is solid, not partial; not 5 because synonyms and file names a user would mention are missing from the description.

Distinctiveness Conflict Risk

Clear niche (deterministic hook-enforced rules vs. prose instructions) with distinct triggers ('hook policy', 'moving an always-do-X rule out of prose into enforcement'); minimal conflict risk with other skills. The neighboring 4 anchor's 'minor overlap risk' does not apply because the scope is sharply bounded.