Conciseness

The body is dense, imperative, and assumes competence — no concept explanations or padding, e.g. "You audit code changes against the project's own standards files... not to invent new rules." Not a 5 because the rule-matching idea is duplicated ("A skill compliance checklist does not apply to a TypeScript converter change" appears in both Standards discovery and What you don't flag), which could be tightened.

Actionability

Concrete, executable guidance throughout: specific violation categories with inline examples ("TodoWrite" instead of "TaskCreate", "learnings-researcher" instead of the fully qualified name), numeric confidence bands (0.80+/0.60-0.79/below 0.60), a two-part evidence requirement with a drop rule, and a copy-paste JSON output block. Not a 5 because it says "matching the findings schema" without showing that schema, and the output example omits where the calibrated confidence value goes — a minor gap for an instruction-only skill.

Workflow Clarity

The sequence is coherent: standards discovery (two paths, one with numbered steps) → match rules to file types → hunt violations → calibrate confidence → apply exclusions → evidence check → JSON output. Checkpoints exist (confidence suppression rule, evidence drop rule, exclusion list). Not a 5 because the overall review order is implied rather than stated as an explicit step sequence, and there is no verify/re-check loop before emitting findings.

Progressive Disclosure

Well-organized sections (discovery, hunting list, calibration, exclusions, evidence, output) with no bundle files to disclose; at ~72 lines of compact persona guidance, nothing clearly belongs in a separate file, so it sits just outside the under-50-line 5 exception. The one buried reference — "matching the findings schema" with no path or included schema — keeps it at 4 rather than 5.

Specificity

"Audits changes against the project's own CLAUDE.md and AGENTS.md standards" names the domain and one concrete action, then enumerates five specific check areas ("frontmatter rules, reference inclusion, naming conventions, cross-platform portability, and tool selection policies"). Falls short of 5 because there is only one verb (audits) and coverage is a category list rather than multiple distinct concrete actions; clearly above 3's '1-2 concrete actions'.

Completeness

The 'what' is clear and concrete; the 'when' is present via "Always-on code-review persona", which is explicit trigger guidance (so the missing-'Use when' cap of 3 does not apply), but it is less explicit and specific than a full 'Use when reviewing changes against project standards...' clause. Fits anchor 4 exactly: both present, 'when' could be more explicit.

Trigger Term Quality

Strong natural terms users would actually say: "CLAUDE.md", "AGENTS.md", "code-review", "frontmatter", "naming conventions", "portability". Missing common variations and synonyms ("conventions", "review my PR", "project rules", "agent standards"), so it does not reach 5's comprehensive synonym coverage; well above 3's partial keyword set.

Distinctiveness Conflict Risk

Auditing against a project's own standards files (CLAUDE.md/AGENTS.md) is a clear niche distinct from general code review. Minor overlap remains with generic code-review skills, since "code-review persona" could fire on any review request; not the minimal-conflict clear-niche case of 5.