Conciseness

The body is lean and telegraphic: a three-command workflow block, per-step behavior in a few sentences each, a hard-rules list, an exit-code table, and an annotated file list. No concept Claude already knows is explained (no WCAG primer, no library comparison), and every line maps to a command, a rule, or a file — matching the 5 anchor. It is above the 4 anchor because there is no padded or over-explained passage to trim.

Actionability

Guidance is fully executable and copy-paste ready: concrete commands with flags ("python scripts/finish_lint.py src/ --output json > lint.json", "node assets/capture_snapshot.mjs http://localhost:3000/invoices --width 320 --out s320.json"), documented exit codes per script ("0 clean · 2 findings · 3 bad input · 4 blocked"), and a no-Playwright fallback with the exact console snippet "copy(JSON.stringify(designQaSnapshot()))". This matches the 5 anchor with specific commands covering the common cases.

Workflow Clarity

The three steps are clearly sequenced (source lint → state matrix → rendered audit), and each step is itself a validation gate with documented blocking exit codes, making checkpoints explicit rather than implicit. Error handling is addressed through hard rules ("A waiver names the WCAG exception or a ticket… the scripts reject a waiver or NA with no reason", "Unverified is not pass") with the keyboard pass as the verification follow-up, and the sign-off file acts as the completion checklist — fitting the 5 anchor. It is not the 4 anchor because validation is built into every step via exit codes, not merely described.

Progressive Disclosure

The body is a concise overview and all detail lives one level deep: three reference files and six assets, each listed with a one-line description of its content in a "References and assets" section, and all listed paths exist in the bundle. Reference files contain no further document references (only an external source citation), so navigation is flat and well-signaled — matching the 5 anchor.

Specificity

The description enumerates many concrete defect classes it detects — "controls that do nothing, placeholder copy, removed focus rings, missing loading/empty/error states, low contrast, small targets, hidden focus and broken reflow measured in the real browser" — plus what it does about each (blocks vs. "flags template defaults without blocking on taste"). This is comprehensive coverage matching the 5 anchor; it exceeds the 4 anchor because no check category is missing.

Completeness

It explicitly answers both questions: "Use when a UI screen, page or component is about to be called done, handed off or shipped, or when generated UI looks generic or half-finished" gives concrete when-triggers, and "Gates the rendered result, not the design file - finds…" states clearly what it does. This matches the 5 anchor with explicit trigger phrases for both what and when.

Trigger Term Quality

Natural phrases a user would say are present — "about to be called done, handed off or shipped", "generated UI looks generic or half-finished" — giving good keyword coverage. It falls short of the 5 anchor because common variations like "design review", "polish", or "UI audit" are absent, but sits above the 3 anchor since the included triggers are varied and phrased the way users actually talk.

Distinctiveness Conflict Risk

The pre-ship framing ("Gates the rendered result, not the design file") carves out a clear niche distinct from design-generation skills, but the defect list includes "low contrast, small targets, hidden focus and broken reflow" — terms that overlap with a generic accessibility-audit skill's triggers. Mostly distinct with minor overlap risk against closely related a11y skills, fitting the 4 anchor rather than 5.