agentsclimarketplace

Design qa

Skill DKeken/codex-skills-alternative/skills/design-qa

17 vendor-neutral Agent Skills (SKILL.md) reimplementing the Codex Creative Production + Product Design plugin workflows. Works with Claude Code, opencode, Cursor, Codex CLI, Gemini CLI, Copilot.

Install
npx -y skills add DKeken/codex-skills-alternative --skill design-qa

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 13 stars13 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Check a prototype against its original design direction and requirements before shipping. Compares implementation to the mockup/spec/context, flags drift in layout, spacing, type, color, states, and behavior, and reports a pass/fix list. Use when the user says "design QA", "does it match the design", "check against the spec", "conformance check", or before sharing/handing off a prototype. Part of the Product Design set. Uses a browser tool for visual comparison.

SKILL.md

3.9 KB, as published. Nobody here has run it

Design — Design QA

Verify the built prototype actually matches the intended design + requirements. Catch drift before it ships. Unlike design-audit (open-ended critique of any UI), QA is a conformance check against a known source of truth — it answers "does this match what we agreed to build?"

When to use

  • A prototype is built and you need to confirm it matches the mockup/spec before review.
  • Final gate before design-share or dev handoff.
  • "Does it match the design?", "conformance check".

When NOT to use

  • No spec/reference to compare against → that's open critique → design-audit.
  • Nothing built yet → build it first.

Inputs

  • The built prototype (local URL or path).
  • The source of truth: reference image/mockup, ./design/context.md, ./design/ideate/*.
  • Acceptance/success criteria (from context).

Workflow

  1. Capture the implementation via a browser-automation tool — screenshots of key screens and states (loading/empty/error/success).
  2. Visual conformance — compare to the reference: layout, type scale, spacing rhythm, palette, radii/shadows, iconography. Note every drift with the expected vs actual value.
  3. Behavioral conformance — walk the flow: do interactions, states, and navigation match the intended UX?
  4. Requirements conformance — does it satisfy the goal + success criteria from context? Any scope item missing?
  5. Accessibility sanity — contrast, focus order, target size, keyboard reachability (flag what needs manual AT testing — automated checks aren't full WCAG).
  6. Write ./design/qa.md: PASS items + prioritized fix list (severity × effort) + a verdict.

Output: ./design/qa.md

prototype: http://localhost:5173   source: ./mocks/onboarding.png + context.md
visual:
 ✓ layout matches  ✗ body type 14px (spec 16px)  ✗ accent #7C3 (spec #6366F1)
behavior:
 ✓ happy path  ✗ no error state on empty name (spec requires inline error)
requirements (vs success criteria):
 ✓ 3-step wizard  ✓ progress saved  ✗ missing "skip" affordance (scope v1)
a11y:
 ✗ CTA contrast 3.1:1 (<4.5)   manual: screen-reader pass still required
verdict: NOT READY — 4 fixes (2 HIGH: error state, contrast). Loop to design-prototype.

Quality bar

  • Compared against the locked source of truth, not personal taste.
  • Every miss cites which source it violates (mock / context / success criteria) + expected vs actual.
  • A11y findings distinguish automated checks from what needs manual AT validation.
  • A clear verdict: READY, or NOT READY with the blocking items named.

Common pitfalls

  • QA by opinion — drifting into "I'd prefer..." instead of "spec says X, build shows Y".
  • Skipping states — only checking the happy path; error/empty/loading are where drift hides.
  • Passing HIGH drift — don't green-light something with blocking visual/behavioral mismatches.
  • Overclaiming a11y — automated contrast/focus ≠ full WCAG conformance; say so.

Handoff

PASS → design-share. FAIL → loop the fix list back to design-prototype (or design-image-to-code for visual drift), then re-QA. Don't share stakeholder-facing prototypes with unresolved HIGH items without explicitly flagging them.

Tooling

Needs a browser-automation tool to capture screens/states for comparison. Without one, do a code-level + manual review and flag that visual conformance is unverified.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.