visual-qa

Files (2)

• SKILL.md
• checklist.md

SKILL.md (first lines)

# Visual QA

Diff against **intent**, not against taste. Intent = page-spec + DIRECTION + TOKENS + design-rules.

## Process

### 1. Load

Page-spec, `DIRECTION.md`, `TOKENS.md`, `docs/agents/qa.md`, `docs/agents/design-rules.md`.

Note the `## Rule exceptions` table in `DIRECTION.md` — those deviations are already accepted.
Everything else is graded on its merits: `DIRECTION.md` prose is intent to check *against*, not a
licence to downgrade a design-rule miss.

### 2. Gather evidence

Follow qa.md: screenshots or live pass at listed breakpoints. If you can read images from the environment, do so. Use [checklist.md](./checklist.md).

### 3. Grade findings

| Grade | Meaning |
| --- | --- |
| **blocker** | Violates hero budget, brand-first, wrong page job, broken CTA, or unusable on a target breakpoint |
| **major** | Clear drift from direction/tokens/spec sections; should fix before ship |
| **nit** | Polish; optional |

State the **expected** (cite spec/direction line) vs **observed**. No vague "make it pop".

### 4. Report

Write `docs/site/qa/<slug>.md` (or issue comment if that is the project norm) with:

- Scope + breakpoints checked
- Pass/fail on acceptance checks
- Findings table (grade, location, expected, observed, fix hint)
- Recommended next skill:
  - Findings open → `/implement-ui`
  - Clean, and `IA.md` still lists un-built pages outside `Deferred` → `/page-spec` for the next one
  - Clean, and the sitemap is covered → `/ship-checklist`

**Completion criterion:** every acceptance check marked pass/fail; every fail has a grade and fix hint.

From Alfred-Lau/skills, MIT.

More from Alfred-Lau/skills