visual-qalisted

# Visual QA Visual QA is a review protocol. The project adapter owns capture; the skill supplies the state matrix, independent expert lenses, refutation, synthesis, and optional verified fixes. ## Inputs Infer these when safe; otherwise ask. - **adapter**: project-specific capture and test instructions in `adapters/<project>.md` - **tier**: `quick`, `full`, or `deep`; default `full` - **mode**: `review` or `fix`; default `review` - **scope**: flows and screens; default to the adapter's declared flows - **focus**: optional lens or risk to emphasize ## Capability discovery Before capture: 1. Inspect the adapter and project scripts for an existing screenshot command. 2. Inventory actually available browser, device, native-UI, and image-reading tools. 3. Prefer the project's deterministic capture harness. Do not assume Playwright, a browser MCP, or a particular agent runtime exists. 4. If capture or image inspection is unavailable, stop and name the missing capability. Never fabricate screenshots or visual findings. This repository does not ship a universal browser driver. Start from [`adapters/_template.md`](adapters/_template.md) when a project has no adapter. ## Pipeline 1. **Capture** — render a deterministic matrix of flows, states, breakpoints, and themes; add a clearly marked exploratory pass. 2. **Validate** — run `python3 scripts/validate_manifest.py path/to/manifest.json --check-files`. 3. **Review** — give each selected expert only its lens, negative constrai