/cs-design-qa¶
Gate $ARGUMENTS before it is called done.
If there is no running URL, run steps 1 and 2 and say that the rendered audit is still owed. Do not call the screen done without it.
python ${CLAUDE_PLUGIN_ROOT}/skills/design-qa/scripts/finish_lint.py <source path>
python ${CLAUDE_PLUGIN_ROOT}/skills/design-qa/scripts/state_matrix_checker.py --matrix <states.csv>
node ${CLAUDE_PLUGIN_ROOT}/skills/design-qa/assets/capture_snapshot.mjs <url> --width 320 --out s320.json
node ${CLAUDE_PLUGIN_ROOT}/skills/design-qa/assets/capture_snapshot.mjs <url> --width 1440 --out s1440.json
python ${CLAUDE_PLUGIN_ROOT}/skills/design-qa/scripts/rendered_audit.py --snapshot s320.json
python ${CLAUDE_PLUGIN_ROOT}/skills/design-qa/scripts/rendered_audit.py --snapshot s1440.json
No state matrix yet? Draft one from the source (which surfaces fetch data, which are forms, which start async work) using ${CLAUDE_PLUGIN_ROOT}/skills/design-qa/assets/states.example.csv, and mark only what you can see as DONE.
Report¶
- Verdict per step (the script's first line), then DONE or NOT DONE.
- Blockers, grouped by fix, each with file:line or element id and the one-line fix.
- Missing states, blocking ones first.
- Template traits and generic risk. Say whether the brief asks for the look; never block on it.
- What the scripts could not check: unverified contrast, the screen-reader pass, and
a11y-auditfor source accessibility.