Prompt file imported from krzemienski/awesome-list-site (
.codex/prompts/verify-ds.md). Fill in{{arguments}}before use. Copyright stays with the author.
description: Run the awesome-list-site design-system verification and report an evidence-backed verdict argument-hint: [offline|full|deep] [routes]
Verify design-system work in awesome-list-site using the awesome-ds-verify skill.
Arguments: {{arguments}} — first word is the mode (offline default, full, or deep), the optional second is a comma-separated route list for the live probe.
- Read
~/.agents/skills/awesome-ds-verify/SKILL.mdand follow it. - Map the mode to the runner and execute it from the repo root:
offline→node ~/.agents/skills/awesome-ds-verify/scripts/verify-ds.mjsfull→ add--mode full(and--routes <list>if given). If the app is not reachable on :5000, start it (npm run dev, in the background, wait until it answers) — never substitute a mock or fixture.deep→ add--mode full --deep.
- Open the printed evidence directory. Read
VERDICT.mdand the.logof every failed gate. On a full run, view the Editorial and Brutalist screenshots underlive-probe/. - Triage each failure against the matching stage of
.agents/skills/verify-design-system/SKILL.md(checkreferences/rulebook-drift.mdin the skill first for stages 3, 7, 8, 9). - Report the verdict block with BLOCK / FIX / UNVERIFIED sections, each finding citing stage, gate and
file:linefrom the log, plus the run id and evidence path.
Rules: UNVERIFIED is never PASS. An offline run cannot justify "done". Do not edit awesome-list-site-ds/, do not grow a baseline, do not fix anything unless the user asked — report first.
