| name | arm-design |
| description | Use when the user asks Codex to design, build, improve, or critique a polished visual artifact where visual taste, hierarchy, typography, layout, brand feel, or marketing positioning are main evaluation criteria. Covers premium marketing sites, one-page landing pages, design taste tests, frontend visual design, polished product pages, restrained premium software, clickable prototypes, browser-native slide decks, motion animations, MP4/GIF exports, design variants, infographics, and expert reviews. Use this before imagegen; call imagegen only from inside this workflow when a raster asset materially improves the page. Do not use for backend features, production SEO, or ordinary frontend bug fixes unless the user asks for visual direction. |
Arm Design
Use this skill for high-fidelity visual deliverables made primarily in HTML.
Always
- Identify the artifact type: prototype, deck, motion/video, infographic, variant exploration, or critique.
- If the task names a current product, brand, person, event, version, or technical spec, verify facts before designing.
- If the task involves a brand or product, create or update
product-facts.md and brand-spec.md before high-fidelity work.
- For concept-only products invented by the user, use prompt-sourced
product-facts.md and brand-spec.md; browse only when real-world grounding is requested.
- Use artifact-scoped fact/spec filenames when the workspace contains unrelated design tests.
- Build HTML-first unless the user explicitly requests another format.
- Use this workflow before
imagegen; call imagegen only when a bespoke raster asset materially improves the artifact.
- Run visual verification before final delivery whenever the environment allows it.
- Record assumptions only when they affect the artifact.
- Use design rationale, not private reasoning traces.
Do Not Use This Skill For
- Backend implementation.
- Production SEO work.
- Ordinary frontend bug fixes.
- Generic website implementation where the user did not ask for visual direction or a design artifact.
- Pure copywriting with no visual output.
Interaction Modes
Default to Execute Mode. Make reasonable, reversible assumptions and proceed.
Use Interview Mode only when the user explicitly wants to choose direction first, or when the prompt is too vague to create a useful first draft.
Use Checkpoint Mode for decks of five or more pages, branded launch animations, client-facing visual systems, or other high-risk deliverables. Create a small style proof before scaling.
Load References By Task
- Ambiguous request:
references/router.md
- Brand or product work:
references/fact-and-asset-protocol.md
- Vague visual direction:
references/design-direction-consultant.md
- App or mobile prototype:
references/app-prototypes.md
- Slide deck:
references/slide-decks.md
- Animation or video:
references/motion-video.md
- Review or critique:
references/critique.md
- Final validation:
references/verification.md
- Any design artifact: apply
references/anti-ai-slop.md
Done When
Prototype: source HTML opens locally, all requested screens exist, key interactions work, a verification screenshot exists, and console errors are absent or documented.
Deck: HTML deck works in browser, a representative style proof exists before bulk production for decks of five or more pages, requested exports are produced, and readability is checked at target viewport.
Animation/video: timeline plan exists, HTML animation runs locally, requested video/GIF/audio exports succeed or limitations are documented, and final file paths are listed.
Critique: scores or judgments are evidence-backed, fixes are prioritized, and quick wins are actionable.
