Install
$ agentstack add skill-dndungu-agent-skills-web-design ✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
About
web-design — mockups in the framework you ship
Same workflow as /ios-design, different medium: direction -> tokens -> component gallery -> screens, iterated on real rendered screenshots. Mockups are authored in the TARGET framework (React/Next + Tailwind, Vue, Svelte, or plain HTML) so the mockup language IS the implementation language — the same-language rule from /design-fidelity-loop.
Why not just use Claude Design: it renders HTML .dc.html artifacts inside its own runtime. That is excellent for standalone deliverables and decks, and this skill borrows its verify loop wholesale — but a .dc.html is not your Next.js app, so shipping it still costs a translation. Route to Claude Design (via DesignSync) when the deliverable IS a standalone page or deck; use this skill when the deliverable becomes product code.
Read WEB.md before Phase 1 — the web delta to the two canons. Read RENDERER.md before the first render — capture harness and gotchas. Read ~/.claude/skills/ios-design/VERIFY.md before showing any render — the fresh-eyes loop is medium-agnostic and applies unchanged.
The rubric (three files, one hierarchy)
ios-design/TASTE.md— beauty laws (15), medium-agnostic.ux-design/CANON.md— experience laws (18), medium-agnostic.WEB.md— the delta: what changes, what is web-only, the click budget.
On craft disagreements the Vercel web-interface-guidelines skill wins; on aesthetics TASTE.md wins. Record any deliberate law break in DESIGN.md.
Phase 0 — Read first
- Resolve the repo. Detect the stack before proposing anything: framework
(Next/React, Vue, Svelte, plain), styling (Tailwind config, CSS vars, CSS-in-JS), component library (shadcn/ui, Radix, headless), and whether Playwright already exists (most repos with e2e tests have it — reuse it).
- Inventory existing design assets:
tailwind.config.*,globals.css/
theme CSS vars, docs/design/, a component directory, brand assets. An existing token system is EXTENDED, never replaced.
- Inventory experience artifacts:
EXPERIENCE.md/journeys.yamlfrom
/ux-design, or a journeys doc. If they exist, screens come FROM flows — every screen names its flow node. If the user wants flows first, route to /ux-design before Phase 1.
- If the
web-interface-guidelinesskill is installed, say so — it is the
craft authority for Phase 5 handoff.
Phase 1 — Direction (read WEB.md now)
- Interview ( fresh-eyes subagent verdict against the user's asks
verbatim + WEB.md laws -> act. Only a VERDICT: done set reaches the user.
Phase 5 — Iterate, then hand off
- Show the contact sheet. Apply feedback at the right altitude: color /
spacing / type -> tokens; behavior -> component; screen-local fixes are rare. Re-render with a bumped iteration prefix; never overwrite history.
- User approval is the only exit. Then persist
docs/design//
with DESIGN.md (POV, signature detail, 3 checkable laws, the motion spec incl. the reduced-motion variant, decision log, any recorded law breaks), the gallery route, and renders/.
- Hand off: /web-craft for production hardening (and the Vercel
web-interface-guidelines skill for the 100+ implementation rules); /ui-review + /ux-review (web lanes) to audit the built result.
Output format
- Renders named
r---.pnginrenders/,
plus an index.html contact sheet. Schemes always come in pairs.
- Progress messages show screenshots, not code.
Does not do
- Production hardening, a11y/perf audits, framework idioms — /web-craft plus
the Vercel guidelines skill.
- Journeys, flows, navigation architecture — /ux-design (upstream).
- Auditing an app that already exists — /ui-review (pixels) and
/ux-review (experience), both of which have web lanes.
- Native iOS design — /ios-design. Standalone HTML pages, decks, or design
systems browsable at claude.ai/design — Claude Design via DesignSync.
- Theme A/B on an existing component kit — /skin-gallery.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: dndungu
- Source: dndungu/agent-skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.