AgentStack
SKILL verified MIT Self-run

Design

skill-dndungu-agent-skills-design · by dndungu

Design front door — interview first, then route to the right design skill. Use when the user says "/design", "design something", "I want to design an app/screen/feature", "make this beautiful", "which design skill should I use", or gives a design request too vague to route (no clear target, stage, or deliverable). Asks 3-4 multiple-choice questions (what/where/constraints/depth), then invokes the…

No reviews yet
0 installs
0 views
view→install

Install

$ agentstack add skill-dndungu-agent-skills-design

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Security review

✓ Passed

No 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.

Are you the author of Design? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

design — the interview-then-route front door

One job: turn a vague design wish into a precisely-invoked design skill. Never do design work here — this skill only interviews and dispatches.

Step 0 — Read first (silently, before any question)

  1. If the user named a repo, resolve it via ~/.claude/repo-map.md; else

note the cwd repo. Peek for an existing design system (docs/design/*/DESIGN.md, *Theme*.swift, project design skills) — its existence changes the routing default and the questions you ask.

  1. Skip any question whose answer is already in the user's message or in

the repo. Never ask what you can infer; the interview is 3-4 questions MAX, one AskUserQuestion call.

Step 1 — Interview (one AskUserQuestion call, ≤4 questions)

Ask only the questions still open after Step 0, drawn from this bank:

  • Goal (always, unless stated): What are we doing?

a) Design something new — app, feature, or screen (mockups first) b) Improve existing screens — they work but look off c) Choose or iterate a theme/palette on an existing component kit d) Turn approved mockups into production UI with fidelity checks e) Liquid Glass adoption/refactor f) App Store presence — screenshots, listing, review prep g) Review/critique an existing app's design (pixels + source, report) h) Design the experience — journeys, flows, navigation, habit loops i) Review an existing app's UX — taps, flows, dead ends, habit ethics

  • Platform (always, unless obvious from the repo — a SwiftUI repo is

iOS, a Next/Vue repo is web): native iOS, web, or both. This selects the column in the routing table below; goal and platform are orthogonal.

  • Target: which repo/app? Offer the repo-map's likely candidates as

options plus Other.

  • Constraints (multiSelect): existing brand/design system is sacred ·

full visual reset allowed · dark-mode-first · iOS 26+ (Liquid Glass ok) · must ship this week (speed over depth)

  • Depth: quick single pass, or full workflow with direction anchors,

verify loops, and iteration gates?

Step 2 — Route (invoke exactly one skill via the Skill tool)

| Answer | Route (iOS) | Route (web) | Args to pass | |---|---|---|---| | (a) new design | /ios-design | /web-design | repo path + goal + constraint answers ("full reset" vs "extend system", tone hints, depth) | | (b) improve existing | /swiftui-craft | /web-craft | repo path + the screens/views named + "review then apply" or "review only" | | (c) theme A/B | /skin-gallery | /skin-gallery | repo path + kit location | | (d) mockups → production | /design-fidelity | /design-fidelity | mockup dir + target | | (e) Liquid Glass | /liquid-glass | (iOS only) | repo path + adopt vs refactor vs review | | (f) App Store | /app-store | (iOS only) | mode hint: OPTIMIZE / PREPARE / AUDIT | | (g) design review | /ui-review | /ui-review | repo path + screens + PLATFORM + "can it run?" answer | | (h) experience/UX | /ux-design | /ux-design | repo path + hero-job hint + existing journeys doc if any | | (i) UX review | /ux-review | /ux-review | repo path + journeys doc + PLATFORM + "can it run?" answer |

The review skills (g, i) and the experience skills (d, h) are one skill with two lanes — always pass the platform in the args so the routed skill picks the right capture and rubric. Only (e) and (f) are iOS-only; a web request there routes to /web-craft (glass-style effects) or is out of scope.

Compose the args string from the interview so the routed skill can skip its own redundant questions — pass the answers, not a summary of them (e.g. Reimagine the home screen of ~/Code/foo — full palette reset, dark-first, full workflow with anchors).

Edge cases:

  • Goal (a) but a design system exists and "brand is sacred": still

/ios-design or /web-design — both extend existing systems (Phase 0); say so.

  • Web goal (a) where the deliverable is a STANDALONE page, deck, or a design

system to browse rather than product code: that is Claude Design via DesignSync, not /web-design. Say which and why.

  • Goal (b) but the repo has NO design system and screens are beyond

polish: offer the user the (a) path before routing — polishing chaos wastes /swiftui-craft.

  • Mixed goals: route the FIRST deliverable only; name the follow-on skill

in one sentence for after it lands.

Output format

After the routed skill is invoked, this skill's job is done — no summary, no double-reporting; the routed skill owns the conversation from there. If the user aborts the interview, print the routing table above as plain text so they can self-route next time.

Does not do

  • Any actual design, review, rendering, or code — the routed skills own

that. If the user's request was already specific enough to route without questions, invoke the target skill directly and say why (interview skipped) — asking questions you can answer is a defect, not diligence.

  • Cross-repo programs (kit-first fleets) — that is /design-fidelity-loop;

mention it only when the user describes multi-agent scale.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.