Design Engineering Skills
An AI partner for product designers working in code: turn client intent into a runnable prototype, refine the experience, and build something people can actually use.
Pairs with the Design Engineering guide. Use a skill before work or to revisit what already exists — no required sequence, no need to load the whole pack for one task.
Start with your work
Use $prototype to build a client scheduling demo in this existing app.
Use $craft to improve this screen, then inspect the rendered result.
Use $slice to connect this workflow to real data and server permissions.
Use $readiness to assess whether this preview can support a real pilot.
$intent helps when the goal itself is unclear. An authorized build request should produce working artifacts, not end with another plan.
Skills
Product definition — frame the goal, build the prototype, pin the behavior:
| Skill | Use it to |
|---|---|
$intent | Clarify the client goal or revisit the original promise. |
$prototype | Build and refine a runnable prototype with honest simulations. |
$requirements | Turn a brief or feature into testable behavior. |
Project setup — foundation, stack, collaboration, and agent instructions:
| Skill | Use it to |
|---|---|
$stack | Compare technical paths when the existing choice is unresolved. |
$setup | Establish or improve a runnable repository foundation. |
$git | Coordinate reviewable changes and shared-file ownership. |
$context | Keep verified agent instructions concise and current. |
UI composition — tokens, structure, system health, and overall craft:
| Skill | Use it to |
|---|---|
$tokens | Create or extend the smallest useful visual vocabulary. |
$layout | Define or revisit regions, widths, scroll, and responsive structure. |
$system | Maintain shared UI decisions and plan consumer migrations. |
$craft | Compose, build, improve, or audit the visible interface. |
IA and flows — where things live and how tasks run:
| Skill | Use it to |
|---|---|
$sitemap | Organize routes, navigation, URL state, and access boundaries. |
$flow | Trace a task through decisions, interruptions, and recovery. |
$hierarchy | Rank one page's content and actions. |
Build and iterate — components, states, and one real workflow:
| Skill | Use it to |
|---|---|
$component | Define or revise a component's API and behavior. |
$states | Find and resolve missing or misleading states. |
$slice | Implement one real workflow through UI, data, permissions, and recovery. |
QA and delivery — the go/no-go question:
| Skill | Use it to |
|---|---|
$readiness | Assess intended real use with evidence, blockers, and unknowns. |
$craft covers the whole visible experience; $layout, $tokens, and $component suit narrower decisions. Each skill's Boundaries section says what it does not own.
Install
npx skills add mattjmdesign/dengskills
For a local checkout:
npx skills add /path/to/dengskills
One skill only: npx skills add mattjmdesign/dengskills --skill craft. Choose agents and project/global scope in the installer; use one installation method to avoid duplicate discovery. Invocation syntax depends on the client — the examples use $skill; otherwise ask for the skill by name.
Claude Code and Codex plugin installs:
claude plugin marketplace add mattjmdesign/dengskills
claude plugin install dengskills@dengskills
codex plugin marketplace add mattjmdesign/dengskills
codex plugin add dengskills
The portable plugin.json (Agent Plugins 1.0) also loads directly in ChatGPT, Cursor, GitHub Copilot, Kiro, and VS Code — no repackaging. Claude plugin skill names are namespaced by the pack.
OpenCode (including the desktop app) uses skills rather than plugin manifests — install the skills for it:
npx skills add mattjmdesign/dengskills --agent opencode
Maintain
bash scripts/check-sync.sh— metadata, groupings, evals, manifests.python3 scripts/audit-security.py— static security scan.bash scripts/bump-version.sh major|minor|patch— version all manifests together.docs/SKILL-AUTHORING.md— rules for changing skills.CHANGELOG.md— release history.
A mock is not a real integration, an automated check is not complete accessibility evaluation, and a readiness review does not authorize deployment or client messages.