Motion, craft and
AI experiments.

Skills & agents

Open-source tools I build so AI works the way designers do: MCP servers, skills, agent systems and rules for Claude and Cursor.

01
mcp-design-system

A plugin that builds an MCP server out of your design system, extracting tokens and components straight from Figma so AI code agents ship UI that is consistent by construction.

MCP server ✳ demo
02
skill-ai-execution-contract

A structured execution-contract workflow for AI products: turns fuzzy discovery into clear definitions an agent can actually execute against.

Skill ✳ demo
03
lib-microinteractions

A motion skill for Claude and Cursor: an agent-ready microinteraction pattern library (Dan Saffer framework) that analyzes static screens and proposes transitions grounded in real motion knowledge.

Library ✳ demo
04
agents-spec-orchestrator

A spec-driven delivery orchestrator: 46 coordinated agents taking software from spec to shipped, with templates for Cursor and Claude.

Agent system
05
agents-ui-craft

Spec-driven agents for UI craft — a constitution, quality gates and craft foundations that hold AI-built interfaces to a designer's bar.

Agent system
06
skill-figma-annotations

Documents UX and business rules as native Figma annotations via MCP — the file becomes its own spec.

Skill
07
skill-figma-semantic-organization

Cleans Figma files semantically — renaming, grouping and applying Auto Layout — so both humans and agents can navigate them.

Skill
08
agentic-experience-handoff

Handoff for agentic experiences: documentation, templates and evaluation criteria for products where the interface is an agent.

Skill
09
rules-design-bible

The Design Bible as Cursor rules: absolute UI/UX principles every AI-generated screen must respect.

Rules
10
rules-figma

Figma best practices encoded for AI agents: Auto Layout, constraints, variables and tokens.

Rules
11
rules-accessibility-handoff

An accessibility handoff annotation system for Figma, written as rules agents can apply.

Rules