# AI Agent Context & Skills Reference Load this when configuring AI agents for a project. --- ## Context Files Agents may create a `.ai/` folder to store project context: | File | Purpose | Format | |------|---------|--------| | `context.yaml` | Structured project metadata (stack, conventions, skills) | YAML | | `IDENTITY.md` | Brand/design identity | Markdown | | `CLAUDE.md` | Human-readable project overview (root level) | Markdown | **`context.yaml`** is the canonical config. Agents should: - Create it if missing - Update it when project context changes - Reference it for all project decisions --- ## Example context.yaml ```yaml project: name: MyApp description: Personal finance tracker stack: runtime: bun frontend: sveltekit backend: elysia database: neon orm: drizzle auth: clerk styling: tailwind + plain css conventions: - Use Svelte 5 runes ($state, $derived), not legacy syntax - Tailwind for layout, plain CSS for animations - API routes via Elysia, not SvelteKit endpoints - Drizzle for all database queries skills: - name: ui-ux-pro-max installed: true brand: identity_file: .ai/IDENTITY.md ``` > **Tip:** Keep `context.yaml` under 50 lines if possible. It's config, not documentation. --- ## Skills Skills are installable modules that give AI agents specialized capabilities. **Install pattern:** ```bash bunx init --ai claude # Bun npx init --ai claude # Node ``` **Suggested skills:** | Skill | Use Case | Install | |-------|----------|---------| | **UI UX Pro Max** | Design systems, palettes, typography | `bunx uipro-cli init --ai claude` | | **ClaudeKit Skills** | Comprehensive skill collection | See below | --- ## ClaudeKit Skills A curated collection of 50+ specialized skills for Claude Code agents. **Installation:** ```bash # Add marketplace /plugin marketplace add mrgoonie/claudekit-skills # Install skill categories as needed /plugin install ai-ml-tools@claudekit-skills /plugin install web-dev-tools@claudekit-skills /plugin install devops-tools@claudekit-skills ``` **Categories:** | Category | Contents | |----------|----------| | **ai-ml-tools** | Gemini API, context engineering, multimodal | | **web-dev-tools** | React, Next.js, Tailwind, shadcn/ui, Three.js | | **devops-tools** | Cloudflare, Docker, GCP, databases | | **backend-tools** | Node.js, Python, Go, authentication | | **debugging-tools** | Systematic debugging, root-cause tracing | | **problem-solving-tools** | Advanced thinking techniques | | **document-processing** | Word, PDF, PowerPoint, Excel | **When to suggest:** - **Web projects** → `web-dev-tools`, `ui-styling` - **Backend APIs** → `backend-tools`, `devops` - **AI/ML tasks** → `ai-ml-tools` - **Debugging** → `debugging-tools` - **Complex problems** → `problem-solving-tools` **Repository:** https://github.com/mrgoonie/claudekit-skills --- ## MCP Servers (Optional) MCP servers extend AI agent capabilities beyond the project. Unlike skills (which install files), MCP servers run in your IDE environment. | Server | Use Case | Setup | |--------|----------|-------| | **Sequential Thinking** | Improves reasoning quality | See MCP docs | | **Magic MCP** | Generate UI components from 21st.dev | `npx @21st-dev/cli@latest install ` | | **shadcn MCP** | Browse/install shadcn/ui components | `pnpm dlx shadcn@latest mcp init` | | **Framelink (Figma)** | Fetch Figma designs | See Framelink docs | **Key differences:** - **MCP servers** run as background processes in IDE (per-developer) - **Skills** install static files into project (shared across team) **Agent behavior:** - Agents can **suggest** MCP servers when relevant - Agents **cannot install** MCP servers automatically (requires IDE config)