agent-skill-loader

# Architecture & Context 🏗️ This document provides technical context for AI agents working on this codebase. ## System Overview **Agent Skill Loader** is a Node.js-based MCP Server. It does not maintain a persistent database; it scans the file system in real-time (or near real-time) to discover "Skills". ### Core Definitions * **Skill**: A directory containing a `SKILL.md` file. The `SKILL.md` contains the instructions (system prompt fragment) for the AI. * **Skill Library**: A root directory containing multiple Skill directories. ## Logic Flow (`src/index.ts`) 1. **Initialization**: * The server starts using `StdioServerTransport`. * It loads configuration from `.env`. * It reads `MCP_SKILL_PATHS` to determine where to scan. 2. **Tool: `list_skills`**: * Recursively scans `SEARCH_PATHS`. * Looks for `SKILL.md`. * Parses frontmatter (YAML-style) to extract `description`. * Returns a simplified JSON list to save context tokens. 3. **Tool: `read_skill`**: * Finds the specific skill by name. * Reads `SKILL.md`. * Returns the raw text Content. 4. **Tool: `install_skill`**: * Locates the source directory. * **Security**: Validates that target path is within current workspace. * Uses `fs.cpSync` to recursively copy the content to the User's active workspace. ## Key Design Decisions * **Recursive Scanning**: We scan subdirectories to find skills, allowing for categorized folder structures (e.g., `dax/skills/writing-dax-measures`). * **No Database**: To keep it lightweight and stateless, we scan the FS. Performance is acceptable for <1000 skills. * **TypeScript**: Used for type safety with the MCP SDK. ## Development Guidelines * **Building**: `npm run build` runs `tsc`. * **Modifying Tools**: Add new tools in `src/index.ts` using `server.tool()`. * **Error Handling**: All filesystem operations should be wrapped in try/catch blocks to prevent server crashes.