Status: Archived as of January 2025
Reason: This project has been superseded by modern alternatives like MCP (Model Context Protocol) and LSP-based tooling.
Value: This repo remains as a reference implementation for VS Code extension architecture, multi-adapter patterns, and Tree Provider APIs.
Original Description:
Forked from Copilot DocsSync.
Download framework documentation and activate AI agent capabilities through context packs.
This project was designed to help local AI models (TinyLlama, Phi-3, Qwen2.5) handle complex coding tasks by providing pre-digested, condensed documentation that fits within their limited context windows (4K-8K tokens). While modern cloud models have made this less critical, the architecture patterns here remain relevant for:
- VS Code extension development
- Multi-adapter system design
- Tree Data Provider implementations
- Webview integration patterns
A context pack is a self-contained knowledge bundle that:
- Downloads framework documentation offline
- Generates instructions that activate multiple agent capabilities
- Works with any AI agent (OpenCode, Claude Code, GitHub Copilot)
npm install
npm run compile
# Press F5 to test in VS Code Extension Development Hostsrc/
extension.ts - Entry point, command registration
codebaseAnalyzer.ts - Detect frameworks and patterns
smartScanner.ts - Intelligent documentation fetching
contextPackService.ts - Orchestrate documentation sync
scraperService.ts - Web scraping (axios, cheerio)
cacheService.ts - Caching with 7-day TTL
instructionsGenerator.ts - Generate agent instructions
packages/core/llm/ - LLM providers for relevance filtering
llmProvider.interface.ts
vscodeLmProvider.ts - Claude Code, Copilot support
opencodeLmProvider.ts - OpenCode support
relevanceAnalyzer.ts - Filter relevant docs
packages/adapters/ - Agent-specific integrations
opencode.ts
claude-code.ts
github-copilot.ts
Frontend: React, Vue.js, Angular, Next.js, Nuxt, Svelte, Astro, SolidJS, Qwik
Backend: Express, FastAPI, Django, Rails, Hono, NestJS, Fastify
ORM: Prisma, TypeORM, Sequelize, Drizzle, Kysely, SQLAlchemy
Auth: Auth0, Firebase Auth, NextAuth, Passport, Better Auth, Clerk, Lucia
CSS: TailwindCSS, Styled Components, Emotion, UnoCSS
Testing: Jest, Vitest, Mocha, Pytest, Cypress, Playwright, Testing Library
State: Redux, Zustand, MobX, Jotai, Signals
Forms: React Hook Form, Formik, Zod, Yup
UI: Shadcn UI, Radix UI, Headless UI, Mantine, Chakra UI, Ant Design, Vuetify
| Agent | Integration | Priority |
|---|---|---|
| OpenCode | Skill registration, OPENCODE_RUNTIME | 5 |
| Claude Code | .claude/config.json, 200K tokens |
10 |
| GitHub Copilot | .github/copilot-instructions.md |
30 |
Auto-detection chooses the highest priority available agent.
Analyzes your codebase and fetches only relevant documentation:
- Scans for dependencies, imports, patterns
- Uses LLM to score documentation relevance
- Downloads relevant sections only (reduces ~80% of unnecessary docs)
Add custom frameworks that Smart Scan should detect and fetch docs for:
{
"agent-contextpacks.customFrameworks": [
{
"name": "MyFramework",
"docUrl": "https://myframework.dev/docs",
"patterns": ["*.myfw", "myfw.config.js"],
"keywords": ["my-framework", "@myorg/fw"]
}
]
}Detection methods:
keywords- Match in package.json dependenciespatterns- Match file names in workspacekeywords- Match in source file contents
.github/docs/
<framework>/
page1.md
page2.md
...
.cache/
cache.json
agent-instructions.md
{
"agent-contextpacks.defaultOutputDirectory": ".github/docs",
"agent-contextpacks.cacheExpiration": 24,
"agent-contextpacks.maxPagesPerSite": 100
}npm run compile # Build once
npm run watch # Auto-recompile on changes
npm run lint # Run ESLint
npm test # Run tests (64 tests)MIT License. See LICENSE file.