AgentPack Docs
AgentPack is a local context engine for AI coding agents. It ranks relevant repository files and builds compact task-focused context packs for Claude Code, Codex, Cursor, Windsurf, Antigravity, MCP tools, CI jobs, and markdown-based LLM workflows.
Use these docs when you want local/offline repo analysis, MCP-first routing, CI-friendly context packs, and benchmarkable file-selection quality without hosted indexing or embeddings.
Get started
- Commands: CLI reference and common workflows.
- Configuration: config, scoring weights,
.agentignore, and git integration. - How AgentPack works: route, pack, retrieve, learn, and benchmark flow.
- Demo assets: generated README GIF/MP4 and regeneration command.
Core onboarding uses one three-command project loop:
agentpack work "describe the change"
agentpack finish
agentpack doctor
Run agentpack learn --json after delivery when you want evidence-backed
learning topics. route, pack, benchmark, and other specialized commands
remain available for deeper context, release, learning, and diagnostics.
Agents and IDEs
- Integrations: setup paths for Claude Code, Codex, Cursor, Windsurf, Antigravity, and generic agents.
- Agent and IDE plugins: thin plugin/rule distribution layer for Codex, Cursor, Windsurf, Copilot, Cline, Kiro, OpenCode, and more.
- Codex plugin: packaged Codex plugin skeleton and
$agentpack-*skills. - Claude Code context engine: Claude Code setup and MCP-first context.
- Cursor context packing: Cursor setup and context workflows.
- MCP context engine: MCP tools for fresh task context.
- AgentPack for AI agents: short guide for agent maintainers.
Guides
- AI coding agent context packing: why ranked task context helps agent workflows.
- Reduce Claude Code token usage: token-focused usage guide.
- Agent behavior before and after: concrete cold-start examples.
Evidence
- Benchmarking: quality bar, release gate, sample fixtures, and public artifacts.
- Benchmark learnings: current tuning decisions and known bottlenecks.