APM – Agent Package Manager
An open-source, community-driven dependency manager for AI agents.
Think package.json, requirements.txt, or Cargo.toml — but for AI agent configuration.
GitHub Copilot | Claude Code | Grok Build | Cursor | OpenCode | Codex | Gemini | Windsurf | Kiro
Documentation · Quick Start · CLI Reference · v0.29 Roadmap
Portable by manifest. Secure by default. Governed by policy.
One file describes every agent's context; one command reproduces it everywhere; one policy controls what an org will allow.
Why APM
AI coding agents need context to be useful — standards, prompts, skills, plugins — but today every developer sets this up manually. Nothing is portable nor reproducible. There's no manifest for it.
APM fixes this. Declare your project's agentic dependencies once in apm.yml, and every developer who clones your repo gets a fully configured agent setup in seconds — with transitive dependency resolution, just like npm or pip. It's also the first tool that lets you author plugins with a real dependency manager and export standard plugin.json packages.
# apm.yml — ships with your project
name: your-project
version: 1.0.0
dependencies:
apm:
# Skills from any repository
- anthropics/skills/skills/frontend-design
# Plugins
- github/awesome-copilot/plugins/context-engineering
# Specific agent primitives from any repository
- github/awesome-copilot/agents/api-architect.agent.md
# A full APM package with instructions, skills, prompts, hooks...
- microsoft/apm-sample-package#v1.0.0
mcp:
# MCP servers -- installed into every detected client
- name: io.github.github/github-mcp-server
transport: http # MCP transport name, not URL scheme -- connects over HTTPS
git clone <org/repo> && cd <repo>
apm install # every agent is configured
Coming from npx skills add? Drop-in:
apm install vercel-labs/agent-skills # whole bundle, like npx skills add
apm install vercel-labs/agent-skills --skill deploy-to-vercel # one skill, persisted to apm.yml
Same install gesture. You also get a manifest, lockfile, and reproducibility.
Zero-config Copilot:
apm compile -t copilot # writes .github/copilot-instructions.md
One command, no configuration -- VS Code and GitHub Copilot read the file automatically. APM dogfoods this target on its own repository.
The three promises
1. Portable by manifest
One apm.yml describes every primitive your agents need — instructions, skills, prompts, agents, hooks, plugins, MCP servers — and apm install reproduces the exact same setup across every client on every machine. apm.lock.yaml pins the resolved tree the way package-lock.json does for npm.
- One manifest for everything -- declared once, deployed across Copilot, Claude, Grok Build, Cursor, OpenCode, Codex, Gemini, Windsurf, Kiro
- Install from anywhere — GitHub, GitLab, Bitbucket, Azure DevOps, GitHub Enterprise, Gitea, Gogs, any git host
- Transitive dependencies — packages can depend on packages; APM resolves the full tree
- Author plugins — build Copilot, Claude, and Cursor plugins with dependency management, then export standard
plugin.json
- Marketplaces — install plugins from curated registries in one command, deployed across all targets and locked
- Pack & distribute —
apm pack bundles your configuration as a zipped package or a standalone plugin
- CI/CD ready — GitHub Action for automated workflows
2. Secure by default
Agent context is executable in effect — a prompt is a program for an LLM. APM treats it that way. Every install scans for hidden Unicode that can hijack agent behavior; the lockfile pins integrity hashes; transitive MCP servers are gated by trust prompts.
- Content security —
apm install blocks compromised packages before agents read them; apm audit runs the same checks on demand
- Lockfile integrity —
apm.lock records resolved sources and content hashes for full provenance
- SBOM export —
apm lock export --format cyclonedx|spdx emits a standard inventory of what reached disk, straight from the lockfile — provenance for procurement, not a compliance attestation
- Drift detection —
apm audit rebuilds your agent context in scratch and diffs it against your working tree to catch hand-edits before they ship
- MCP trust boundaries — transitive MCP servers require explicit consent
3. Governed by policy
apm-policy.yml lets a security team say "these are the only sources, scopes, and primitives this org will allow" and have every apm install enforce it — with tighten-only inheritance from enterprise to org to repo, a published bypass contract, and audit-mode CI gates.
apm-policy.yml governs what gets installed; your agent harness governs what runs. The two planes do not overlap.
Get Started
macOS with Homebrew
brew install apm
No custom tap is required. Homebrew manages installation and updates:
use brew upgrade apm, not apm self-update.
Linux / macOS without Homebrew
curl -sSL https://aka.ms/apm-unix | sh
Windows
irm https://aka.ms/apm-windows | iex
Native release binaries are published for macOS, Linux, and Windows x86_64.
For standalone installations, apm self-update reuses the matching platform installer.
You do not need Homebrew to use APM.
Other install methods: pip, WinGet, and Scoop
Linux / macOS
# pip (Python 3.10+)
pip install apm-cli
Windows
# WinGet
winget install --id Microsoft.APM --exact --source winget
# Scoop
scoop bucket add apm https://github.com/microsoft/scoop-apm
scoop install apm
# pip
pip install apm-cli
Already using the Microsoft Homebrew tap? See the
migration guide.
Then start adding packages:
apm install microsoft/apm-sample-package#v1.0.0
Or install from a marketplace:
apm marketplace add github/awesome-copilot
apm install azure-cloud-development@awesome-copilot
Or add an MCP server (wired into Copilot, Claude, Cursor, Codex, OpenCode, Gemini, Windsurf, and Kiro):
apm install --mcp io.github.github/github-mcp-server --transport http # connects over HTTPS
See the Getting Started guide for the full walkthrough.
Works with agentrc
agentrc analyzes your codebase and generates tailored agent instructions — architecture, conventions, build commands — from real code, not templates.
Use agentrc to author high-quality instructions, then package them with APM to share across your org. The .instructions.md format is shared by both tools — no conversion needed when moving instructions into APM packages.
Created by @danielmeppiel. Maintained by @danielmeppiel and @sergio-sisternes-epam.
Built on open standards: AGENTS.md · Agent Skills · MCP
Trademarks
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.
APM（Agent Package Manager）是一款开源的 AI 编程代理上下文依赖管理器——就像 package.json 或 Cargo.toml，但管理的是指令、技能、提示词和 MCP 服务器等代理配置。