agentpack

The package manager for AI coding agents

Pin, resolve, and sync skills & plugins across Claude Code, Cursor, OpenCode, and Codex — from one manifest.

---

The Problem

Every AI coding agent has its own way of managing skills, plugins, commands, and rules — different file layouts, different config formats, different discovery mechanisms. If you use more than one agent (or want your team to share a curated set of tools), you're stuck copying files around manually.

The Solution

agentpack gives you one manifest (agentpack.toml) that declares your agent dependencies. It resolves versions, caches content, converts artifacts per-harness, and launches each agent with the right configuration — automatically.

agentpack.toml  ──>  pack.lock  ──>  staged bundles  ──>  launch
   (you write)        (pinned)        (per-harness)        (agent runs)

Quick Start

# Install via Homebrew
brew tap OlegHQ/agentpack
brew install agentpack

# Or from source
cargo install --path .

# Initialize in your project
agentpack init

# Add a skill from GitHub
agentpack add anthropics/skills/skills/canvas-design

# Launch your agent with everything bundled
agentpack claude    # Claude Code
agentpack agent     # Cursor Agent
agentpack opencode  # OpenCode
agentpack codex     # Codex

How It Works

1. agentpack.toml — Declare dependencies (GitHub repos, subdirectories, or local paths) with version constraints.
2. pack.lock — Deterministic lockfile pins every package to an exact commit + content hash.
3. sync — Downloads, caches, and converts artifacts into per-harness staging directories.
4. Launchers — agentpack claude, agent, opencode, codex — each starts the target agent with the staged bundle injected via the agent's native extension mechanism (plugin dirs, config roots, or HOME overrides).

Cross-Harness Artifact Conversion

agentpack doesn't just copy files — it converts between formats:

Source artifact	Claude Code	Cursor	OpenCode	Codex
Commands	Claude frontmatter	Plain markdown	OF frontmatter	Skill fallback
Agents	Claude agent MD	Cursor agent MD	OF agent MD	Skill fallback
Skills	Normalized skill	Normalized skill	Normalized skill	Codex skill
Rules	Skill fallback	.mdc preserved	Skill fallback	Skill fallback

Manifest Example

name = "my-project"
version = "0.1.0"

[dependencies]
"github.com/anthropics/skills/skills/canvas-design" = { branch = "main" }
"github.com/anthropics/claude-plugins-official/plugins/hookify" = { version = "^1.0.0" }
"github.com/my-org/internal-tools/skills/deploy" = { tag = "v2.1.0" }

[overrides."github.com/someorg/heavy-pack"]
disable = ["commands/noise.md", "hooks"]

Key Features

• One manifest, four agents — Write agentpack.toml once, launch Claude, Cursor, OpenCode, or Codex with the same skill set.
• Deterministic lockfile — pack.lock pins exact commits and content hashes. Reproducible across machines.
• Transitive resolution — Dependencies can declare their own agentpack.toml; agentpack resolves the full tree.
• Content-addressed cache — Downloaded packages are stored once in $AGENTPACK_HOME/cache/ and shared across projects.
• No workspace pollution — Pack content lives in staging directories, not in your git repo. Nothing is materialized into .cursor/ or .claude/.
• Offline-first — Local mirrors ($AGENTPACK_HOME/local/) and cached metadata reduce network calls. Git protocol fallback when REST API is throttled.
• Fast launch path — Launchers skip full re-sync when inputs haven't changed since the last successful run.
• Selective overrides — Disable specific commands, hooks, or directories per-package via [overrides].
• Project-local agent config — Optional .agents/ directory (dot-agents style) merges project-specific rules, skills, and agents into staged bundles.

Comparison

Feature	agentpack	Manual copy	Symlinks	Git submodules
Multi-agent support	Claude, Cursor, OpenCode, Codex	One at a time	One at a time	One at a time
Version pinning	Semver, branch, tag, commit	None	None	Commit only
Transitive deps	Yes	No	No	Manual
Artifact conversion	Automatic per-harness	Manual	No	No
Cache / dedup	Content-addressed	None	N/A	Partial
Workspace isolation	Full (staging dirs)	Files in repo	Symlinks in repo	Dirs in repo
Lockfile	Deterministic	None	None	.gitmodules

Commands

Command	Description
init	Create agentpack.toml and pack.lock in the current project
add <spec>	Add a dependency, resolve, lock, and sync
remove <spec>	Remove a dependency and re-sync
lock	Resolve manifest and regenerate pack.lock
sync	Download, cache, and rebuild staging for all harnesses
claude	Sync + launch Claude Code with --plugin-dir
agent	Sync + launch Cursor Agent with staged HOME
opencode	Sync + launch OpenCode with staged config root
codex	Sync + launch Codex with staged CODEX_HOME

Prerequisites

• Rust toolchain (edition 2021), or Homebrew rust when installing via brew
• GitHub CLI for release automation (optional)

Documentation

See AGENTS.md for full technical details: environment variables, data layout, staging internals, harness-specific behavior, and the lockfile format.

License

MIT — see LICENSE.