Atomic packages for AI workflows.
One pack, compiled to every platform you use — with governance built in.
AgentPack is an open packaging standard for AI agent behavior. One AGENTPACK.yaml compiles to Claude Code, Codex, Cursor, ChatGPT Apps, and a generic AGENTS.md target that agents like Google Antigravity read — with permissions and risk visible before you install.
# Not on npm yet — clone & build once: # git clone https://github.com/jckeen/agent-pack # cd agent-pack && pnpm install && pnpm build # Install a pack straight from GitHub # (no registry or account needed) agentpack install \ github:jckeen/agent-pack@master#examples/pr-quality # Import your existing Claude Code setup agentpack import --from claude-code ~/.claude \ --id you.your-pack --out my-pack # Compile it for any target agentpack pack export my-pack \ --target codex --out dist/codex
Featured packs
Curated AgentPacks across editor, review, and team-workflow surfaces.
Pull Request Quality Pack
Cross-platform PR review, security review, formatting, and summary workflow.
Claude Code Starter Pack
Starter project instructions, skills, hooks, and settings for Claude Code.
Codex AGENTS.md Starter Pack
Starter AGENTS.md, Codex config, skills, and hooks.
Cursor Rules Starter Pack
Project rules, frontend/testing standards, and MCP config for Cursor.
Newsroom Editorial Workflow Pack
Editorial standards, fact-checking workflow, headline/social command, and human approval rules.
Grant Research Workflow Pack
Prospect research, fit scoring, LOI drafting, budget narrative, and review checklist.
Compile, don't replace
Targets existing platform standards: CLAUDE.md, AGENTS.md, .cursor/rules, MCP. AgentPack is the layer above, not another silo.
Permission transparency
Every pack lists what it can read, write, run, and connect to — broken down by install profile, before you say yes.
Built for supply chain
Atoms, lockfiles, checksums, and review states. Treat agent customization like every other dependency in your repo.