What is Agents Pack?

Agents Pack gives Claude Code, Codex, and Cursor a shared set of project instructions, skills, and subagents. You choose the agents and components once; Agents Pack renders the right native files for each provider and keeps track of what it owns.

It is a local-first CLI and versioned content pack. There is no account, hosted control plane, or background process.

The product is open source: github.com/farfarawaylabs/agentspackai.

The problem it solves

Coding agents become more useful after you teach them how you work. Each agent expects instructions, skills, and subagents in different places and formats.

Without a shared lifecycle, you typically have to:

• duplicate the same capability across provider-specific files;
• learn several configuration systems;
• keep copies synchronized by hand; and
• decide which generated files are safe to edit.

Agents Pack keeps one maintained content pack and adapts it for Claude Code, Codex, and Cursor.

What it manages

Agents Pack manages three kinds of content:

• Instructions — durable rules that should guide normal project work.
• Skills — focused workflows such as debugging, API development, plan review, frontend design, documentation refresh, or security auditing.
• Subagents — specialized roles such as code reviewer, backend developer, UX enhancer, or researcher.

You choose:

1. where the content should apply (repository or global scope);
2. which coding agents you use; and
3. which components you want.

Agents Pack then renders provider-native files and records exactly what it wrote so future updates can be previewed, validated, applied, or rolled back safely.

For ownership rules — what is official, what is user-owned, and what is generated — see Ownership.

How you use it

You normally interact with Agents Pack in two ways:

1. Ask your coding agent. After initialization, describe what you want in normal language. The installed management skill guides the agent through the correct CLI operations. This is the recommended day-to-day workflow.
2. Run the CLI yourself. Useful when you want exact control, are writing automation, or are troubleshooting.

The first initialization is the exception: you run agents-pack init directly because the management skill is installed by that command.

CLI and content pack are separate

The agents-pack executable and the official content pack are versioned independently. Reinstalling or upgrading the CLI does not change the instructions, skills, or subagents in an initialized scope. Content updates use agents-pack update.

Next steps

1. Install the CLI
2. Initialize a project
3. Try first workflows

Browse the skill catalog when you want to see what the core pack includes.