Skip to content

Initialize a project

Initialization is the one step you run yourself before day-to-day management through a coding agent. From the project you want to configure:

cd /path/to/your-project
agents-pack init

Agents Pack asks you to choose:

  • repository or global scope;
  • Claude Code, Codex, Cursor, or a supported combination; and
  • Recommended, All, or a custom component selection.

It shows the exact selected components and filesystem plan before applying anything. The confirmation defaults to No; select Yes to apply the plan.

Use the arrow keys to move through each menu. Press Space to toggle agents or components, and Enter to confirm the visible selection. Repository scope, all compatible agents, and recommended components are selected by default.

With CLI 0.3.0, choosing Recommended opens an additional menu for optional categories and individual components. Nothing optional is preselected; press Enter to keep only the Recommended set. A category can include skills, subagents, and instructions. Select a category with Space to toggle its components together, or move to an individual item to toggle just that component.

For the complete design workflow, add Design / Creative and Design / Critique. Also select ap-frontend-design and ap-frontend-review under Engineering / Frontend; they are optional too.

To install the recommended repository components for all three agents:

agents-pack init \
--scope repository \
--agents claude,codex,cursor \
--components recommended \
--yes

--agents all is shorthand for --agents claude,codex,cursor.

For a global Claude Code and Codex installation:

agents-pack init \
--scope global \
--agents claude,codex \
--components recommended \
--yes

To apply an initialization from a script or another non-interactive shell, provide the scope, agents, component choice, and --yes. Use --dry-run instead of --yes when the script should only print the plan.

Content-changing commands support a dry run:

agents-pack init \
--scope repository \
--agents claude,codex,cursor \
--components recommended \
--dry-run

A dry run loads and validates the pack and prints the same plan, but does not write provider files, installation state, transaction state, or the pack cache.

The core pack labels components as required, recommended, or optional.

Choice What it installs
recommended Every required and currently recommended component. Best starting point for most users.
all Every compatible component in the current pack version. One-time expansion, not a subscription — later optional components are not silently installed.
Explicit IDs A tailored list such as ap-debug,ap-code-reviewer,ap-frontend-design. Required components are added automatically.

Choose Custom to select categories or individual components from the full compatible catalog, with recommended components preselected. Required components are always included. Your final component IDs are stored explicitly. Passing --components recommended, all, or explicit IDs skips the interactive component menus.

See Component selection and the skill catalog for more detail.

init normally uses the official registry. For development, private content, or air-gapped work, pass a directory:

agents-pack init \
--scope repository \
--agents claude,codex,cursor \
--components recommended \
--pack /path/to/local-pack \
--yes

The installation records whether its source is official or local. A local installation never silently switches to the public registry.

Coding agents discover the installed instructions, skills, and subagents through their native mechanisms. Check health with:

agents-pack status

Before your first real task, set up the opinionated workflow: a PRD.md, TECHNICAL_REQUIREMENTS.md, and TODOs.md at the repository root that the installed core instructions read automatically at the start of every session.

Then try first workflows — usually by asking your coding agent what you want next.

For every flag and option, see the CLI reference.