AgentKit Recommended for new setups ak CLI

What Is AgentKit?

A beginner mental model: AgentKit is the next generation of ClaudeKit — kit-based workflows, more runtimes, the ak CLI, and Desktop App.

Context

ClaudeKit is now AgentKit

Both a rebrand and an expansion. Kits, Skills, Agents, Hooks, and structured workflows stay central — with a wider product surface.

New native CLI

ak replaces ck for the new line

More runtimes

Via adapters — no parity promise

Desktop App

Local visual control center

Why it exists

What problem does it solve?

You want verified AI workflows installed into the coding assistant you already use, then run tasks with that runtime’s models, tools, and permissions — while you stay responsible for reviewing plans, diffs, and external actions.

Product surface

Four main surfaces

Each surface has a job. Desktop does not replace the terminal; the CLI does not replace your coding assistant.

CLI

AgentKit CLI (ak)

Authenticates, resolves, verifies, installs, updates, diagnoses, and removes AgentKit-owned content.

KIT

Kits

Versioned packages of Skills, Agents, Hooks, and supporting artifacts — the unit you install.

ADP

Runtime adapters

Project Kit content into the structure a coding agent understands.

APP

Desktop App

Native UI for setup, Kit browsing, project registration, and local activity.

Critical boundary

Install a Kit ≠ run a workflow

Install a Kit

AgentKit installs and adapts content. It does not start a task.

  • CLI / Desktop own lifecycle
  • Adapter projects into the runtime
  • You choose project or global scope

Invoke a Skill

The coding assistant runs the AI session with its own model, tools, and permissions.

  • Syntax varies (/ak:…, $ak:…)
  • Hooks / events differ by runtime
  • You review the result

Runtimes

Supported runtime model

Paths

Where should you start?

Continue learning