docs.alakazam.gg/skill.md.
Once installed, your agent can:
- Generate playable worlds from a premise or an image
- Create talk-to characters (stances, memory, voice)
- Embed worlds correctly (the two-token rule)
- Program a world’s state-machine graph, deterministically or in natural language
- Wire the
/v1endpoints and the@alakazamworld/embedSDK into your app
Prerequisites
-
An Alakazam API key. Create an app at play.alakazam.gg/?view=developer, mint a secret key (start with
sk_test_…: a free, mocked sandbox), and export it in the shell your agent inherits: - A supported agent: Claude Code, Codex, Cursor, Windsurf, OpenCode, Gemini CLI, GitHub Copilot, and 50+ others read skill files.
-
Node.js to run the
npx skillsinstaller (or install manually withcurl).
Installation
Method 1: Skills CLI (one command, every agent)
The skills CLI detects your agent and drops the skill where that agent reads it:-g to install into the current project instead of globally.
Method 2: Manual install with curl
Pick your client and run the matching command. Use the project path for an in-repo skill, or the global path for a cross-project one.
- Claude Code
- Codex
- Cursor
- Any agent (.agents)
Verify it loaded
Open a fresh agent session and ask:“Using Alakazam, how do I embed a world in a browser without exposing my secret key?”A correctly loaded skill answers with the two-token rule: your server mints a short-lived session token (
POST /v1/sessions/token with the secret key), and the browser uses that token with @alakazamworld/embed, never the key itself. If the agent suggests putting the API key in the browser, the skill didn’t load, confirm the file exists at the path your agent reads.
Next steps
Agent quickstart
Pair the skill with a one-paste onboarding prompt.
Quickstart
Create your first world and embed it.

