AI Service Setup

The first time you open the editor with kumo installed, setup.py runs once. It looks at your project folder for a .claude/ (or a CLAUDE.md / AGENTS.md) and copies the bundled skill file into Claude’s skills directory.

After that, Claude knows about kumo. Ask it to create or edit an Unreal asset and it calls cli.py itself.

You can re-run setup any time:

python Plugins/Kumo/setup.py

Supported assistants

Assistant	Skill location	In-editor chat panel
Claude	.claude/skills/unreal/	✅ Bundled
Codex	.agents/skills/unreal/	✅ Bundled

Claude and OpenAI Codex both drive the chat panel. First-class integration for Cursor, GitHub Copilot and Windsurf ships next; for now, those tools drive kumo via the bring-your-own CLI path below.

Signing in to Codex

Codex signs in with an OpenAI API key. Create one at platform.openai.com/api-keys (usage is billed to your OpenAI account), then:

1. Open the chat panel. On the sign-in screen, pick Codex. If you are already signed in to Claude, choose Switch account first.
2. Paste the key and confirm. The first time, kumo installs the Codex runtime for you; that takes about a minute.
3. The chat opens on Codex. kumo installs its skill into .agents/skills/ for this project as you sign in, so Codex can work in the editor from its first message.

Signing in sets Codex’s own sign-in on this machine to that key, as codex login --with-api-key would. The key itself is kept in your user settings folder, never in the project.

Once you are signed in to both agents, the Agent selector under the chat input switches between them. The conversation on screen carries over to the agent you switch to, and the session history lists both agents’ sessions; opening one resumes it in the agent it belongs to. If OpenAI refuses a request (a mistyped key, no credit left), the chat shows OpenAI’s own message.

Uninstalling

To remove the installed skill:

python Plugins/Kumo/setup.py --uninstall

How the skill works

A skill file is a short markdown document that describes kumo to the AI assistant:

• When to engage — any request to create, edit, or inspect an Unreal asset
• What’s available — a pointer to the CLI’s self-documenting schema (not the schema itself — keeps the per-session cost low)
• How to call it — invoke cli.py schema to learn operations, then cli.py apply to run a plan

Claude reads the skill on session start. From your side this means you can say “create a pickup Blueprint” and Claude knows to use kumo rather than writing Python or C++ from scratch.

Note

If you update kumo to a version that adds new operations, re-run setup.py to refresh the skill. The installer is idempotent — already-current skills are skipped, so it’s safe to run every editor launch (and indeed it does).

Any other AI tool

Any tool that can shell out to a CLI can drive kumo. Point it at the CLI and let the self-documenting schema do the rest:

CLI="python Plugins/Kumo/Scripts/cli.py"

# Discover every operation and its parameters
$CLI schema

# Apply an operations plan from stdin
echo '[{"op":"create_blueprint","path":"/Game/Pickups","name":"BP_Pickup","parent":"Actor"}]' \
    | $CLI apply --stdin

# Inspect an existing asset
$CLI inspect /Game/Pickups/BP_Pickup

Pipe cli.py schema into your tool’s system prompt once; the model produces valid operation plans on its own from then on. See the CLI Commands reference for the full command list.