ComfyUI Skill for OpenClaw

Run local ComfyUI workflows via the HTTP API. The agent edits workflow JSON (prompt, style, seed) then queues and polls using the bundled script.

Prerequisites

• ComfyUI installed and runnable (e.g. ~/ComfyUI with a venv). The skill can guide the user to install it if the server is not reachable.
• Python 3 with the ComfyUI venv (use ~/ComfyUI/venv/bin/python; do not rely on bare python on PATH).

What this skill does

• Instructs the agent to read workflow JSON, find prompt/style/sampler nodes, edit them, and write a temp workflow file.
• Runs comfyui_run.py to queue the workflow and poll until completion.
• Tells the agent to send generated images back to the user and how to recover if the ComfyUI server is not running.

Layout

comfyui/
├── SKILL.md           # Main instructions for the agent
├── README.md          # This file
├── CONTRIBUTING.md    # Where to report issues
├── .clawhub/
│   └── origin.json    # ClawHub slug and version
├── scripts/
│   ├── comfyui_run.py       # Queue workflow, poll, print images
│   └── download_weights.py  # Download model weight URLs to ComfyUI/models/
└── assets/
    ├── default-workflow.json  # Example workflow
    └── tmp-workflow.json      # Written by agent (editable)

Publishing to ClawHub

The skill only appears on ClawHub after you publish it. Until then it exists only locally.

1. Install the CLI: npm i -g clawhub
2. Log in: clawhub login (GitHub account must be at least one week old to publish)
3. From the workspace root (parent of skills/):

   clawhub publish ./skills/comfyui --slug comfyui --name "ComfyUI" --version 1.0.1 --changelog "Describe changes"

4. For later updates, bump the version (e.g. 1.0.2) and run the same command with a new --changelog.

Finding your skill: Open clawhub.ai, search for comfyui or ComfyUI, or check your profile for published skills. Install with: clawhub install comfyui