Integrations

OpenClaw skills

Define OpenClaw skills for ComfyUI or Python pipelines so agents can generate, save, and reuse assets in GenAsset.

Overview

OpenClaw skills let your agents follow one repeatable flow: generate in your chosen tool, save to GenAsset, then reuse approved versions later. You can start from either ComfyUI or Python.

ComfyUI path

Run graph, capture output + workflow metadata, save as next version.

Python path

Run script/pipeline, capture prompt + params, save as next version.

Skill contract

Keep one shared contract so every agent run saves consistent fields. This makes versions comparable across tools and teammates.

OpenClaw skill contract
yaml
name: genasset_save_and_reuse
description: Save outputs to GenAsset and reload approved recipes for reuse.
inputs:
  - workspace_token
  - mode            # comfyui | python
  - asset_name
  - prompt
  - source
  - model
  - image_input
outputs:
  - asset_id
  - version_id
  - signed_preview_url
steps:
  - workspace_status
  - (generate in selected mode)
  - save_image_to_genasset
  - load_version_recipe

ComfyUI skill example

Use this template when generation happens in ComfyUI and OpenClaw orchestrates save + reuse calls through MCP.

OpenClaw ComfyUI skill template
yaml
name: openclaw_comfyui_to_genasset
mode: comfyui
description: Run a ComfyUI graph, then save image + recipe metadata to GenAsset.
run:
  - action: run_comfyui_graph
    with:
      graph: sdxl-character-edit.json
  - action: save_image_to_genasset
    with:
      workspace_token: "${workspace_token}"
      asset_name: "${asset_name}"
      image_url: "${comfyui.output_url}"
      prompt: "${comfyui.prompt}"
      model: "${comfyui.model}"
      source: "comfyui"
      metadata:
        seed: "${comfyui.seed}"
        steps: "${comfyui.steps}"
        cfg: "${comfyui.cfg}"
        sampler: "${comfyui.sampler}"
        scheduler: "${comfyui.scheduler}"
        workflow_json: "${comfyui.workflow_json}"

Python skill example

Use this template when generation happens in Python. Save the output image and structured runtime metadata back to GenAsset.

OpenClaw Python skill template
yaml
name: openclaw_python_to_genasset
mode: python
description: Generate in Python, then persist output + recipe metadata in GenAsset.
run:
  - action: run_python_pipeline
    with:
      script: generate_character_variant.py
  - action: save_image_to_genasset
    with:
      workspace_token: "${workspace_token}"
      asset_name: "${asset_name}"
      image_path: "${python.output_path}"
      prompt: "${python.prompt}"
      model: "${python.model}"
      source: "python-openclaw"
      metadata:
        seed: "${python.seed}"
        runtime_seconds: "${python.runtime_seconds}"
        parameters: "${python.parameters}"

Usage prompts

After skills are loaded, these are practical prompts for day-to-day team usage:

Prompt examples
text
# ComfyUI-first flow
Generate a cinematic rain portrait in ComfyUI and save to GenAsset as
asset_name "spring-hero". Then load the latest version recipe.

# Python-first flow
Generate a red-coat variant with Python from the approved GenAsset version,
save it as the next version of "spring-hero", and return asset_id/version_id.

# Reuse flow
Find the approved version of "spring-hero", explain what changed from v2 to v3,
and recommend whether to use v2 or v3 for final campaign output.

Saved metadata

For reproducibility, always save common fields across both paths: prompt, source, model, seed when available, and provider/runtime details in metadata.

Closed-model note
ComfyUI runs usually provide full replay metadata. Closed providers may hide seed/sampler internals. Save all available provider metadata and label replay quality as partial when needed.