Comfy Tools Setup

comfy-tools-setup

Use this skill to install or validate the Python tools that back the comfy-agent-tools skills. The public install flow is skills-first; the agent installs the CLIs only when they are needed.

Detect Mode

First determine whether you are inside the comfy-agent-tools repo:

test -f pyproject.toml && test -d comfy_agent_tools && test -d skills

If true, prefer repo/dev commands:

uv sync --extra dev
uv run comfy-models validate

If not in this repo, use globally installed tools from uv tool.

Install Tools

Check for uv before installing:

command -v uv

If uv is missing, ask the user to install uv first. Do not install package managers or system packages without explicit user approval.

Install the CLIs:

uv tool install git+https://github.com/quinteroac/comfy-agent-tools

Upgrade the CLIs:

uv tool upgrade comfy-agent-tools

Validate that the expected commands exist:

command -v comfy-imagegen
command -v comfy-imagedescribe
command -v comfy-videogen
command -v comfy-musicgen
command -v comfy-models

Model Config

If .comfy-agent-tools.json is missing in the current project, ask the user where they want to store or use ComfyUI model files for these tools. Do not pick /mnt/models/comfyui automatically; mention it only as an example.

After the user provides a path, initialize config and set that exact path:

comfy-models init
comfy-models set-models-dir 

Then validate:

comfy-models validate
comfy-models show

In repo/dev mode, prefix CLI commands with uv run, for example:

uv run comfy-models init
uv run comfy-models set-models-dir 
uv run comfy-models validate

If .comfy-agent-tools.json already exists, respect its configured models_dir and do not ask again unless validation fails or the user wants to change it.

If validation reports missing models, switch to comfy-model-downloader when the user is trying to run a supported built-in capability. Use comfy-model-onboarding for custom checkpoints, changed defaults, or unsupported profiles.

Skill Installation

Users install the agent guidance first:

npx skills add quinteroac/comfy-agent-tools

After that, the skills can bootstrap the Python tools on demand.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: quinteroac
• Source: quinteroac/comfy-agent-tools
• License: MIT

Install and usage instructions live in the source repository linked above.