managing-models

Downloads ComfyUI model weights from policy-approved sources with sha256 verification.

Updated Apr 8, 2026
One-click install
npx skills add https://github.com/ShinyGua/ComfyUI-Agent --skill managing-models-shinygua
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: managing-models
Source: https://github.com/ShinyGua/ComfyUI-Agent/tree/main/.claude/skills/managing-models
Command: npx skills add https://github.com/ShinyGua/ComfyUI-Agent --skill managing-models-shinygua

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires huggingface_hub, and includes scripts (resource) components.

What problem does it solve? It provides a sanctioned, policy-driven path for downloading ComfyUI model weights, preventing unverified or arbitrary downloads by requiring entries in policies/model_sources.yaml and explicit user opt-in. ## Core Features & Use Cases - Policy-gated downloads: Every download is keyed to an entry in policies/model_sources.yaml, so no ad-hoc URL guessing is allowed. - Integrity verification: Downloaded files are checked against an expected sha256 hash, and mismatches raise a loud structured error instead of silently proceeding. - Deferred activation gate: The skill stays dormant unless the user sets allow_model_download = true in host settings, making weight downloads an explicit halt point. - Use Case: A planner emits a model_download sub-goal for a checkpoint; the executor invokes this skill, which verifies the user opt-in, downloads via huggingface_hub or direct HTTPS URL, verifies the sha256, and returns a ModelHandle JSON. ## Quick Start Enable allow_model_download in your host settings, then ask the agent to download the approved model file listed in policies/model_sources.yaml.

Frequently Asked Questions about managing-models

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I download ComfyUI model weights with this skill?▼

Set allow_model_download to true in your host settings, ensure the model is listed in policies/model_sources.yaml with its sha256, then let the planner dispatch a model_download sub-goal. The script reads a JSON payload on stdin and returns a ModelHandle on success.

Why does the model download fail with a downloads_disabled error?▼

The skill is dormant by default and refuses to run unless allow_model_download is set to true in host settings. This gate exists so model weight downloads are always an explicit user decision, never a silent automated action.

Does it support HuggingFace model downloads?▼

Yes, HuggingFace repos are downloaded via the huggingface_hub package, while direct HTTPS URLs use urllib.request. Both paths verify the file's sha256 hash before returning a handle.

What happens when the sha256 checksum does not match?▼

A sha256 mismatch raises a loud structured error envelope rather than silently proceeding. This ensures corrupted or tampered weight files are never placed into the ComfyUI models folder unnoticed.

Can I download a model from an arbitrary URL?▼

No, every download must be keyed to an entry in policies/model_sources.yaml. There is intentionally no path for guessing URLs, which keeps all model sources reviewed and approved.