The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The skill provides installation instructions and links for third-party GitHub repositories that are not included in the 'Trusted GitHub Organizations' list.
Evidence: references/PURZ_EXAMPLES.md provides git clone commands for https://github.com/purzbeats/purz-comfyui-workflows and https://github.com/purzbeats/ComfyUI-Purz.
Evidence: references/OFFICIAL_DOCS.md links to templates and resources at https://github.com/Comfy-Org/ and https://github.com/jtydhr88/.
[PROMPT_INJECTION] (LOW): The skill exhibits an indirect prompt injection surface as it is designed to ingest untrusted user Python code and transform it into executable ComfyUI nodes.
Ingestion points: The skill body in SKILL.md identifies user-provided Python snippets as primary input for conversion (Step 1).
Boundary markers: Absent. There are no instructions to the agent to treat user-provided snippets as potentially adversarial or to ignore embedded instructions.
Capability inventory: The resulting nodes are intended for execution on the user's ComfyUI server and have significant capabilities, including filesystem access via the folder_paths module and model management via comfy.model_management (documented in references/NODE_TEMPLATE.md).
Sanitization: Absent. No instructions are provided to sanitize, escape, or validate the user-provided Python logic before wrapping it into the node template.

comfy-nodes