The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including community-provided workflow JSON files and prompt text, which serve as ingestion points for potential malicious instructions.
Ingestion points: workflow_json, workflow_path, and user-supplied prompt text from community sources (SKILL.md).
Boundary markers: The skill does not specify any delimiters or warnings to ignore instructions embedded within the processed workflow data.
Capability inventory: The skill utilizes tools like comfyui_image and comfyui_video, which perform network operations to a ComfyUI server and can trigger model downloads/file placement (SKILL.md).
Sanitization: No evidence of escaping, filtering, or validation of the external workflow content is mentioned.
[EXTERNAL_DOWNLOADS]: The skill describes a failure handling mechanism that involves reading a missing_models payload containing download URLs for external models, which are then placed in local filesystem paths (SKILL.md).
[COMMAND_EXECUTION]: The skill instructs the agent to advise users to install 'custom nodes' via ComfyUI Manager or author-documented paths, which involves the execution and integration of third-party Python code into the local environment (SKILL.md).

comfyui