The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied workflow_json and workflow_path inputs, which are complex data structures that can serve as vectors for indirect prompt injection. \n
Ingestion points: Untrusted JSON workflows provided by users or imported from community sources as described in the 'Choosing a Workflow' and 'Server Contract' sections. \n
Boundary markers: The instructions lack specific boundary markers or 'ignore' directives to prevent the agent from following instructions accidentally embedded within the JSON nodes. \n
Capability inventory: The skill coordinates the use of tools (comfyui_image, comfyui_video, comfyui_music) that perform network requests to a local ComfyUI server and manage local file artifacts. \n
Sanitization: There is no requirement mentioned for the agent to sanitize or validate the content of the nodes within the user-provided workflows before processing them. \n- [EXTERNAL_DOWNLOADS]: The skill outlines a process for resolving missing models using download URLs provided in a missing_models data structure. Although this is a functional requirement for the task, it introduces a dependency on external sources for binary model files.

comfyui