The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute several local Python scripts for workflow management tasks. These include scripts/workflow_catalog.py for building workflow catalogs, scripts/comfyui_workflow.py for inspecting and patching workflow files, and scripts/comfyui_control.py for live validation.
[INDIRECT_PROMPT_INJECTION]: The skill processes external workflow definitions and handoff data, which constitutes a potential vulnerability surface for indirect prompt injection.
Ingestion points: The agent reads JSON-formatted workflow definitions from the comfyui_workflows/ library and project-specific locations like 04_generation/comfyui/.
Boundary markers: The instructions rely on the structured nature of ComfyUI JSON files but do not specify explicit delimiters or "ignore instructions" markers for natural language fields (like prompts) contained within those workflows.
Capability inventory: The skill uses subprocess calls to local Python scripts (scripts/workflow_catalog.py, scripts/comfyui_workflow.py, scripts/comfyui_control.py) to write files and perform network operations against the local ComfyUI API.
Sanitization: The skill mandates rigorous validation, including verifying node classes against the server's /object_info endpoint, performing graph inspections, and running workflow-family contract checks to ensure safety and functionality.

comfyui-workflow