The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection via the processing of user-supplied prompt text.
Ingestion points: User-provided prompt strings are ingested into the agent context through natural language requests and passed to scripts/prepare_workflow.py.
Boundary markers: The skill does not implement delimiters or safety warnings when interpolating untrusted prompt text into the workflow JSON templates.
Capability inventory: The skill possesses capabilities for network operations (via the ComfyUI MCP) and file system writes to generate patched workflow files.
Sanitization: No escaping or safety-focused validation is applied to the prompt text before it is used in downstream model generation tasks.
[OBFUSCATION]: Static detection flagged hidden Unicode characters in assets/workflows/i2v-turbo.ui.json and assets/workflows/i2v.ui.json. Specifically, the U+2028 (Line Separator) character was found at the end of prompt strings within the widgets_values field of node 105. Hidden Unicode characters in prompt fields can be used for steganographic instruction delivery, though in this case, it may be a formatting artifact.

minimax-h3-comfyui