flux3-klein-stills
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill includes two Python scripts,
scripts/comfy_client.pyandscripts/preflight.py, which are used to interact with a ComfyUI server. These scripts rely solely on Python standard libraries and handle authentication via an environment variable (COMFY_API_KEY) in a standard manner.- [SAFE]: The ComfyUI workflows provided in theworkflows/directory are standard API-format JSON files. They reference specific model weights and define the image generation process, including a prompt-upsampling stage that uses an internal system prompt to guide a text-generation model.- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text prompts which are passed to a ComfyUI workflow. This represents a surface for indirect prompt injection, as the prompts are processed by an upsampling model within the workflow. - Ingestion points: User prompts are ingested via the
--setparameter inscripts/comfy_client.pyand applied to workflow nodes such as4.textor18:17.value. - Boundary markers: The workflows do not employ explicit delimiters or instructions to isolate user input from the rest of the prompt logic.
- Capability inventory: The skill's scripts perform network requests to a specified ComfyUI server and write generated files to the local filesystem via the
fetchcommand. - Sanitization: No validation or sanitization of the input prompt strings is performed before they are sent to the remote server.
Audit Metadata