camera-lens-lut

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external video and image files and accepts text-based prompts which are then interpolated into ComfyUI workflows. This creates a surface for indirect prompt injection if the processed media contains hidden instructions.
  • Ingestion points: The LoadVideo node in workflows/backlot_direct.api.json and various input fields like prompt and text_prompt defined in assets/comfyui-backlot/web/backlot_menu.js.
  • Boundary markers: There are no explicit delimiters or instructions provided to the agent or the ComfyUI nodes to ignore potentially malicious content embedded within the processed media.
  • Capability inventory: The skill utilizes scripts/comfy_client.py which has the ability to upload files to the server, execute workflows, and download resulting files, providing a path for data manipulation or exfiltration if the agent is subverted.
  • Sanitization: The provided scripts and workflows do not perform sanitization or validation of the input media metadata or prompt strings before they are processed by the ComfyUI backend.
  • [EXTERNAL_DOWNLOADS]: The skill instructions and documentation reference external repositories for necessary node packs.
  • It directs the user to install the DepthAnythingV2Preprocessor from a public repository (https://github.com/Fannovel16/comfyui_controlnet_aux).
  • The assets/comfyui-backlot pack is identified as originating from the Trenchplanet/mogfield project.
  • [COMMAND_EXECUTION]: The skill requires the agent to execute bundled Python scripts to manage the ComfyUI workflow.
  • scripts/comfy_client.py and scripts/preflight.py are used to communicate with the ComfyUI API, performing network requests and file system operations such as reading input files for upload and writing output files.
  • These scripts also handle the COMFY_API_KEY environment variable to authenticate requests to the server.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 08:11 PM