reframe-aspect

Fail

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: HIGHDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The scripts/comfy_client.py script contains a path traversal vulnerability in its fetch function. It uses os.path.join to combine an output directory with a filename provided by the remote server without validation. If the server provides an absolute path or a path with parent directory references (e.g., ../../.ssh/authorized_keys), the script will write the downloaded content to that location, potentially compromising system integrity.
  • [DATA_EXFILTRATION]: The skill provides a command-line utility for uploading local files to a remote ComfyUI server. This creates a capability for an agent to exfiltrate sensitive data (such as API keys, credentials, or private configuration files) if it is instructed to do so or if it incorrectly identifies a sensitive file as a valid input for the video workflow.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its text and video processing inputs.
  • Ingestion points: The prompt input (node 5138.text) and the source video file input (node 1.file) in workflows/ltx23_video_reframe.api.json.
  • Boundary markers: No delimiters or safety instructions are present to prevent the agent from following commands embedded in these inputs.
  • Capability inventory: The skill has extensive filesystem access (read/write) and network capabilities via the comfy_client.py script.
  • Sanitization: There is no sanitization or validation of the text prompt or video content before processing.
  • [EXTERNAL_DOWNLOADS]: The documentation and workflow files reference external resources for models and node packs from well-known providers such as Hugging Face and the Lightricks GitHub repository. These are legitimate dependencies required for the skill's intended video processing functionality and are documented neutrally.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 4, 2026, 08:11 PM