voiceover

Warn

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes Python scripts (scripts/comfy_client.py and scripts/preflight.py) and instructs the agent to execute them to interact with a remote ComfyUI server. These scripts handle file system operations and network communication.
  • [COMMAND_EXECUTION]: The fetch function in scripts/comfy_client.py is vulnerable to path traversal. It uses os.path.join(out_dir, f["filename"]) to save files downloaded from the server, where f["filename"] is a string provided by the server response. A malicious or compromised server could return a filename containing traversal sequences (e.g., ../../target_file) to overwrite arbitrary files on the user's system.
  • [DATA_EXFILTRATION]: The upload command in scripts/comfy_client.py reads local files and transmits them to a remote server. This functionality could be abused to exfiltrate sensitive files, such as credentials or configuration files, if the agent is directed to upload unauthorized paths to an attacker-controlled URL.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text for narration and voice-over generation, interpolating it directly into ComfyUI workflow parameters without sanitization.
  • Ingestion points: Workflow node parameters (e.g., node 1.text in voice_chatterbox_multilingual.api.json).
  • Boundary markers: None; the instructions tell the agent to pass the user's text verbatim.
  • Capability inventory: The bundled comfy_client.py script provides capabilities to read/write local files and communicate with external network endpoints.
  • Sanitization: No validation or escaping is performed on the input text before transmission to the API.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 4, 2026, 08:11 PM