skills/openclaw/skills/runware/Gen Agent Trust Hub

runware

Fail

Audited by Gen Agent Trust Hub on Feb 14, 2026

Risk Level: HIGHDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION] (HIGH): The load_image_as_datauri function in scripts/video.py reads local file contents based on user-supplied paths and encodes them as Base64 to be sent to the Runware API.
  • Ingestion points: The images positional argument in cmd_img2vid (scripts/video.py).
  • Boundary markers: None. File paths are used directly to open files.
  • Capability inventory: urllib.request.urlopen (Network POST) and open(path, "rb").read() (Local file read).
  • Sanitization: None. The script only checks if the file exists and uses the extension for MIME type mapping. This allows an attacker to exfiltrate sensitive files (e.g., ~/.aws/credentials, .env) by passing them as 'images' to the tool.
  • [COMMAND_EXECUTION] (LOW): The skill documentation encourages the execution of local Python scripts with various arguments. While the script itself uses argparse correctly and avoids os.system or shell=True for its internal logic, the pattern of executing local scripts based on agent-generated command lines is a standard capability that requires strict supervision of the agent's shell access.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 14, 2026, 06:02 PM