flyworks-avatar-video

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • Indirect Prompt Injection (LOW): The skill ingests untrusted data from users (text, audio, and images) that is then processed by an external API. This creates a surface for injection where malicious instructions in the input could attempt to influence agent behavior. \n
  • Ingestion points: The create_video (--text, --audio), create_talking_photo (--image), and clone_voice (--audio) tools in scripts/hifly_client.py. \n
  • Boundary markers: None documented; the tool interface does not specify delimiters to isolate user content. \n
  • Capability inventory: The skill performs network requests and file operations. \n
  • Sanitization: Not verifiable as the logic resides in the external hifly_client.py script.\n- Data Exfiltration (LOW): The skill communicates with hfw-api.hifly.cc. While this is the legitimate service endpoint, it is not on the trusted domain whitelist, and communication involves transmitting user-provided data externally.\n- Command Execution (LOW): The skill's primary function is executed via a Python script (scripts/hifly_client.py) called with user-supplied arguments. This is a standard pattern for agent skills but relies on the script to safely handle these inputs.\n- External Downloads (LOW): Installation instructions reference a non-whitelisted GitHub repository (Flyworks-AI/skills). Although the organization is untrusted, the severity is kept at LOW as this is the primary source for the skill's own code.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:45 PM