doubao-external-audio-video

Pass

Audited by Gen Agent Trust Hub on Oct 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/doubao_cdp_browser.py) and shell scripts (scripts/doubao_lock_soundtrack.sh) to perform browser automation, file handling, and media post-processing.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external files which could contain malicious instructions designed to influence the agent or the target platform.
  • Ingestion points: Reads input from PROMPT.md, SONG.mp3, and REFERENCE.png.
  • Boundary markers: The skill includes an instruction to avoid pasting local paths into prompts, but lacks programmatic sanitization of the input content.
  • Capability inventory: The agent can execute shell commands, perform file reads/writes, and control a live browser session via CDP.
  • Sanitization: No explicit sanitization or validation of the external file contents is implemented.
  • [DYNAMIC_EXECUTION]: The skill invokes system utilities like ffmpeg and ffprobe at runtime to mux audio streams and verify video file integrity.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 19, 2026, 06:23 PM