bailian-gen
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted multi-modal data including images, audio, video, and text prompts. This data is fed into multi-modal models (e.g., Qwen-VL, Qwen-Audio), creating a surface where hidden instructions in media could influence agent behavior.\n
- Ingestion points: Found in SKILL.md, reference/omni.md, reference/vision.md, and reference/speech.md via flags such as --image, --audio, --video, and --prompt.\n
- Boundary markers: No explicit delimiters or instructions to ignore embedded content are provided in the skill instructions.\n
- Capability inventory: The skill can write files to the local system (--out-dir, --download), execute commands via the bl CLI, and suggests user execution of ffmpeg and media players.\n
- Sanitization: No sanitization or validation of the content within the ingested media is performed by the skill.\n- [EXTERNAL_DOWNLOADS]: The skill relies on the external bl CLI tool and mentions ffmpeg for media processing. These are legitimate tools for the skill's stated purpose.\n- [COMMAND_EXECUTION]: The skill operates by executing commands through the bl binary to interact with Aliyun Bailian services.
Audit Metadata