qwen-mm-plugins-video-memory
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The
script/build_memory/build_memory.shscript automatically installs the Python packagesopencv-python-headless,numpy, anddashscopefrom the Aliyun PyPI mirror if they are not detected on the system. - [COMMAND_EXECUTION]: The Python script
script/build_memory/build_graph.pyand the shell scriptscript/build_memory/build_memory.shexecute external commands includingffmpeg,ffprobe, andpythonto process video files and manage the build pipeline. - [CREDENTIALS_UNSAFE]: The skill reads from a configuration file located at
~/.qwen-mm-plugins/configto retrieve API keys and environment variables, which involves accessing a sensitive path in the user's home directory. - [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it extracts text from videos via OCR and ASR and includes this untrusted content in LLM prompts.
- Ingestion points: Video files are processed by
build_graph.pyto extract OCR and ASR text content. - Boundary markers: Prompt templates in
script/build_memory/prompts.pyuse structured formatting but lack explicit instructions to the model to ignore potential malicious content embedded in the extracted text. - Capability inventory: Subprocess execution (ffmpeg), local filesystem access (memory directory), and network requests to Aliyun and DashScope APIs.
- Sanitization: No sanitization or filtering of the raw extracted text is performed before it is interpolated into LLM prompts for analysis.
Audit Metadata