blog-narrator
Warn
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [DATA_EXFILTRATION]: The
embed_imagesfunction inscripts/narrator_core.pyis vulnerable to path traversal. It processes image tags starting with/pics/but fails to sanitize parent directory sequences (..). This allows a malicious Markdown file to include and expose sensitive local files by embedding them as Base64 strings in the generated HTML. - [EXTERNAL_DOWNLOADS]: Fetches ONNX models and configuration files from ModelScope (
modelscope.cn), a well-known AI model repository, to support the audio-to-text alignment functionality. - [COMMAND_EXECUTION]: Invokes the
ffmpegsystem utility viasubprocess.runto decode audio data for the ASR (Automatic Speech Recognition) component. - [COMMAND_EXECUTION]: Uses the
webbrowsermodule to automatically launch the user's default browser and display the resulting HTML file upon completion.
Audit Metadata