soia-pkm-clip-douyin
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads media files (MP4) from Douyin's official content delivery networks. It also identifies a dependency on the well-known 'playwright' library and requires the installation of browser binaries to perform its core function of metadata extraction.
- [COMMAND_EXECUTION]: The script contains logic to detect missing dependencies and provides the agent with the specific shell commands needed to install them (e.g.,
pip install playwright && python -m playwright install chromium). - [DATA_EXPOSURE]: The skill manages data across the local file system by writing Markdown notes to an Obsidian vault and saving video files to the system's Downloads directory. It demonstrates security awareness by explicitly redacting signing tokens (msToken, a_bogus) from its output and storage to prevent credential exposure.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from the web (video descriptions and titles from Douyin) and interpolates it into persistent Markdown files.
- Ingestion points: Data is fetched via Playwright from the Douyin web API (scripts/archive_douyin.py).
- Boundary markers: The untrusted content is clearly delineated within Markdown headers and callout blocks.
- Capability inventory: The skill has file-write access and utilizes the Playwright library for browser-based automation.
- Sanitization: Filenames are sanitized using regex to remove illegal characters, though the caption text itself is stored as raw text in the '## 原文' section.
Audit Metadata