reading-and-book-club
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by extracting text from EPUB files and storing it for processing, creating a surface for potential instruction injection. 1. Ingestion points: The agent is instructed in SKILL.md to read EPUB files from '~/读书笔记/books/' and search WeChat cache directories at '/home/ubuntu/.hermes/cache/documents/'. 2. Boundary markers: No specific delimiters or safety instructions are provided to the agent to distinguish between book content and system instructions when processing extracted text. 3. Capability inventory: The skill utilizes file system modification (os.makedirs, file writing), package installation (pip), and network communication (WeChat) as indicated in the command snippets. 4. Sanitization: The provided Python snippets use basic regular expressions to strip HTML tags but do not perform deep content validation or filtering of potential instructions.
- [EXTERNAL_DOWNLOADS]: The skill installs the 'ebooklib' and 'beautifulsoup4' packages from the official PyPI registry (pypi.org/simple/). This is a well-known and trusted source for software dependencies.
- [COMMAND_EXECUTION]: Several shell commands are provided for file inventory auditing and directory management, including the use of 'ls' and 'cat' to verify progress and library contents.
- [DYNAMIC_EXECUTION]: The skill contains Python code blocks meant for runtime execution by the agent to parse EPUB structures and extract text into individual chapter files.
Audit Metadata