The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill parses untrusted external files (EPUB, MOBI, PDF) and returns their raw text content to the agent. This content could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Files are read in extract_epub.py, extract_mobi.py, and extract_pdf.py.
Boundary markers: Absent; there are no delimiters or instructions to ignore embedded text.
Capability inventory: The text is returned to the agent for processing.
Sanitization: Only HTML tags are removed (in EPUB); text content is unvalidated.
[Command Execution] (LOW): The script extract_mobi.py executes the system command ebook-convert. While it uses subprocess.run with a list of arguments (mitigating shell injection), it relies on an external binary to process potentially malformed user data.
[External Downloads] (LOW): The setup.sh script installs third-party Python packages (ebooklib, beautifulsoup4, PyMuPDF) and suggests installing Calibre. These are standard, well-known tools, making the risk low per [TRUST-SCOPE-RULE].

ebook-extractor