The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external system binaries using subprocess.run() to perform document extraction and analysis. This includes tools like pdftotext and pdfinfo for PDF files, and ebook-convert (from Calibre) for MOBI/AZW formats. These operations are implemented using list-style arguments to prevent shell injection risks.
[REMOTE_CODE_EXECUTION]: The script scripts/book_to_skill/dependencies.py contains functionality to install Python packages using pip (sys.executable -m pip install). While the skill defaults to reporting the command rather than executing it, it allows for unattended installation if the BOOK_SKILL_INSTALL_MISSING environment variable is set or the --install-missing yes flag is used.
[EXTERNAL_DOWNLOADS]: Through its dependency management system, the skill can download and install external Python packages from the Python Package Index (PyPI). Additionally, the skill provides links and instructions for downloading external tools like Calibre.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted document content (e.g., PDFs, Ebooks) and uses it to generate structured instructions for an AI agent. This creates a surface for indirect prompt injection if the source documents contain malicious instructions. The skill implements multiple robust mitigations:
Ingestion points: extract_document.py reads user-provided file paths for processing.
Boundary markers: The skill structure uses Markdown headers to separate data, and a validator (scripts/book_skill_validator.py) scans for common injection phrases in the output.
Capability inventory: The skill can write to the local filesystem, execute system commands, and install Python packages.
Sanitization: scripts/book_to_skill/sanitize.py strips adversarial invisible Unicode characters, and scripts/book_to_skill/zip_safety.py prevents XML/ZIP-based attacks.

book-to-skill