The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the system's pandoc utility to perform the initial file conversion.
Evidence: In scripts/convert_book.py, the extract_epub_with_pandoc function uses subprocess.run with a list of arguments to execute pandoc safely.
Analysis: By passing arguments as a list rather than a single string with shell=True, the script avoids common shell injection vulnerabilities even when handling user-provided file paths.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted content from EPUB files which creates a potential surface for indirect prompt injection.
Ingestion points: Content is extracted from user-provided EPUB files into raw/book-parsed.md in the project directory.
Boundary markers: There are no specific delimiters or guardrail instructions to prevent the subagents from following instructions potentially embedded within the book text.
Capability inventory: The subagents are primarily tasked with text formatting and file writing, while the main agent executes the provided Python scripts.
Sanitization: The skill does not sanitize the extracted Markdown text before passing it to subagents for formatting.
[SAFE]: The skill implements a robust architectural pattern by offloading the processing of potentially large and untrusted book content to isolated subagents. This 'subagent isolation' helps protect the main agent's context and limits the potential impact of any malicious content found within the books being processed.

book-converter