The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several shell commands to extract and process book data, including unzip, find, sort, pdftotext, and the gbrain CLI utility.
[DYNAMIC_EXECUTION]: A Python script is dynamically generated via a shell heredoc to perform HTML text extraction using BeautifulSoup. This is used for local processing of book content.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from EPUB and PDF files alongside sensitive personal context. A maliciously crafted book could attempt to influence the subagents' behavior.
Ingestion points: EPUB and PDF files provided by the user.
Boundary markers: Use of a dedicated CLI runtime and read-only tool restrictions for subagents.
Capability inventory: Access to personal brain pages, shell execution, and writing to the media/books/ directory.
Sanitization: HTML tags are removed during extraction, but text content is not otherwise sanitized for prompt injection patterns.
[DATA_EXFILTRATION]: The skill assembles a 'context pack' containing sensitive files like USER.md and SOUL.md along with recent personal reflections. This aggregated data is then processed by an external language model (Claude Opus) via the gbrain CLI to generate the analysis.

book-mirror