The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a variety of shell commands including unzip, find, sort, pdftotext, and a custom CLI tool gbrain book-mirror. It also executes a multi-line Python script passed via a heredoc to python3 to process extracted HTML content.
[EXTERNAL_DOWNLOADS]: The skill explicitly instructs the agent to install external Python packages at runtime using pip3 install beautifulsoup4 lxml if they are missing.
[DATA_EXFILTRATION]: The skill collects highly sensitive personal information from the user's environment, including USER.md, SOUL.md, and the last 14 days of daily reflections from wiki/personal/reflections/. While this data is intended to be used as context for the personalization feature, it represents a significant exposure of private user data to the underlying LLM provider.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It extracts text from untrusted external files (EPUB/PDF books) and processes this content in a high-capability environment that has access to the user's personal history and the ability to write new pages.
Ingestion points: EPUB and PDF files provided by the user (extracted via unzip and pdftotext).
Boundary markers: No explicit markers or sanitization are mentioned to distinguish untrusted book content from the personal context pack during the analysis phase.
Capability inventory: Access to shell commands, Python execution, and the ability to write pages via put_page (orchestrated by the CLI).
Sanitization: None. The skill relies on the assumption that the gbrain book-mirror CLI handles trust narrowing, which is an unverified claim.

book-mirror