The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes a bundled Python script (scripts/libgen.py) to search for and download PDF documents from libgen.li. This is an unofficial, third-party source not associated with the vendor or a trusted organization.
[COMMAND_EXECUTION]: The orchestrator uses shell commands to execute the bundled libgen.py script and to spawn subagents for parallel processing of downloaded documents. It constructs paths dynamically using a user-provided 'slug'.
[REMOTE_CODE_EXECUTION]: The skill performs automated network operations to an unverified external domain (libgen.li) to fetch binary data. While the data is intended to be PDFs, downloading and processing files from non-trusted external sources introduces a vector for secondary attacks.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection.
Ingestion points: Untrusted PDF files downloaded from a shadow library are ingested and read 'cover to cover' by subagents.
Boundary markers: Absent. The instructions to subagents do not include delimiters or specific ignore-instructions for the external content.
Capability inventory: The agent has capabilities to execute shell scripts, write to the filesystem, and perform network requests.
Sanitization: No validation or sanitization of the PDF content or the resulting synthesized text is mentioned, allowing malicious instructions embedded in a book to potentially influence the agent's behavior during synthesis.
[COMMAND_EXECUTION]: The skill generates new executable 'skills' in the ~/.claude/skills/ directory. While this is the intended purpose, it involves script generation that will be executed in future sessions.

learn-craft