The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/setup.sh script downloads an external binary executable (annas-mcp) from an unverified public repository (https://github.com/iosifache/annas-mcp/releases) and assigns execution privileges (chmod +x). Running untrusted pre-compiled binaries presents a supply-chain execution risk.
[COMMAND_EXECUTION]: The scripts/book.py script leverages subprocess.run to invoke the local annas-mcp binary to complete searches and book downloads. While arguments are handled safely via a structured array, it relies on dynamically located executables from the host platform environment.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an unvalidated input processing boundary.
Ingestion points: Book titles, author fields, summaries, and book search query results are downloaded from Z-Library's EAPI and Anna's Archive and loaded directly into memory or displayed on screen.
Boundary markers: None are present in the script instructions to ensure content downloaded from outer mirrors is handled purely as data rather than containing hidden instructions.
Capability inventory: Local subprocess invocation (annas-mcp), configuration file-writes (~/.claude/book-tools/config.json), and file-system modifications (~/Downloads).
Sanitization: Filenames are sanitized using regex re.sub(r'[<>:"/\\|?*]', '_', filename) before being written to disk, which handles path traversal but does not sanitize potential hidden content from within processed catalogs.

zlib-download