The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external sources (Calibre books) and processes that data to generate new instructions (SKILL.md).
Ingestion points: Data is pulled from the local library via calibre_get_content, calibre_search, and calibre_semantic_search (SKILL.md).
Boundary markers: The instructions do not specify explicit delimiters or 'ignore' instructions for the ingested book content.
Capability inventory: The skill can write files to the filesystem (creating new skill directories) and modify library metadata via calibre_update_book.
Sanitization: The skill relies on natural language instructions ('Own words only', 'Zero verbatim sentences') to prevent direct leakage of source content into the generated instructions, which acts as a manual/LLM-based filter.
[COMMAND_EXECUTION]: The skill utilizes the calibre-mcp toolset to perform operations on the local filesystem and library database, including searching, reading content, and optionally updating book metadata.
[DYNAMIC_EXECUTION]: The skill generates new executable instruction files (SKILL.md) and metadata (distill.manifest.yaml) based on processed input. While this is the intended purpose, it involves the runtime generation of agent capabilities.

calibre-distill-topic