The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses the ~/.hermes/.env file to retrieve NVIDIA_API_KEY or OPENROUTER_API_KEY. While this is a standard configuration pattern for the skill, the .env file is a sensitive path often containing credentials.
[DATA_EXFILTRATION]: Text chunks from the library and user API keys are transmitted to external endpoints at integrate.api.nvidia.com and openrouter.ai to generate embeddings. These are well-known services and the behavior is essential for the skill's functionality.
[PERSISTENCE]: The SKILL.md file includes instructions and examples for users to configure persistent automated indexing using system tools like cron (Linux/macOS) or launchd (macOS).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from EPUB and PDF documents. This creates an attack surface for indirect prompt injection, where malicious instructions embedded in a book could be retrieved and influence the agent's behavior during a search.
[DYNAMIC_EXECUTION]: The skill uses SQLite's extension loading mechanism (conn.enable_load_extension(True)) to dynamically load the sqlite-vec extension at runtime, which is required for vector search operations.

library-rag