The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation references commands that download tools and dependencies during development setup.
The make setup script is documented to fetch rustup via curl https://sh.rustup.rs | sh. This targets a well-known service and the documentation explicitly advises manual verification of the script before execution.
The make deny target is documented to download and install cargo-deny from official Rust package registries if it is not already present.
[INDIRECT_PROMPT_INJECTION]: The documented database engine possesses an attack surface for indirect prompt injection through its LLM query planning feature.
Ingestion points: SQL query strings and responses from external LLM providers (e.g., Ollama) are processed by the query planner as documented in query-planner.md.
Boundary markers: The engine implements sanitization for questions and applies hard limits to prompt prefix sizes to mitigate injection risks.
Capability inventory: The database supports execution of WASM User Defined Functions (UDFs), which are implemented with sandbox restrictions and resource limits as detailed in wasm-udf.md.
Sanitization: The documentation describes how the engine sanitizes user questions by removing control characters and applying length constraints before they are used in LLM prompts.

fandhe-db