securetools

Security-hardened tool definitions for R LLM agents: path scoping, parameterized SQL, domain allow-lists, and rate limits.

R → ian-flores/securetools

Left alone, an LLM can call system(), write to any path, and run arbitrary SQL. securetools ships pre-built securer tool factories that let you give agents real capabilities without giving them the keys to the kingdom: file access scoped to specific paths, parameterized queries, domain allow-lists for HTTP, size limits, and rate limiting.

guarded_tool() wraps any of them with secureguard checks, so one tool definition carries both the capability and its policy.


← all projects