Adding a Tool

v1 status: stub. Step-by-step walkthrough TBD. In the meantime, see examples/custom_tool.rs in the rantaiclaw repo  and read Concepts → Tools & Skills.

The shape

1. Create src/tools/<your_tool>.rs and implement the Tool trait
2. Define a strict JSON Schema for parameters — what the LLM is told is what the runtime validates
3. Return ToolResult, never panic
4. Register in src/tools/mod.rs
5. Add tests for success, validation failure, and security-denied paths

Anti-patterns to avoid

• Tools that wrap arbitrary shell commands. Use the existing shell tool with an allowlist; do not write a new “shell-with-extra-flags” tool.
• Tools that bypass the approval gate. The gate runs before execute; you do not opt out.
• Tools that mutate gateway / security policy directly. That is an architecture boundary violation; the policy lives in src/security/ for a reason.

Reference implementations

• src/tools/file_read.rs — minimal read-only tool
• src/tools/shell.rs — shell with sandbox + allowlist
• src/tools/web_search.rs — outbound HTTP with structured result
• src/tools/memory_recall.rs — semantic search over memory