What problem does it solve? Poorly designed tools cause agents to pick the wrong tool, hallucinate IDs, retry permanently-failing calls, and burn context on oversized results. This Skill provides a review checklist and design rules for tools exposed to an LLM so every name, description, parameter schema, error message, and return shape is optimized for model selection and token efficiency. ## Core Features & Use Cases - Pre-ship review checklist: Validates verb-first namespaced names, disambiguating descriptions, enum-constrained parameters, minimal required sets, and declared read_only metadata. - Error message design: Treats errors as prompts that state what was wrong, the valid space, the next action, and whether the failure is transient or terminal. - Token-efficient returns: Enforces summary-plus-handle patterns, projection support, hard-capped output with announced truncation, and machine-actionable structure. - Tool-set sizing: Applies the 30-50 tool confusion threshold with an escalation ladder covering deferred loading, tool search, and code execution, plus MCP security posture guidance. - Use Case: When defining an MCP server or function-calling schema, run the checklist to catch near-synonym tools, opaque ID parameters, and uncapped payloads before they degrade agent behavior. ## Quick Start Ask the agent to review your tool definitions against the agent tool design checklist and fix any naming, description, error message, or return shape defects it finds.
