Module 06 of 8~16 minPro
Designing Tools, Resources, and Prompts an LLM Can Actually Drive
Design for the model as the user: tool names, descriptions, and schemas an LLM can choose correctly, structured results, recoverable errors, idempotency and safe-by-default destructive actions, and tool vs resource vs prompt decisions.
§ You will learn
- Write tool names, descriptions, and JSON schemas that an LLM selects and fills correctly on the first try, instead of guessing or flailing across your server
- Return structured, parseable results and recoverable error surfaces so the agent can self-correct without a human in the loop
- Decide correctly when to expose something as a tool, a resource, or a prompt, and design destructive actions to be idempotent and safe by default
- Build discoverability habits (naming conventions, scoping, examples) that make a multi-tool server like LightRAG's 30 tools navigable to the model
§ Sealed entry
This module is part of the Pro record.
5 sections · ~16 min · objectives above are the preview