A planner agent must choose among dozens of operational tools depending on the incident type. Hard-coding a fixed tool list in every prompt is unmaintainable. Which approach best builds the tool ecosystem?
Select an answer to reveal the explanation.
Short Explanation
You need function calling with specified and dynamic tool use. Filter the tool catalog by incident class, validate arguments, and execute only allow-listed tools. That keeps the planner accurate and safe as the catalog grows.
Full Explanation
Correct answer — the function-calling approach with specified and dynamic tool use: Building tool ecosystems includes function calling, specified tool use, and dynamic tool use. Filtering available tools by context, validating arguments, and enforcing allow lists scales better than static mega-prompts or unconstrained privileges.
Why the free-text shell-command approach is wrong: Free-text shell commands lack structured calling, validation, and safe automation.
Why the full-admin registration approach is wrong: Admin privileges on every agent violate least privilege.
Why the full-OpenAPI-every-turn approach is wrong: Unfiltered full OpenAPI injection is costly and degrades tool selection.
Exam tip: Dynamic catalogs plus validation beat “all tools always.”