Designing Tools for LLM Agents Quiz

Answers and explanations appear as you take the quiz.

1. What does the model actually use to decide how to call a tool?
   • The tool's source code
   • The name, description, schema and returned text
   • The server's latency history
   • The programming language the tool is written in
2. Why might an agent use prefixes such as github_create_issue and jira_create_issue?
   • Prefixes make tools run faster
   • APIs require prefixes
   • They keep similar tools from different services apart
   • They reduce the size of JSON Schema files
3. Which parameter design is most likely to produce correct arguments?
   • An enum listing the allowed values
   • A free-text string called 'value'
   • An untyped field with no description
   • A parameter named 'x'
4. What makes a good tool error message?
   • A numeric error code only
   • A generic 'something went wrong'
   • Nothing, so no tokens are spent
   • It explains what went wrong and suggests what to try next
5. Why can adding many tools reduce agent performance?
   • Models can only call one tool per conversation
   • Each definition takes up context, and overlapping tools make selection harder
   • Tools stop working once there are more than five
   • Tool count has no effect
6. Compared with mirroring every API endpoint one-to-one, what does the guide recommend?
   • Fewer, higher-level tools shaped around real tasks
   • Exposing every endpoint so the agent has full coverage
   • Using only one tool for everything
   • Letting the model write its own HTTP requests
7. Why should a tool paginate or truncate its output by default?
   • To hide data from users
   • Because models cannot read numbers
   • To avoid filling the context window with low-value data
   • Because JSON Schema requires it
8. Why make tools idempotent or give them an idempotency key?
   • It makes descriptions shorter
   • It lets the model skip schemas
   • It speeds up model inference
   • A retried call will not create duplicate records