Instructions and covered skills
Read each scenario carefully before selecting a response. Focus on the stated execution flow, data constraints, and responsibilities of the model and application. Choose the option that best supports reliable and controlled tool use. Do not rely on assumptions that are not included in the question. Stay focused on one question at a time and turn off notifications where possible. Review your selections before submitting the test.
Key Areas
This test evaluates the practical skills required to connect a language model with application functions and external services. Candidates should understand how a tool definition communicates a function name, purpose, parameters, required fields, and data types to a model. They should be able to distinguish between a model proposing a tool call and the application authorizing and executing that request.
Strong performance requires knowledge of JSON Schema concepts used in tool arguments, including object properties, arrays, enums, required fields, and constraints. The test also covers server-side validation because model-generated arguments must be treated as untrusted input. Candidates should recognize when an application needs to reject, repair, or clarify malformed arguments rather than passing them directly to a downstream service.
Workflow design is another key area. This includes preserving tool call identifiers, returning results in the expected conversation format, managing multi-step calls, and deciding when calls can run concurrently or must follow a dependency. Candidates should understand timeouts, retry behavior, idempotency, and error responses that allow a model to continue a conversation without inventing an outcome.
Security and operational control are central to dependable orchestration. Relevant practices include least-privilege credentials, allowlisted tools, authorization checks, confirmation before consequential actions, audit logs, rate limits, and protection of sensitive data in tool results. The test also addresses how tools should expose only the information needed for the next conversational step.
Recommended Preparation
Review the tool-calling documentation for a language model API and trace a complete request lifecycle: user message, model tool request, application validation, function execution, tool result, and final model response. Practice writing concise tool descriptions and parameter schemas for realistic tasks such as order lookup, calendar availability, knowledge-base search, and support-ticket creation.
Study API reliability patterns, particularly idempotency keys, bounded retries, timeout handling, correlation identifiers, and structured errors. Practice identifying which controls belong in the model prompt and which must be enforced by application code. Finally, examine how authorization and user confirmation should be applied when a chatbot can access customer records or perform actions that change external systems.