Build AI agents: free practice questions

When an agent Run reaches `requires_action` status (typically due to a tool call that needs external input), execution is paused and the caller must submit tool outputs to resume. By modeling the compliance approval as a tool call whose output requires human review, you can pause the Run at that point, surface the pending action to the compliance officer, and only resume by calling the submit tool outputs API after approval — implementing a true human-in-the-loop gate. Setting `max_completion_tokens` to zero does not pause the Run in a reviewable state; it would cause the Run to fail or produce empty output. Treating a submit_tool_outputs action as a manual step is close but option C specifically describes the correct full flow: polling for `requires_action` then submitting approval — option B is incomplete because it omits polling and the approval gate logic. A second interceptor agent adds complexity, has no native pause mechanism, and does not guarantee the primary Run is actually paused.

Enabling diagnostic logging on the Azure AI Foundry project with a diagnostic setting pointing to a Log Analytics workspace is correct. Azure AI Foundry integrates with Azure Monitor through Azure Diagnostics settings, allowing teams to stream agent telemetry—including Run traces, tool call details, and token metrics—directly into Log Analytics for querying with KQL and building dashboards in Azure Monitor. Instrumenting a custom Semantic Kernel logging plugin to Blob Storage is a partial, non-standard approach that bypasses the platform's native observability integration and requires custom parsing. Polling the REST API on a schedule is fragile, introduces latency, and is not a scalable observability strategy for production workloads. Injecting a TraceWriter into the system prompt is not a real pattern—system prompts are natural language instructions to the model, not configuration for telemetry infrastructure.

Submitting tool outputs via the submit_tool_outputs endpoint is correct. When a Run reaches `requires_action` status, the application must call the submit tool outputs API, providing the tool call ID returned by the model and the result of executing the function. This is the designed handshake that allows the Run to proceed. Deleting and recreating the Run is incorrect because that would lose conversation context and is not the intended workflow. Adding the tool output as a new user message is incorrect; that would corrupt the conversation structure — tool outputs must be submitted through the dedicated tool output mechanism. Updating the Agent definition is incorrect; the Agent definition is for static configuration such as instructions and available tools, not for passing runtime function results.

The `tool_resources` parameter in the Foundry Agent Service create/update agent API is used to attach resource objects — specifically vector store IDs for the File Search tool and file IDs for the Code Interpreter tool — to the agent. This tells the service which pre-created indexes or files the tools should operate on. It does not specify Azure service resource IDs for connectivity (that is handled via connections in the project). It does not control compute allocation for Code Interpreter sandboxes (this is managed by the service). It does not define managed identity permissions for custom functions (those are governed by Azure RBAC and the function endpoint's own authorization).

File Search is the correct choice. It enables the agent to perform semantic/vector-based retrieval over uploaded files (including PDFs), returning grounded, cited answers to free-form questions without manual chunking. Code Interpreter is designed for executing code and analyzing structured data—it is not optimized for semantic search over large prose documents. A custom function tool would require building and maintaining a separate retrieval service, adding unnecessary complexity. An OpenAPI tool integrates with external HTTP APIs using an OpenAPI schema; it does not provide built-in document retrieval semantics.