FEFreeExamDumps.in

CCA-F Practice Questions — Page 8

Question 71

Open question ↗

Scenario: Agent Architecture and Orchestration

A coordinator must guarantee that `verify_identity` always runs and succeeds before `process_payment` can be called. The team is debating whether to enforce ordering via the system prompt or programmatically.

Which approach gives a deterministic guarantee?

  • A.A detailed system prompt that explains the required order with examples.
  • B.A programmatic precondition that blocks `process_payment` until `verify_identity` has returned a verified result.
  • C.Setting `tool_choice: "any"` so the model must always call a tool first.
  • D.Listing `verify_identity` before `process_payment` in `allowed_tools`.

Question 72

Open question ↗

Scenario: Agent Architecture and Orchestration

A coordinator prompt for a research system reads: "Step 1: call search_web exactly once. Step 2: call analyze_doc exactly once. Step 3: write report." In practice, complex topics need several searches and some need no document analysis at all, and report quality suffers.

How should the coordinator prompt be written instead?

  • A.Keep the rigid step list but add "repeat if necessary" to each step.
  • B.Express goals and quality criteria (e.g., complete coverage with citations) and let the coordinator choose how many subagents to invoke.
  • C.Hard-code five searches and three analyses to cover the worst case.
  • D.Remove the coordinator and let a single agent do everything in one turn.

Question 73

Open question ↗

Scenario: Agent Architecture and Orchestration

A multi-agent system aborts the entire workflow whenever any single subagent fails, discarding all the work other subagents already completed successfully.

Which error-handling principle is being violated?

  • A.Subagents should retry indefinitely until they succeed.
  • B.The system should continue with partial results and annotate the gaps rather than discard everything on one failure.
  • C.All errors should be silently suppressed so the workflow always completes.
  • D.Each subagent should escalate every error directly to the end user.

Question 74

Open question ↗

Scenario: Agent Architecture and Orchestration

On escalation, a support agent passes only the line "Customer is upset, please help" to the human operator. The operator has no access to the conversation transcript and must re-gather everything.

What should the handoff contain?

  • A.A self-contained structured summary: customer ID, issue, order ID, actions already taken, and recommended action.
  • B.A link to the raw model logs so the operator can read the full reasoning.
  • C.Just the customer's most recent message, to keep the handoff short.
  • D.The agent's self-rated confidence score for the case.

Question 75

Open question ↗

Scenario: Agent Architecture and Orchestration

A subagent doing web search hits a transient 503 from one provider. The team is deciding where retry logic should live.

What is the recommended pattern?

  • A.Retry indefinitely inside the subagent until the provider responds.
  • B.Do 1–2 local retries in the subagent; if still failing, propagate a structured error with partial results to the coordinator.
  • C.Immediately propagate every error to the coordinator with no local retry.
  • D.Suppress the error and return an empty result marked as success.

Question 76

Open question ↗

Scenario: Agent Architecture and Orchestration

A synthesis subagent occasionally needs to confirm a single date or name while merging results. Today it hands control back to the coordinator, which calls the search agent and re-runs synthesis—adding 2–3 round trips. About 85% of these checks are trivial fact lookups; 15% need deeper investigation.

What design reduces overhead while preserving reliability?

  • A.Give the synthesis agent full access to every web-search tool.
  • B.Give the synthesis agent a narrow `verify_fact` tool for simple checks, and keep routing complex verification through the coordinator.
  • C.Batch all verification needs and send them to the coordinator only at the very end.
  • D.Pre-cache large amounts of source context speculatively around every claim.

Question 77

Open question ↗

Scenario: Tool Design and MCP Integration

Two tools have nearly identical descriptions: `analyze_content` ("Analyzes content and returns insights") and `analyze_document` ("Analyzes a document and returns insights"). The agent frequently picks the wrong one.

What is the most effective first fix?

  • A.Set `tool_choice: "any"` so the model is forced to call one of them.
  • B.Rewrite the descriptions to clearly distinguish purpose, inputs, and when to use each instead of the other.
  • C.Add a routing classifier in front of the agent to pick the tool.
  • D.Lower the temperature to make selection more deterministic.

Question 78

Open question ↗

Scenario: Tool Design and MCP Integration

A single "do-everything" agent has been given 18 tools spanning search, refunds, file editing, and analytics. Its tool-selection accuracy has dropped noticeably compared with earlier versions that had 4–5 tools each.

What is the most likely cause and fix?

  • A.Too many tools degrade selection reliability; scope each agent to only the tools relevant to its role.
  • B.The model simply needs a longer system prompt listing all 18 tools in detail.
  • C.Tool count has no effect on selection; the issue must be the model version.
  • D.Setting `tool_choice: "auto"` is the problem; forcing `"any"` will fix accuracy.

Question 79

Open question ↗

Scenario: Tool Design and MCP Integration

You are extracting metadata and must guarantee that the model returns a tool call producing structured output rather than a free-text answer, but you are fine with the model choosing among several extraction tools.

Which `tool_choice` setting fits?

  • A.`{"type": "auto"}`
  • B.`{"type": "any"}`
  • C.`{"type": "tool", "name": "extract_metadata"}`
  • D.Omit `tool_choice` entirely.

Question 80

Open question ↗

Scenario: Tool Design and MCP Integration

An extraction pipeline must always run `extract_metadata` as the very first tool call before any enrichment step, with no exceptions.

How do you guarantee that specific first action?

  • A.`tool_choice: {"type": "tool", "name": "extract_metadata"}` for the first request.
  • B.`tool_choice: "any"` and hope the model picks it first.
  • C.List `extract_metadata` first in the `tools` array.
  • D.`tool_choice: "auto"` with a strong system-prompt hint.