Anthropic Claude Certified Developer-Foundations - CCDV-F Exam Practice Test

A. Truncate every response to the first few characters, validate that the truncated output matches the expected JSON structure, and log any mismatches for review.

B. Restart the application after every turn and monitor whether the format remains consistent across subsequent interactions.

C. Identify the failure mode as format drift, examine how the conversation context evolves over turns, and address the drift through context management or output validation.

D. Switch to a smaller Claude model and re-test the application to determine whether the drift persists across conversation turns.

A. A memory pattern that stores the entire research history in advance, before any subtask begins execution.

B. A single tool-use loop that includes every tool the agent might need across all subtasks.

C. An orchestrator and subagent pattern with specialized subagents assigned to each subtask.

D. A context-window pruning pattern that drops each subtask's content after the agent moves on.

A. Cross-site scripting and SQL injection, because these traditional web application threats apply with equal weight to any application that uses Claude in any way.

B. Prompt injection, data leakage from prompts or context, jailbreak attempts, and unsafe model output that bypasses application controls.

C. Network-level denial of service and physical infrastructure attacks, because these categories cover the threats most likely to affect any Claude application in production.

D. Supply chain attacks on the Claude SDK because the SDK itself is the only point of vulnerability that a Claude application introduces beyond traditional web application threats.

A. Truncate the conversation so the model sees only the most recent turn during each subsequent response.

B. Apply compaction to summarize older portions of the conversation so the gist remains while the specifics carry less weight.

C. Reset the conversation after every turn so the model loses all prior turns when generating a response.

D. Increase the context window size so all turns of the conversation remain visible to the model in full detail.

A. Examine production traces to identify whether the issue is hallucination by the model, context loss, prompt injection, or another failure mode before recommending a fix.

B. Apply a retrieval-augmented generation pattern to ground the responses in source content before any further investigation of the production traces.

C. Replace the current model with a larger one to reduce the chance of hallucination, on the grounds that larger models tend to hallucinate less in typical applications.

D. Add a system prompt instruction telling the model not to invent information, on the grounds that prompt-level instructions are the fastest fix for hallucination concerns.

A. Use a different LLM provider's SDK and translate the responses into Claude's API shape so the application can switch providers in the future.

B. Call the REST API directly with raw HTTP requests so the application avoids the SDK's abstraction between the application code and the API.

C. Use the Claude Python SDK and let it handle authentication, retries, and response parsing through its standard documented patterns for Python integrations.

D. Skip the SDK and embed Claude calls in shell commands invoked from Python, so that the application runs the calls outside the main Python process.

A. Tell users that validation failures are unavoidable and instruct them to perform manual accuracy checks before relying on outputs.

B. Pass validation failures directly to downstream systems and let each downstream system decide how to handle the malformed output.

C. Disable output validation until the underlying cause of validation failures has been identified and addressed in a future release.

D. Treat validation failures as a recognized error path that triggers retry, repair, or fallback logic before failing the user request.

A. Agree with the teammate because tool names are sufficient for the model to choose the right tool on every request the agent handles.

B. Explain that the model uses the tool description to decide when to call the tool, and descriptions disambiguate cases where the tool name is not enough.

C. Suggest replacing the descriptions with example calls embedded in the tool definition, treating example calls as a complete substitute for the prose description.

D. Suggest moving the descriptions out of the tool definition and into a separate documentation file the team maintains so the tool definitions stay short and the descriptions remain available.

A. Use distinct keys for each environment so a compromise in one environment does not affect the others during normal operation across the application's lifecycle.

B. Rotate the same key across environments at random intervals on the grounds that random rotation provides isolation between environments without requiring distinct keys.

C. Use the same key across all environments for simplicity and treat the propagation of any compromise as a known operational tradeoff for the team's key management approach.

D. Use a single development key everywhere on the grounds that production keys are too risky to deploy across the application's three environments during normal operation.

A. Prompt caching, which speeds up the cost profile of future requests and does not affect the delivery timing of the first response.

B. Structured JSON output, which delivers responses only after the model has finalized the JSON shape across the entire response.

C. The Batch API, which delivers full responses after a delay suitable for non-interactive workloads.

D. Streaming responses, which deliver tokens incrementally as the model generates them so users see output progressively.