Question 670
A team maintaining a Python Claude API backend has two different engineering requirements: (1) apply consistent reviewed tool permissions to an entire Claude Code repository team, and (2) reproduce production behavior across deploys instead of silently adopting a new model release. Which TWO independent changes directly meet them? Choose TWO.
- Use the user's own Claude Code settings file.
- Pin a supported explicit model version where stability is required.
- Track enabled plugins and their compatible versions as release dependencies.
- Store team settings in the project's shared configuration file.
- Enforce per-tenant token and request budgets in the backend.
Explanation
Correct Answers
B:To reproduce production behavior across deploys instead of silently adopting a new model release as part of a design that must also apply consistent reviewed tool permissions to an entire Claude Code repository team, explicit version selection limits unexpected changes; a release process should evaluate upgrades and retirement timelines deliberately; this is why the action is one of the two required selections.
D:To apply consistent reviewed tool permissions to an entire Claude Code repository team as part of a design that must also reproduce production behavior across deploys instead of silently adopting a new model release, repository-level settings can be versioned and reviewed, unlike private preferences that vary across each developer's machine; this is why the action is one of the two required selections.
Incorrect Answers
A:This alternative serves the separate goal to customize one developer's default Claude Code behavior across local projects because user settings are scoped to one person, so they should not be relied on for enforcing team-wide policy; the required outcomes are instead to apply consistent reviewed tool permissions to an entire Claude Code repository team and to reproduce production behavior across deploys instead of silently adopting a new model release.
C:This alternative serves the separate goal to avoid a plugin update unexpectedly changing tools available to a coding agent because plugin capability changes can affect tool names and permissions, so production configuration should be reproducible and reviewed; the required outcomes are instead to apply consistent reviewed tool permissions to an entire Claude Code repository team and to reproduce production behavior across deploys instead of silently adopting a new model release.
E:This alternative serves the separate goal to keep one customer from consuming an entire shared inference budget because application quotas isolate noisy tenants and control spending; provider-wide rate limits alone do not guarantee fairness between customers; the required outcomes are instead to apply consistent reviewed tool permissions to an entire Claude Code repository team and to reproduce production behavior across deploys instead of silently adopting a new model release.
Question 671
A design review for a model selection trial asks for support to allow supported models to vary internal reasoning effort according to task difficulty, together with a way to teach a distinctive output format with one representative example. Select the two appropriate changes. Choose TWO.
- Iterate the SDK or HTTP event stream until its terminal event.
- Revalidate prompts, schemas, safety constraints, and tools after model migration.
- Select an available effort level after measuring task performance.
- Configure supported adaptive thinking behavior rather than forcing a fixed pattern.
- Provide one accurate input-output demonstration.
Explanation
Correct Answers
D:To allow supported models to vary internal reasoning effort according to task difficulty as part of a design that must also teach a distinctive output format with one representative example, adaptive reasoning responds to problem complexity, but supported modes differ by model and release and should not be assumed universal; this is why the action is one of the two required selections.
E:To teach a distinctive output format with one representative example as part of a design that must also allow supported models to vary internal reasoning effort according to task difficulty, a single example illustrates structure and conventions, but it may not reveal how to handle difficult edge cases; this is why the action is one of the two required selections.
Incorrect Answers
A:This alternative serves the separate goal to process partial model output as it is delivered over a streaming connection because streaming clients must handle both content deltas and final status rather than assuming the first chunk is the complete response; the required outcomes are instead to allow supported models to vary internal reasoning effort according to task difficulty and to teach a distinctive output format with one representative example.
B:This alternative serves the separate goal to avoid assuming a model-version change preserves tool output and writing behavior because new releases can change response distributions or supported behaviors, making a model upgrade a real software change; the required outcomes are instead to allow supported models to vary internal reasoning effort according to task difficulty and to teach a distinctive output format with one representative example.
C:This alternative serves the separate goal to trade answer latency and computation against solution quality on supported models because reasoning effort is an inference-time tradeoff; lowering it may reduce costs or latency while harming difficult-case accuracy; the required outcomes are instead to allow supported models to vary internal reasoning effort according to task difficulty and to teach a distinctive output format with one representative example.
Question 672
In a context-rich document assistant, a release requirement is to analyze a large source set without exposing every raw document to the parent agent. Which implementation addresses that specific requirement?
- State explicit priority and conflict resolution for the task constraints.
- Have a scoped subagent process sources and return a concise verified handoff.
- Revise one prompt component at a time against a regression evaluation set.
- Retrieve authoritative policy passages and cite those used.
Explanation
Correct Answer
B:To analyze a large source set without exposing every raw document to the parent agent in this context-rich document assistant, separate contexts let specialists work deeply while the coordinator sees only claims and references needed for the final decision; the proposed implementation directly supplies that capability.
Incorrect Answers
A:The proposed alternative helps to prioritize mandatory legal requirements over optional stylistic preferences because without a priority rule, competing directives can create ambiguous outputs; hard requirements should guide how optional preferences are applied; that mechanism does not meet the different requirement to analyze a large source set without exposing every raw document to the parent agent.
C:The proposed alternative helps to improve a prompt while preserving performance on previously solved cases because controlled prompt iteration distinguishes genuine quality gains from regressions or overfitting to one attractive example; that mechanism does not meet the different requirement to analyze a large source set without exposing every raw document to the parent agent.
D:The proposed alternative helps to answer a company-policy question from current approved records instead of relying on general knowledge because retrieval supplies task-specific facts that pretraining may not include, while source attribution lets the application verify the response; that mechanism does not meet the different requirement to analyze a large source set without exposing every raw document to the parent agent.
Question 673
For a shared tool gateway for Claude, a design is acceptable only if it can process several requested independent tools without mixing their outputs while also being able to connect a Claude tool request to an internal service whose logic the application owns. Which paired implementation is sufficient?
- Perform both steps: (1) Correlate each tool result with its matching tool_use_id; (2) implement a client-side handler and return its result with the tool-use identifier.
- Perform both steps: (1) Use the MCP stdio transport with properly framed protocol messages; (2) implement a client-side handler and return its result with the tool-use identifier.
- Perform both steps: (1) Use the MCP stdio transport with properly framed protocol messages; (2) define a descriptive tool name and validated input schema.
- Perform both steps: (1) Correlate each tool result with its matching tool_use_id; (2) define a descriptive tool name and validated input schema.
Explanation
Correct Answer
A:The first change enables the team to process several requested independent tools without mixing their outputs because multi-tool responses can interleave operations, so identifier-based correlation is essential for safe continuation; the second addresses the separate need to connect a Claude tool request to an internal service whose logic the application owns because a declared client tool does not execute itself; application code runs it and feeds results back into the conversation.
Incorrect Answers
B:The second step meets the requirement to connect a Claude tool request to an internal service whose logic the application owns because a declared client tool does not execute itself; application code runs it and feeds results back into the conversation; the first step would only help to launch a local MCP server as a subprocess managed by a compatible client because standard input and output suit local process integration, where protocol output must not be mixed with arbitrary logging noise, leaving the separate requirement to process several requested independent tools without mixing their outputs unmet.
C:Both steps target different results—launch a local MCP server as a subprocess managed by a compatible client and expose a business lookup with predictable typed arguments—because standard input and output suit local process integration, where protocol output must not be mixed with arbitrary logging noise and well-described schemas help Claude select a capability and emit usable arguments; the backend still validates authorization and meaning; the required capabilities are instead to process several requested independent tools without mixing their outputs and to connect a Claude tool request to an internal service whose logic the application owns.
D:The first step is appropriate for the requirement to process several requested independent tools without mixing their outputs because multi-tool responses can interleave operations, so identifier-based correlation is essential for safe continuation; the other step instead aims to expose a business lookup with predictable typed arguments because well-described schemas help Claude select a capability and emit usable arguments; the backend still validates authorization and meaning, which does not meet the second requirement to connect a Claude tool request to an internal service whose logic the application owns.
Question 674
A design review for a high-volume Messages API gateway asks for support to identify damage that could result from adversarial documents or tool access, together with a way to turn a support desk's responsiveness expectation into an engineering objective. Select the two appropriate changes. Choose TWO.
- Treat stop_reason end_turn as a completed assistant turn.
- Create a threat model covering trust boundaries, permissions, and misuse paths.
- Set an end-to-end latency target and a measurement method.
- List upstream systems with permissions, quotas, and failure behavior.
- Configure an available server-side tool with the required API definition.
Explanation
Correct Answers
B:To identify damage that could result from adversarial documents or tool access as part of a design that must also turn a support desk's responsiveness expectation into an engineering objective, aI application risk includes injected content and powerful downstream functions; those flows need explicit mitigations before implementation; this is why the action is one of the two required selections.
C:To turn a support desk's responsiveness expectation into an engineering objective as part of a design that must also identify damage that could result from adversarial documents or tool access, a latency target is a nonfunctional requirement; model choice, network overhead, tool calls, and rendering all affect whether the service meets it; this is why the action is one of the two required selections.
Incorrect Answers
A:This alternative serves the separate goal to detect that the assistant has naturally finished a response because the completion signal is explicit in the response metadata; it should not be inferred solely from a sentence ending in punctuation; the required outcomes are instead to identify damage that could result from adversarial documents or tool access and to turn a support desk's responsiveness expectation into an engineering objective.
D:This alternative serves the separate goal to identify which search services and internal systems are critical to the workflow because a dependency inventory exposes the operational and authorization constraints that a purely model-centered design misses; the required outcomes are instead to identify damage that could result from adversarial documents or tool access and to turn a support desk's responsiveness expectation into an engineering objective.
E:This alternative serves the separate goal to use provider-hosted search or execution without implementing the tool body locally because provider-executed tools are handled in Anthropic's environment and differ from client functions that the caller must run; the required outcomes are instead to identify damage that could result from adversarial documents or tool access and to turn a support desk's responsiveness expectation into an engineering objective.
Question 675
A design review for a tool-using incident triage agent asks for support to require human approval before an autonomous task crosses a high-impact decision boundary, together with a way to keep a large source-reading task from crowding the coordinator's working context. Select the two appropriate changes. Choose TWO.
- Insert an approval gate before the high-risk action.
- Delegate source analysis to a subagent and return a concise result.
- Inspect the tool_use block, invoke the authorized handler, and return tool_result.
- Record stage-level traces and intermediate validation outcomes.
- Chain dependent operations serially with explicit input and output contracts.
Explanation
Correct Answers
A:To require human approval before an autonomous task crosses a high-impact decision boundary as part of a design that must also keep a large source-reading task from crowding the coordinator's working context, human-in-the-loop control interrupts execution at a defined boundary; an agent's natural-language assurance cannot substitute for authorization; this is why the action is one of the two required selections.
B:To keep a large source-reading task from crowding the coordinator's working context as part of a design that must also require human approval before an autonomous task crosses a high-impact decision boundary, isolated workspaces let a specialist process extensive material while the parent receives only task-relevant evidence instead of every intermediate token; this is why the action is one of the two required selections.
Incorrect Answers
C:This alternative serves the separate goal to execute a requested internal business function inside the application's trusted process because client tools are executed by the application, not by merely declaring a schema; every returned result must be matched to its tool-use identifier; the required outcomes are instead to require human approval before an autonomous task crosses a high-impact decision boundary and to keep a large source-reading task from crowding the coordinator's working context.
D:This alternative serves the separate goal to pinpoint which known workflow stage failed when the final artifact is incomplete because deterministic stages are operationally useful when their boundaries are observable; a final answer alone cannot reveal which step produced the fault; the required outcomes are instead to require human approval before an autonomous task crosses a high-impact decision boundary and to keep a large source-reading task from crowding the coordinator's working context.
E:This alternative serves the separate goal to ensure each later tool receives a validated artifact from its preceding stage because later steps cannot safely start until upstream data exists; passing validated outputs prevents races and malformed downstream input; the required outcomes are instead to require human approval before an autonomous task crosses a high-impact decision boundary and to keep a large source-reading task from crowding the coordinator's working context.
Question 676
For a secure tool-execution boundary, the missing capability is to investigate user actions without leaking API keys into operational records. What should the developers implement?
- Redact secrets and protect log access with retention controls.
- Retrieve keys through a managed secrets facility at runtime.
- Enforce a pre-tool hook that denies the disallowed call.
- Apply an input-side policy gate and route flagged cases to a safe path.
Explanation
Correct Answer
A:To investigate user actions without leaking API keys into operational records in this secure tool-execution boundary, logs support accountability only when access and retention are restricted; copying all prompt content into logs creates another exposure surface; the proposed implementation directly supplies that capability.
Incorrect Answers
B:The proposed alternative helps to keep Claude API credentials out of repository source and deployment images because central secret handling improves rotation and auditing and avoids accidental credential disclosure through source control; that mechanism does not meet the different requirement to investigate user actions without leaking API keys into operational records.
C:The proposed alternative helps to stop a coding agent from running commands that match a prohibited destructive pattern because a deterministic hook checks the proposed command before execution, providing a stronger control than asking the agent to self-police; that mechanism does not meet the different requirement to investigate user actions without leaking API keys into operational records.
D:The proposed alternative helps to identify prohibited requests before an automated agent invokes business tools because early filtering reduces inappropriate work, but later checks remain necessary because tool outputs can also introduce risks; that mechanism does not meet the different requirement to investigate user actions without leaking API keys into operational records.
Question 677
An end-to-end test for a multi-tenant support application checks whether it can handle a safety refusal as a distinct application outcome and map out-of-order batch outputs back to source records reliably. Which plan should be approved to satisfy both checks?
- Perform both steps: (1) Detect refusal and apply the product's allowed fallback or messaging policy; (2) assign and preserve unique custom identifiers for batched requests.
- Perform both steps: (1) Use stop_sequences and inspect the returned stop_sequence field; (2) cache an eligible unchanged prefix with cache_control.
- Perform both steps: (1) Use stop_sequences and inspect the returned stop_sequence field; (2) assign and preserve unique custom identifiers for batched requests.
- Perform both steps: (1) Detect refusal and apply the product's allowed fallback or messaging policy; (2) cache an eligible unchanged prefix with cache_control.
Explanation
Correct Answer
A:The first change enables the team to handle a safety refusal as a distinct application outcome because a refusal is not the same as a network failure; blindly retrying identical input may not produce an appropriate user experience; the second addresses the separate need to map out-of-order batch outputs back to source records reliably because asynchronous results need stable correlation keys; relying on processing order can attach an output to the wrong input.
Incorrect Answers
B:Both steps target different results—enforce a caller-defined textual boundary on generation and reuse a long stable instruction or document prefix over repeated requests—because a stop sequence halts generation at a specified marker, distinct from natural completion and separate from a JSON schema validator and prompt caching reduces repeated input processing when the prefix matches; changes before the cache boundary can prevent a hit; the required capabilities are instead to handle a safety refusal as a distinct application outcome and to map out-of-order batch outputs back to source records reliably.
C:The second step meets the requirement to map out-of-order batch outputs back to source records reliably because asynchronous results need stable correlation keys; relying on processing order can attach an output to the wrong input; the first step would only help to enforce a caller-defined textual boundary on generation because a stop sequence halts generation at a specified marker, distinct from natural completion and separate from a JSON schema validator, leaving the separate requirement to handle a safety refusal as a distinct application outcome unmet.
D:The first step is appropriate for the requirement to handle a safety refusal as a distinct application outcome because a refusal is not the same as a network failure; blindly retrying identical input may not produce an appropriate user experience; the other step instead aims to reuse a long stable instruction or document prefix over repeated requests because prompt caching reduces repeated input processing when the prefix matches; changes before the cache boundary can prevent a hit, which does not meet the second requirement to map out-of-order batch outputs back to source records reliably.
Question 678
The technical owner of a Claude inference cost review has to preserve critical service during a temporary primary-model problem. A separate requirement is to avoid assuming a model-version change preserves tool output and writing behavior. Which two-part remediation covers both?
- Perform both steps: (1) Serialize the documented nested request shape rather than flattening it; (2) budget total context for instructions, history, tools, and output.
- Perform both steps: (1) Design an approved fallback model path with compatibility checks; (2) budget total context for instructions, history, tools, and output.
- Perform both steps: (1) Serialize the documented nested request shape rather than flattening it; (2) revalidate prompts, schemas, safety constraints, and tools after model migration.
- Perform both steps: (1) Design an approved fallback model path with compatibility checks; (2) revalidate prompts, schemas, safety constraints, and tools after model migration.
Explanation
Correct Answer
D:The first change enables the team to preserve critical service during a temporary primary-model problem because fallbacks improve resilience only when downstream schemas, cost limits, and policy controls work with the alternate model; the second addresses the separate need to avoid assuming a model-version change preserves tool output and writing behavior because new releases can change response distributions or supported behaviors, making a model upgrade a real software change.
Incorrect Answers
A:Both steps target different results—preserve arrays of typed content blocks when sending multimodal input and know why a long prompt plus a proposed response may exceed the model's accepted context—because typed content and role fields carry protocol meaning that is lost if developers simply concatenate them into plain text and a context window bounds what can be processed in one request; large tool results consume it along with user messages; the required capabilities are instead to preserve critical service during a temporary primary-model problem and to avoid assuming a model-version change preserves tool output and writing behavior.
B:The first step is appropriate for the requirement to preserve critical service during a temporary primary-model problem because fallbacks improve resilience only when downstream schemas, cost limits, and policy controls work with the alternate model; the other step instead aims to know why a long prompt plus a proposed response may exceed the model's accepted context because a context window bounds what can be processed in one request; large tool results consume it along with user messages, which does not meet the second requirement to avoid assuming a model-version change preserves tool output and writing behavior.
C:The second step meets the requirement to avoid assuming a model-version change preserves tool output and writing behavior because new releases can change response distributions or supported behaviors, making a model upgrade a real software change; the first step would only help to preserve arrays of typed content blocks when sending multimodal input because typed content and role fields carry protocol meaning that is lost if developers simply concatenate them into plain text, leaving the separate requirement to preserve critical service during a temporary primary-model problem unmet.
Question 679
The team responsible for an autonomous repository maintenance job needs to stop iterative tool use when a success condition or resource budget is reached. What is the most direct engineering change?
- Configure the appropriate Anthropic-executed server tool.
- Enforce explicit completion checks and a maximum iteration budget.
- Classify the request and dispatch to an appropriate specialized workflow.
- Apply limits for steps, elapsed time, and token or tool expenditure.
Explanation
Correct Answer
B:To stop iterative tool use when a success condition or resource budget is reached in this autonomous repository maintenance job, bounded loops avoid unending tool calls and uncontrolled spend while making incomplete work visible instead of silently continuing; the proposed implementation directly supplies that capability.
Incorrect Answers
A:The proposed alternative helps to use a Claude platform-provided capability without operating a local handler because server tools execute on the provider side and have distinct authorization and runtime semantics from client tools the application dispatches; that mechanism does not meet the different requirement to stop iterative tool use when a success condition or resource budget is reached.
C:The proposed alternative helps to send distinct request types to specialized handlers while keeping the routing rule inspectable because routing separates distinct pathways without making the entire task autonomously planned; the classification decision must be observable and tested; that mechanism does not meet the different requirement to stop iterative tool use when a success condition or resource budget is reached.
D:The proposed alternative helps to impose predictable maximum cost and wall-clock exposure on autonomous execution because runtime budgets are external constraints on agent activity and must be enforced by the host, not inferred from a cooperative natural-language instruction; that mechanism does not meet the different requirement to stop iterative tool use when a success condition or resource budget is reached.
Question 680
Two acceptance tests block a production inference application: one requires the ability to preserve behavior while replacing a fragile parser in an AI integration; the other requires the ability to detect a changed tool response field before it breaks downstream code. Which plan clears both tests?
- Perform both steps: (1) Use the API's documented POST request with a JSON body; (2) log status, timing, and correlation ids with sensitive fields redacted.
- Perform both steps: (1) Refactor in bounded steps under existing tests; (2) log status, timing, and correlation ids with sensitive fields redacted.
- Perform both steps: (1) Refactor in bounded steps under existing tests; (2) add automated consumer-provider contract tests.
- Perform both steps: (1) Use the API's documented POST request with a JSON body; (2) add automated consumer-provider contract tests.
Explanation
Correct Answer
C:The first change enables the team to preserve behavior while replacing a fragile parser in an AI integration because small changes constrain regression scope; a wholesale rewrite makes it harder to isolate whether model or parsing behavior changed; the second addresses the separate need to detect a changed tool response field before it breaks downstream code because schema contracts make integration changes visible and prevent a renamed field from silently corrupting later model context.
Incorrect Answers
A:Both steps target different results—send a model inference request with the operation expected by the REST endpoint and debug a failed request without exposing the user's confidential document—because a state-changing inference operation is submitted as an HTTP request with an explicit method and payload, not inferred from a browser URL alone and operational records can identify a failure path without retaining raw secrets or protected prompt content; the required capabilities are instead to preserve behavior while replacing a fragile parser in an AI integration and to detect a changed tool response field before it breaks downstream code.
B:The first step is appropriate for the requirement to preserve behavior while replacing a fragile parser in an AI integration because small changes constrain regression scope; a wholesale rewrite makes it harder to isolate whether model or parsing behavior changed; the other step instead aims to debug a failed request without exposing the user's confidential document because operational records can identify a failure path without retaining raw secrets or protected prompt content, which does not meet the second requirement to detect a changed tool response field before it breaks downstream code.
D:The second step meets the requirement to detect a changed tool response field before it breaks downstream code because schema contracts make integration changes visible and prevent a renamed field from silently corrupting later model context; the first step would only help to send a model inference request with the operation expected by the REST endpoint because a state-changing inference operation is submitted as an HTTP request with an explicit method and payload, not inferred from a browser URL alone, leaving the separate requirement to preserve behavior while replacing a fragile parser in an AI integration unmet.