Question 100
A team maintaining an incident involving failed Claude completions has two different engineering requirements: (1) distinguish a syntactically valid but low-quality answer from a transport outage, and (2) explain why an agent loop is stuck after a model requested an external function. Which TWO independent changes directly meet them? Choose TWO.
- Examine correlated request, tool, and transformation traces stage by stage.
- Check that each tool_use identifier receives one matching tool_result.
- Inspect the HTTP 429 status and rate-limit metadata before retrying.
- Compare versioned prompts and representative eval cases against model responses.
- Read the 400 error details and validate the request payload.
Explanation
Correct Answers
B:To explain why an agent loop is stuck after a model requested an external function as part of a design that must also distinguish a syntactically valid but low-quality answer from a transport outage, an unmatched tool call leaves a protocol-level continuation gap, even if the handler actually ran successfully; this is why the action is one of the two required selections.
D:To distinguish a syntactically valid but low-quality answer from a transport outage as part of a design that must also explain why an agent loop is stuck after a model requested an external function, a successful HTTP call only proves the protocol worked; quality regressions require output-level evidence and controlled comparisons; this is why the action is one of the two required selections.
Incorrect Answers
A:This alternative serves the separate goal to locate which component of a multi-tool answer introduced the wrong value because structured traces reveal whether retrieval, tool execution, postprocessing, or generation caused the defect; reading only the final text cannot isolate it; the required outcomes are instead to distinguish a syntactically valid but low-quality answer from a transport outage and to explain why an agent loop is stuck after a model requested an external function.
C:This alternative serves the separate goal to distinguish an upstream throttling response from a malformed model answer because a rate limit is a request-layer capacity problem; the response needs controlled pacing rather than a prompt rewrite; the required outcomes are instead to distinguish a syntactically valid but low-quality answer from a transport outage and to explain why an agent loop is stuck after a model requested an external function.
E:This alternative serves the separate goal to find a request schema error instead of repeatedly changing the system prompt because an invalid API field or incompatible parameter cannot be fixed by natural-language instructions within the malformed request; the required outcomes are instead to distinguish a syntactically valid but low-quality answer from a transport outage and to explain why an agent loop is stuck after a model requested an external function.
Question 101
A planned upgrade to a multi-tenant support application has two requirements that must both pass: reduce repeated TLS setup overhead for high-volume application traffic; trace exactly which prompt template was deployed with a code change. Identify the two direct implementations. Choose TWO.
- Log status, timing, and correlation IDs with sensitive fields redacted.
- Classify permanent client errors separately from transient failures.
- Reuse an appropriately configured HTTP client or SDK connection pool.
- Cache an eligible unchanged prefix with cache_control.
- Version prompt files alongside relevant application code.
Explanation
Correct Answers
C:To reduce repeated TLS setup overhead for high-volume application traffic as part of a design that must also trace exactly which prompt template was deployed with a code change, connection reuse lowers transport overhead but still requires limits and health handling for long-lived clients; this is why the action is one of the two required selections.
E:To trace exactly which prompt template was deployed with a code change as part of a design that must also reduce repeated TLS setup overhead for high-volume application traffic, prompts change software behavior and should be reviewed and diffed like code, rather than edited anonymously in production; this is why the action is one of the two required selections.
Incorrect Answers
A:This alternative serves the separate goal 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; the required outcomes are instead to reduce repeated TLS setup overhead for high-volume application traffic and to trace exactly which prompt template was deployed with a code change.
B:This alternative serves the separate goal to avoid retrying a malformed request that the API consistently rejects because a deterministic validation or authorization error requires fixing the request or credential, not another identical attempt; the required outcomes are instead to reduce repeated TLS setup overhead for high-volume application traffic and to trace exactly which prompt template was deployed with a code change.
D:This alternative serves the separate goal 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; the required outcomes are instead to reduce repeated TLS setup overhead for high-volume application traffic and to trace exactly which prompt template was deployed with a code change.
Question 102
In an internal agent-extension rollout, a release requirement is to execute a high-privilege enterprise action that must remain inside a corporate network. Which implementation addresses that specific requirement?
- Host a restricted client tool in the private application environment.
- Use the MCP stdio transport with properly framed protocol messages.
- Package the connector behind an MCP server.
- Publish a properly described MCP tool and its input contract.
Explanation
Correct Answer
A:To execute a high-privilege enterprise action that must remain inside a corporate network in this internal agent-extension rollout, a private handler provides local network and IAM controls that a generic provider-executed server tool may not possess; the proposed implementation directly supplies that capability.
Incorrect Answers
B:The proposed alternative helps 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; that mechanism does not meet the different requirement to execute a high-privilege enterprise action that must remain inside a corporate network.
C:The proposed alternative helps to share one integration with several compatible AI clients or agent runtimes because MCP standardizes capability exposure across clients, reducing duplicate integration logic compared with proprietary tool schemas everywhere; that mechanism does not meet the different requirement to execute a high-privilege enterprise action that must remain inside a corporate network.
D:The proposed alternative helps to let an MCP client invoke a business capability through a standard server interface because MCP tools represent callable operations with typed parameters; the server executes authorized behavior rather than only providing reference text; that mechanism does not meet the different requirement to execute a high-privilege enterprise action that must remain inside a corporate network.
Question 103
A change request for a latency-sensitive model gateway combines two needs: process partial model output as it is delivered over a streaming connection, and stop a no-longer-needed inference when the user's request is cancelled. Which implementation pair addresses both, not just one?
- Perform both steps: (1) Iterate the SDK or HTTP event stream until its terminal event; (2) propagate cancellation to the active network request when supported.
- Perform both steps: (1) Iterate the SDK or HTTP event stream until its terminal event; (2) check each response block's type before extracting its fields.
- Perform both steps: (1) Use a synchronous client with explicit timeout and error handling; (2) propagate cancellation to the active network request when supported.
- Perform both steps: (1) Use a synchronous client with explicit timeout and error handling; (2) check each response block's type before extracting its fields.
Explanation
Correct Answer
A:The first change enables the team 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 second addresses the separate need to stop a no-longer-needed inference when the user's request is cancelled because cancelling work can free local resources and avoid processing obsolete results, though already-incurred provider work may still be billed.
Incorrect Answers
B:The first step is appropriate for the requirement 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 other step instead aims to prevent a tool_use block from being displayed as if it were natural-language text because typed response objects can contain several block families; assuming every block has a text field causes logic errors, which does not meet the second requirement to stop a no-longer-needed inference when the user's request is cancelled.
C:The second step meets the requirement to stop a no-longer-needed inference when the user's request is cancelled because cancelling work can free local resources and avoid processing obsolete results, though already-incurred provider work may still be billed; the first step would only help to implement a low-volume command-line utility that handles one inference at a time because synchronous calls are simplest for strictly sequential work, where concurrency would add complexity without improving throughput, leaving the separate requirement to process partial model output as it is delivered over a streaming connection unmet.
D:Both steps target different results—implement a low-volume command-line utility that handles one inference at a time and prevent a tool_use block from being displayed as if it were natural-language text—because synchronous calls are simplest for strictly sequential work, where concurrency would add complexity without improving throughput and typed response objects can contain several block families; assuming every block has a text field causes logic errors; the required capabilities are instead to process partial model output as it is delivered over a streaming connection and to stop a no-longer-needed inference when the user's request is cancelled.
Question 104
A test for a production inference application fails because it cannot ensure a rebuilt agent uses the same tested SDK dependency set. Which change fixes the identified failure?
- Track enabled plugins and their compatible versions as release dependencies.
- Specify expected concurrency and enforce admission or queueing limits.
- Queue or reject excess requests using bounded admission control.
- Commit lockfiles and control library upgrades.
Explanation
Correct Answer
D:To ensure a rebuilt agent uses the same tested SDK dependency set in this production inference application, dependency pinning reduces environment drift; transitive SDK updates can change behavior even when prompt text is unchanged; the proposed implementation directly supplies that capability.
Incorrect Answers
A:The proposed alternative helps 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; that mechanism does not meet the different requirement to ensure a rebuilt agent uses the same tested SDK dependency set.
B:The proposed alternative helps to design for a documented burst of simultaneous requests without uncontrolled overload because capacity planning separates average traffic from peak demand and prevents upstream rate limits from becoming cascading application failures; that mechanism does not meet the different requirement to ensure a rebuilt agent uses the same tested SDK dependency set.
C:The proposed alternative helps to shield customer traffic when inference capacity is exhausted because controlled queuing protects system latency and prevents unbounded retries from overwhelming downstream provider quotas; that mechanism does not meet the different requirement to ensure a rebuilt agent uses the same tested SDK dependency set.
Question 105
An architecture review of a knowledge-grounded assistant asks how to require that a generated report have a fixed set of sections. Which action is the best fit?
- Name each mandatory field or section and provide formatting constraints.
- Validate parsed output against the application's declared schema.
- State explicit priority and conflict resolution for the task constraints.
- Recheck current plans against a stored objective and constraints.
Explanation
Correct Answer
A:To require that a generated report have a fixed set of sections in this knowledge-grounded assistant, concrete desired structure is easier to verify than a vague instruction to be professional or comprehensive; the proposed implementation directly supplies that capability.
Incorrect Answers
B:The proposed alternative helps to reject model output missing a field that a downstream service requires because parsing proves a string is legal JSON, but schema validation checks required keys, types, and structural constraints; that mechanism does not meet the different requirement to require that a generated report have a fixed set of sections.
C: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 require that a generated report have a fixed set of sections.
D:The proposed alternative helps to identify when a long agent discussion begins solving a different problem than the user approved because explicit objective comparison detects drift before tool actions diverge from the original task; extra history alone cannot guarantee alignment; that mechanism does not meet the different requirement to require that a generated report have a fixed set of sections.
Question 106
Two approval criteria are still open for an internal coding-assistant workflow: the team must stop old agent histories from burying the information needed for the current step and recover a multi-hour agent workflow after an infrastructure interruption. Which two implementation actions satisfy them? Choose TWO.
- Deploy the agent runner in the organization's own environment.
- Persist checkpoints for tool outputs and accepted decisions.
- Define structured inputs, expected outputs, and failure states for each delegate.
- Return a separate tool_result for every corresponding tool_use identifier.
- Retrieve only task-relevant memories into the active context.
Explanation
Correct Answers
B:To recover a multi-hour agent workflow after an infrastructure interruption as part of a design that must also stop old agent histories from burying the information needed for the current step, durable checkpoints let the orchestrator resume from a known stage rather than rerunning costly or side-effecting actions; this is why the action is one of the two required selections.
E:To stop old agent histories from burying the information needed for the current step as part of a design that must also recover a multi-hour agent workflow after an infrastructure interruption, selective retrieval improves signal-to-noise and keeps token use bounded while retaining access to older information when required; this is why the action is one of the two required selections.
Incorrect Answers
A:This alternative serves the separate goal to keep orchestration runtime and internal tool handlers inside company-managed infrastructure because self-hosting puts tool execution, networking, and operational controls under the application's responsibility, subject to its own deployment constraints; the required outcomes are instead to stop old agent histories from burying the information needed for the current step and to recover a multi-hour agent workflow after an infrastructure interruption.
C:This alternative serves the separate goal to prevent specialists from producing incompatible artifacts for the orchestrator because explicit delegation contracts make multi-agent integration testable; free-form handoffs often obscure whether a child finished its assigned task; the required outcomes are instead to stop old agent histories from burying the information needed for the current step and to recover a multi-hour agent workflow after an infrastructure interruption.
D:This alternative serves the separate goal to handle several independent tool_use blocks in one assistant message because parallel tool execution requires complete, correctly linked results; dropping one identifier leaves the model's outstanding request unresolved; the required outcomes are instead to stop old agent histories from burying the information needed for the current step and to recover a multi-hour agent workflow after an infrastructure interruption.
Question 107
An operational review of a hosted document-processing service reports distinct failures: it cannot preserve an understandable outcome when model generation cannot complete, and it cannot support Claude models accessed through different supported cloud offerings. Which two repairs are required? Choose TWO.
- Specify user-visible degraded responses and recovery paths.
- Include a typed image content block in the message.
- Define measurable answer-format and citation acceptance criteria.
- Specify expected concurrency and enforce admission or queueing limits.
- Use the provider-specific endpoint, authentication, and request conventions.
Explanation
Correct Answers
A:To preserve an understandable outcome when model generation cannot complete as part of a design that must also support Claude models accessed through different supported cloud offerings, a fallback requirement defines what customers see when an API call times out or is refused rather than assuming every inference will succeed; this is why the action is one of the two required selections.
E:To support Claude models accessed through different supported cloud offerings as part of a design that must also preserve an understandable outcome when model generation cannot complete, Claude API, Bedrock, and Vertex-style integrations are not interchangeable URLs; host authentication and service wrappers differ; this is why the action is one of the two required selections.
Incorrect Answers
B:This alternative serves the separate goal to ask Claude to interpret a supported image supplied as part of a request because multimodal input must use the API's image representation and supported sourcing; merely writing a filename in text does not transfer pixels; the required outcomes are instead to preserve an understandable outcome when model generation cannot complete and to support Claude models accessed through different supported cloud offerings.
C:This alternative serves the separate goal to translate a request for citation-backed answers into observable application behavior because a functional requirement specifies what the user must receive and how success is evaluated, not merely that the team will use a language model; the required outcomes are instead to preserve an understandable outcome when model generation cannot complete and to support Claude models accessed through different supported cloud offerings.
D:This alternative serves the separate goal to design for a documented burst of simultaneous requests without uncontrolled overload because capacity planning separates average traffic from peak demand and prevents upstream rate limits from becoming cascading application failures; the required outcomes are instead to preserve an understandable outcome when model generation cannot complete and to support Claude models accessed through different supported cloud offerings.
Question 108
In an internal agent-extension rollout, a release requirement is to supply a read-only policy manual to an MCP client without giving it an action endpoint. Which implementation addresses that specific requirement?
- Expose the manual as an MCP resource rather than a tool.
- Create a Skill containing the procedure and supporting instructions.
- Use the MCP stdio transport with properly framed protocol messages.
- Package the connector behind an MCP server.
Explanation
Correct Answer
A:To supply a read-only policy manual to an MCP client without giving it an action endpoint in this internal agent-extension rollout, a resource is contextual data; representing a passive document as an imperative function adds unnecessary execution semantics; the proposed implementation directly supplies that capability.
Incorrect Answers
B:The proposed alternative helps to make a repeatable multi-step code-review methodology available to an agent because a Skill teaches how to perform a task, whereas a tool is the executable capability that performs an external operation; that mechanism does not meet the different requirement to supply a read-only policy manual to an MCP client without giving it an action endpoint.
C:The proposed alternative helps 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; that mechanism does not meet the different requirement to supply a read-only policy manual to an MCP client without giving it an action endpoint.
D:The proposed alternative helps to share one integration with several compatible AI clients or agent runtimes because MCP standardizes capability exposure across clients, reducing duplicate integration logic compared with proprietary tool schemas everywhere; that mechanism does not meet the different requirement to supply a read-only policy manual to an MCP client without giving it an action endpoint.
Question 109
An architecture decision for a production tool-calling agent must support keep a failed tool from being mistaken for successful completion of an objective and keep orchestration runtime and internal tool handlers inside company-managed infrastructure. Which option gives a complete rather than partial solution?
- Perform both steps: (1) Catch tool failures and report typed errors into the agent loop; (2) implement an application-owned messages api agent loop.
- Perform both steps: (1) Apply limits for steps, elapsed time, and token or tool expenditure; (2) implement an application-owned messages api agent loop.
- Perform both steps: (1) Apply limits for steps, elapsed time, and token or tool expenditure; (2) deploy the agent runner in the organization's own environment.
- Perform both steps: (1) Catch tool failures and report typed errors into the agent loop; (2) deploy the agent runner in the organization's own environment.
Explanation
Correct Answer
D:The first change enables the team to keep a failed tool from being mistaken for successful completion of an objective because explicit failure messages let the model or controller choose recovery; swallowing exceptions makes later actions rely on results that never existed; the second addresses the separate need to keep orchestration runtime and internal tool handlers inside company-managed infrastructure because self-hosting puts tool execution, networking, and operational controls under the application's responsibility, subject to its own deployment constraints.
Incorrect Answers
A:The first step is appropriate for the requirement to keep a failed tool from being mistaken for successful completion of an objective because explicit failure messages let the model or controller choose recovery; swallowing exceptions makes later actions rely on results that never existed; the other step instead aims to require precise control over message assembly, tool dispatch, and stopping behavior because a custom loop explicitly owns each model call, tool result, state update, and budget decision, trading flexibility for more engineering responsibility, which does not meet the second requirement to keep orchestration runtime and internal tool handlers inside company-managed infrastructure.
B:Both steps target different results—impose predictable maximum cost and wall-clock exposure on autonomous execution and require precise control over message assembly, tool dispatch, and stopping behavior—because runtime budgets are external constraints on agent activity and must be enforced by the host, not inferred from a cooperative natural-language instruction and a custom loop explicitly owns each model call, tool result, state update, and budget decision, trading flexibility for more engineering responsibility; the required capabilities are instead to keep a failed tool from being mistaken for successful completion of an objective and to keep orchestration runtime and internal tool handlers inside company-managed infrastructure.
C:The second step meets the requirement to keep orchestration runtime and internal tool handlers inside company-managed infrastructure because self-hosting puts tool execution, networking, and operational controls under the application's responsibility, subject to its own deployment constraints; the first step would only help 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, leaving the separate requirement to keep a failed tool from being mistaken for successful completion of an objective unmet.
Question 110
A team maintaining a multimodal customer portal must meet 2 separate acceptance criteria: (1) avoid assuming a structured SDK response is a plain string; (2) prevent duplicate purchases or tickets if an agent repeats a tool call. Which two implementation actions directly satisfy the stated requirements? Choose TWO.
- Give mutating operations an application-level idempotency key.
- Add automated consumer-provider contract tests.
- Inspect response.content blocks and metadata through the SDK's typed representation.
- Track latency, failure rates, token use, and output-quality signals over time.
- Read the completed stream's usage and final message metadata.
Explanation
Correct Answers
A:The test to prevent duplicate purchases or tickets if an agent repeats a tool call is met because idempotency ensures replays of the same intended action do not create duplicate real-world effects even when orchestration retries; the remaining selection covers the separate need to avoid assuming a structured SDK response is a plain string.
C:The test to avoid assuming a structured SDK response is a plain string is met because the SDK exposes block types, stop reason, and usage separately; naïve string concatenation can lose tool requests or other content; the remaining selection covers the separate need to prevent duplicate purchases or tickets if an agent repeats a tool call.
Incorrect Answers
B:The action is useful for the separate capability 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; however, the required test outcomes are to prevent duplicate purchases or tickets if an agent repeats a tool call and to avoid assuming a structured SDK response is a plain string.
D:The action is useful for the separate capability to detect rising inference errors after a feature reaches production because production health spans infrastructure and model behavior, so a healthy HTTP server alone does not prove the AI feature is functioning; however, the required test outcomes are to prevent duplicate purchases or tickets if an agent repeats a tool call and to avoid assuming a structured SDK response is a plain string.
E:The action is useful for the separate capability to account accurately for token consumption when a response is streamed because early text deltas do not necessarily carry authoritative final usage or stop information; accounting must wait for the appropriate terminal event; however, the required test outcomes are to prevent duplicate purchases or tickets if an agent repeats a tool call and to avoid assuming a structured SDK response is a plain string.