Question 270
A team copies hundreds of examples into every call and exceeds latency targets. What optimization fits?
- Preserve example-set versions and compare outcomes on a fixed test suite
- Build in an example that correctly returns an unknown or escalation category
- Provide contrasting high-severity and low-severity labeled cases with rationales
- Retain a small representative few-shot set and evaluate the trade-off empirically
Explanation
Correct Answer
D:Large demonstrations consume context and tokens. Selective examples can preserve important decision boundaries while reducing request size.
Incorrect Answers
A:Versioned few-shot experiments reveal the impact of changes; supporting measuring few-shot changes with controlled evaluations cannot replace the required controlling context cost from examples.
B:Boundary examples teach that abstention is acceptable; this contributes to calibrating abstention with a boundary example without meeting the need for controlling context cost from examples.
C:Contrastive examples teach the decision boundary between categories; that serves distinguishing severity classes with contrastive examples, whereas the case requires controlling context cost from examples.
Question 271
A multi-agent report merges evidence from workers but loses the distinction between primary documents and secondary summaries. What metadata matters?
- Preserve original-source identifiers and evidence type through every handoff
- Retain an auditable decision record linking tool observations, rules and approvals
- Encode the value as missing with a documented reason and prohibit invented identifiers
- Condition the claim and request stronger corroboration before treating it as established
Explanation
Correct Answer
A:A claim's evidentiary strength depends on where it originated. Maintaining source type prevents successive summaries from masquerading as independent confirmations.
Incorrect Answers
B:Reproducibility requires more than the final recommendation; this may support reconstructing the evidence behind past decisions but cannot alone meet the requirement for retaining primary-source lineage across agents.
C:Confidence labels cannot make absent evidence real; the stated problem requires retaining primary-source lineage across agents rather than documenting absence without manufacturing data.
D:Weak or incomplete evidence warrants explicit uncertainty; a method suited to calibrating claims to evidence strength does not meet the need for retaining primary-source lineage across agents.
Question 272
After a batch finishes, one document has failed while the others succeeded. Which retained identifier plus status information best supports tracing that particular failure to its input?
- Check one representative request shape synchronously before scaling the batch
- Adopt a synchronous interactive request for the latency-sensitive turn
- Identify failed items only by their array position, assuming the batch output matches submission order
- Keep only batch-level success totals and build a general retry queue while discarding each request's custom_id mapping
- Preserve each request's custom_id and associated error classification
Explanation
Correct Answer
E:Stable item identifiers connect failures to inputs and retry history. Without them, a parallel batch becomes difficult to audit or repair precisely.
Incorrect Answers
A:A small dry run catches unsupported parameters and schema errors early; this addresses prevalidating request shape before bulk submission, while the question is about linking each batch error to its source record.
B:A batch job runs asynchronously and completion is not immediate; that relates to keeping interactive requests out of deferred batches, not the necessary linking each batch error to its source record.
C:Batches can complete out of order; array position is an unsafe correlation key. Persist each item's custom_id alongside its individual success or error status.
D:Batch totals and an uncorrelated retry queue do not reveal which input document produced a given error. Retain the individual request custom_id with its result status and error classification.
Question 273
A report has an urgent one-page executive conclusion and optional deep appendices. How should tasks be prioritized?
- Give separate evaluators with distinct rubrics and merge after both finish
- Select direct deterministic lookup and rule evaluation for the simple stages
- Complete mandatory high-value deliverables before optional expansions
- Schedule exploration, plan approval and implementation as dependent tasks
Explanation
Correct Answer
C:A critical-path plan ensures that the required outcome exists even if time runs short. Optional exploration should not consume the budget reserved for core deliverables.
Incorrect Answers
A:Different quality dimensions need different tests; the unsatisfied requirement is prioritizing the critical reporting path, not decomposing evaluation into distinct quality checks.
B:Not every step requires another model or subagent; in this scenario, separating deterministic steps from inference is secondary to prioritizing the critical reporting path.
D:When addressing decomposing dependent coding stages, staging the work lets decisions use verified repository context; the present task instead calls for prioritizing the critical reporting path.
Question 274
A report generator omits risks because the prompt emphasizes successes only. What structure would improve completeness?
- Include a clear output structure with explicit labels for evidence and risk rationale
- Specify required positive findings, limitations, risks and outstanding decisions
- State explicit eligibility conditions and required evidence for each refund decision
- Specify the defect categories and require a cited code location for every finding
Explanation
Correct Answer
B:Balanced section requirements make omissions more visible during validation. A one-sided objective can produce polished output that neglects material implementation risks.
Incorrect Answers
A:Combining an evaluation rubric with an output contract makes responses comparable; this concerns standardizing the structure of risk assessments, leaving the requirement for eliciting both benefits and risks in reports outstanding.
C:Concrete criteria make a model's judgment testable against policy; the action supports translating broad intent into checkable refund rules but falls short of eliciting both benefits and risks in reports in this case.
D:Evidence-based criteria distinguish observed defects from speculative criticism; this can serve preventing unsupported PR review findings, but leaves the requirement for eliciting both benefits and risks in reports unmet.
Question 275
A reviewer finds an edge-case failure after the feature is otherwise complete. How should the agent proceed?
- Introduce a focused regression test that reproduces the bug before or with the fix
- Perform contract or compatibility tests for existing consumers
- Maintain changes bounded to the requested behavior and a small verified diff
- Run failure-path tests and add explicit handling where warranted
- Focus on adding the failing edge case to tests and implement the smallest correction
Explanation
Correct Answer
E:Treating review feedback as a concrete test turns a discovered omission into durable coverage. Guessing at a broad rewrite would be less targeted.
Incorrect Answers
A:A targeted test captures the desired behavior and guards against future regressions; that serves establishing a regression test for a repaired defect, whereas the case requires incorporating edge-case feedback into tests.
B:Because code that compiles can still break callers relying on prior inputs or outputs, the measure supports testing API compatibility after code generation rather than the needed incorporating edge-case feedback into tests.
C:Narrow modifications are easier to review, test and revert; its benefits for limiting a patch to the requested behavior do not resolve the need for incorporating edge-case feedback into tests.
D:Reliable implementation must address realistic negative inputs and dependency errors; supporting validating negative paths after implementation cannot replace the required incorporating edge-case feedback into tests.
Question 276
A delegated work item returns many pages but no acceptance signal. Which planning change improves control?
- Prepare parallel prerequisite branches with a join gate before completion
- Apply direct deterministic lookup and rule evaluation for the simple stages
- Define measurable deliverables and pass-fail conditions for each subtask
- Attempt again or replace only the failed subtask while keeping successful outputs
Explanation
Correct Answer
C:A task boundary is useful when the parent can decide whether its output satisfies requirements. Word count alone does not indicate completion or quality.
Incorrect Answers
A:A fan-out and join pattern preserves concurrency while enforcing that no required verification branch is skipped; the stated problem requires specifying task-level acceptance criteria rather than joining mandatory parallel checks.
B:Not every step requires another model or subagent; a method suited to separating deterministic steps from inference does not meet the need for specifying task-level acceptance criteria.
D:Independence in the task graph permits selective replay; that may serve isolating retry to failed work, but it is no substitute for specifying task-level acceptance criteria.
Question 277
A tool description says simply 'gets data', and Claude chooses it for several unrelated data sources. What change is most useful?
- Employ an unambiguous timestamp representation with timezone requirements
- Break up the overloaded interface into cohesive operations with narrow schemas
- Classify separately the read-only lookup from the authorized state-changing operation
- Describe the target resource, permitted query scope and result semantics
- Specify a stable output schema and version incompatible changes
- Describe explicit success, not-found and error result shapes
Explanation
Correct Answer
D:Effective tool descriptions explain when the operation should be used and what it returns. Vague statements force the model to infer missing API boundaries.
Incorrect Answers
A:Although machine-oriented time fields and explicit timezone semantics prevent interpretation drift, the choice addresses normalizing temporal tool parameters and not the requested exposing the resource scope of retrieval tools.
B:Smaller purpose-built tools make argument requirements and authorization easier to reason about; the deciding condition is exposing the resource scope of retrieval tools, whereas this action concerns reducing an overloaded tool schema.
C:Tool metadata should expose whether a call merely inspects or mutates an order; the option would serve clarifying read versus write tool intent, not exposing the resource scope of retrieval tools as specified in the stem.
E:A predictable contract lets agents and validators consume results consistently; that relates to managing output schema compatibility, not the necessary exposing the resource scope of retrieval tools.
F:Although a clear output contract lets downstream agents tell absent records from service faults, the choice addresses specifying the semantics of empty tool results and not the requested exposing the resource scope of retrieval tools.
Question 278
An MCP integration is reachable but the agent cannot access a particular server tool. What should the operator inspect first?
- Confirm a reviewed project .mcp.json definition into version control
- Apply supported authenticated transport and credential renewal rather than embedding a fixed secret
- Check tool exposure, permission filters and server-side authorization
- Add the integration in a local or user-controlled configuration scope
- Restrict the transport with TLS and authenticate clients before sensitive tool calls
Explanation
Correct Answer
C:Connectivity alone does not grant every capability. The client or server may restrict individual tools or identities even when the transport handshake succeeds.
Incorrect Answers
A:When considering sharing an MCP server configuration with a project, project-scoped MCP configuration can be distributed with the repository; the requirement for diagnosing a connected but unavailable MCP tool remains unmet.
B:When addressing maintaining authenticated MCP transport, remote MCP services need a secure authentication lifecycle; the present task instead calls for diagnosing a connected but unavailable MCP tool.
D:Personal scope keeps experimental endpoints out of the shared project contract; the unsatisfied requirement is diagnosing a connected but unavailable MCP tool, not keeping personal MCP configuration out of shared files.
E:Tool inputs and results may contain confidential operational data; in this scenario, protecting an MCP network connection is secondary to diagnosing a connected but unavailable MCP tool.
Question 279
A developer puts API credentials inside CLAUDE.md to simplify an agent's setup. What should be done instead?
- Classify repository instructions as untrusted and refuse exfiltrating requests
- Remove secrets from instruction text and use secure environment-based credential provisioning
- Establish the intended CLAUDE.md is discovered and loaded for that working context
- Express the stable rules once in reviewed project instructions
Explanation
Correct Answer
B:Instruction files are often shared or exposed as model context. Secrets belong in controlled stores or environment injection, not narrative repository guidance.
Incorrect Answers
A:Codebase text and instructions must not override higher-priority security boundaries; the action supports defending against untrusted repository instructions but falls short of protecting credentials from instruction-file exposure in this case.
C:Rules are useful only when the assistant encounters the relevant file and scope; that addresses verifying instruction-file loading, whereas this case calls for protecting credentials from instruction-file exposure.
D:Persistent project guidance makes the same expectations available across sessions; this concerns persisting stable codebase conventions, leaving the requirement for protecting credentials from instruction-file exposure outstanding.
Question 280
A team's policy script exits with a blocking status before a destructive command. How should the agent runtime respond?
- Produce a blocking control result instead of an informational log event
- Reject closed or suspend the protected action according to explicit policy
- Block over-budget requests in the tool-call authorization boundary
- Honor the hook's documented block decision and return the reason to the agent
Explanation
Correct Answer
D:A hook decision is part of the deterministic policy boundary for tool execution. Swallowing a deny signal would make the guard ineffective and hide the cause from the agent.
Incorrect Answers
A:Emitting telemetry alone does not enforce permission decisions; its benefits for distinguishing logging from access denial do not resolve the need for respecting a hook's deny outcome.
B:When the decision service is unavailable, a high-risk action should not silently proceed; supporting handling a failed mandatory policy check cannot replace the required respecting a hook's deny outcome.
C:Tracking usage and checking it before execution makes a budget enforceable; this contributes to enforcing a quota at the tool boundary without meeting the need for respecting a hook's deny outcome.