Question 210
An invoice-extraction workflow must recover from malformed output. The invoice number appears in two places with different values. Which change most directly addresses this condition?
- Mark the conflict and request human clarification with both source locations
- Stop after the configured retry limit and return a review-needed state
- Require evidence or field-level source matching for supplier identity
- Run business-rule arithmetic validation and request targeted correction
- Report the mismatched field and constraint, then retry within a bounded policy
Explanation
Correct Answer
A:A retry cannot resolve contradictory evidence by itself; the ambiguity belongs in review rather than arbitrary selection.
Incorrect Answers
B:This measure fits a different trigger: the model has failed the same correction twice. In that setting, Unlimited repair attempts can waste resources and may still fabricate a plausible record; bounded retries preserve reliability; it does not resolve the present need to escalate ambiguous source evidence rather than guess.
C:When a corrected output has valid JSON but an invented supplier name, this would be useful because syntax repair does not imply factual correctness; source-backed validation checks whether each critical value is actually present; this item instead calls for a way to escalate ambiguous source evidence rather than guess.
D:A valid numeric type does not prove accounting consistency; arithmetic rules catch errors outside the schema's expressiveness; the alternative would make sense if an extracted total passes type validation but does not equal subtotal plus taxes. It does not directly fix the observed condition: the invoice number appears in two places with different values.
E:Specific schema errors give the model a repair target; blindly regenerating without feedback repeats the same failure mode; the relevant precondition would be a tool payload is not valid against the required JSON Schema. The question instead describes the invoice number appears in two places with different values.
Question 211
An internal knowledge system is being exposed through the Model Context Protocol. A team wants a reusable suggested workflow template for incident triage. What is the most defensible next step?
- Expose the query as an MCP tool with a defined input schema
- Publish a parameterized MCP prompt for the triage workflow
- Refresh or version the resource reference when content changes
- Handle MCP content blocks according to their declared type instead of flattening everything to text
Explanation
Correct Answer
B:Prompts package reusable instructions and messages, whereas a tool would imply executable service behavior. Here it serves the explicit need to offer a reusable interaction prompt without adding a separate unverified inference.
Incorrect Answers
A:Tools are callable actions with arguments; a parameterized database operation is not just a static document resource; the relevant precondition would be Claude must actively run a database query with parameters. The question instead describes a team wants a reusable suggested workflow template for incident triage.
C:When a resource URI resolves to a newer policy than the model saw earlier, this would be useful because a retrievable resource is not permanently immutable; current source identity is necessary for reliable policy decisions; this item instead calls for a way to offer a reusable interaction prompt.
D:This measure fits a different trigger: a connection returns JSON records and images through multiple content parts. In that setting, Typed content lets a client interpret data faithfully and prevents binary or structured outputs from being silently mangled; it does not resolve the present need to offer a reusable interaction prompt.
Question 212
An AI team uses a generator and reviewer for compliance reports. The writer is allowed to alter the reviewer verdict before publication. Which approach satisfies the stated constraint?
- Use an independent reviewer prompt or agent grounded in source evidence
- Provide the reviewer the cited sources and a claim-verification rubric
- Keep review decisions in a distinct controlled stage or human gate
- Preserve both positions and send the evidence to a human approver
Explanation
Correct Answer
C:An independent check loses value if the generator can overwrite or ignore a failed result without accountability. For the stated case, the check is whether the system can separate generation from approval authority; this approach supplies that control.
Incorrect Answers
A:A separate review role with explicit criteria can catch mistakes that a writer may repeat during unstructured self-review; the relevant precondition would be the writer evaluates its own output and overlooks repeated factual errors. The question instead describes the writer is allowed to alter the reviewer verdict before publication.
B:Without reference evidence, the reviewer can only assess plausibility and may approve fabricated claims; the alternative would make sense if the reviewer has no access to source documents and simply judges fluency. It does not directly fix the observed condition: the writer is allowed to alter the reviewer verdict before publication.
D:When a dispute remains after the independent reviewer raises a high-risk concern, this would be useful because persistent disagreement is a signal for accountable judgment, not a reason to keep agents debating indefinitely; this item instead calls for a way to separate generation from approval authority.
Question 213
A support application selects tool-use policies for sensitive operations. An audit-only response must not invoke any available tools while explaining a policy decision. Which mechanism should be selected?
- Keep the refund capability unavailable or execution-blocked for this state
- Set tool_choice to none on the request that should not dispatch a tool
- Permit automatic tool selection within a safe allowlist
- Update the tool registry and descriptions so only the current lookup is eligible
Explanation
Correct Answer
B:The none mode prevents tool calls on that request even when tool definitions are present; later actions still need application-side authorization and controls.
Incorrect Answers
A:Tool choice is not the same as permission; the financial action must be gated on verified identity and policy; the alternative would make sense if a tool can issue refunds but the current conversation has not passed identity verification. It does not directly fix the observed condition: an audit-only response must not invoke any available tools while explaining a policy decision.
C:This measure fits a different trigger: a user asks an open-ended policy question that does not necessarily require a tool. In that setting, Over-forcing one retrieval tool can cause needless calls; controlled automatic choice suits variable informational needs; it does not resolve the present need to prevent tool dispatch for an intentionally tool-free response.
D:When an agent repeatedly chooses an obsolete lookup despite a new supported endpoint, this would be useful because retiring ambiguous or obsolete tools attacks the selection problem at its source instead of adding reminders to avoid the old one; this item instead calls for a way to prevent tool dispatch for an intentionally tool-free response.
Question 214
A Claude Code team shares MCP connections across several repositories. A secret API token is needed by a configured MCP server. What should be implemented at the appropriate boundary?
- Store the connection in the supported project-scoped configuration
- Use a user-scoped MCP configuration for the reusable personal connection
- Audit server provenance, tool definitions, and permission prompts before enablement
- Inject the secret through a managed environment or credential store
Explanation
Correct Answer
D:Project configuration can be shared and reviewed, but credentials belong in protected storage rather than plaintext repository files.
Incorrect Answers
A:A project-level definition keeps repo-specific capabilities with the project instead of exposing them to unrelated workspaces; the relevant precondition would be a repository requires a custom issue tracker MCP server not needed anywhere else. The question instead describes a secret API token is needed by a configured MCP server.
B:User scope supports broadly useful tooling without copying the connection into every repository; the alternative would make sense if a developer uses the same safe personal reference server across many projects. It does not directly fix the observed condition: a secret API token is needed by a configured MCP server.
C:This measure fits a different trigger: a new MCP server exposes a tool that should not be trusted automatically. In that setting, External servers can expose surprising side effects or injected descriptions; installation is not authorization for every capability; it does not resolve the present need to avoid committing credentials in shared configuration.
Question 215
A research assistant uses the Messages API to coordinate evidence gathering. A returned tool_use_id is absent from the outstanding-request map. Which option best meets the operational need?
- Validate each tool-result ID against a pending request before appending it
- Treat tool output as data and enforce higher-priority instructions and tool permissions
- Wait for the first result and derive the second request from verified output
- Detect max_tokens and continue or retry within a controlled budget
- Treat ordinary assistant text as content rather than an executable tool instruction
Explanation
Correct Answer
A:Correlation prevents an orphaned or stale result from contaminating the conversation and makes replay mismatches detectable.
Incorrect Answers
B:This measure fits a different trigger: a search tool returns untrusted web text that says to ignore system instructions. In that setting, External pages may contain prompt injection; tool output should inform answers, not override policy or initiate new privileges; it does not resolve the present need to reject an uncorrelated tool response.
C:When two sequential searches must use the first response to form the second query, this would be useful because when the next arguments depend on earlier evidence, speculative parallel execution can issue an invalid or misleading query; this item instead calls for a way to reject an uncorrelated tool response.
D:The stop reason distinguishes token exhaustion from a finished response; blindly accepting the fragment risks malformed or unsupported findings; the relevant precondition would be a response ends because the output-token limit was reached halfway through a structured answer. The question instead describes a returned tool_use_id is absent from the outstanding-request map.
E:Only explicit tool-use content requests host execution; natural-language statements do not authorize a network call; the alternative would make sense if an assistant text block describes a search but includes no tool_use block. It does not directly fix the observed condition: a returned tool_use_id is absent from the outstanding-request map.
Question 216
Claude Code is working with an unfamiliar repository too large to read at once. Several subagents have inspected separate subsystems. What is the most appropriate design choice?
- Read the two implicated files and retain references to the other hits
- Record directories, ownership boundaries, and key entry points before deep reading
- Retain constraints, decisions, evidence references, and remaining risks
- Merge file-indexed findings with status and unresolved issues
Explanation
Correct Answer
D:A structured handoff preserves useful specialist evidence while limiting the coordinator's context load. This matches the workflow goal to summarize discoveries into a coordinator-friendly artifact rather than introducing an unrelated action.
Incorrect Answers
A:Focused context makes the debugging task tractable without deleting the ability to revisit excluded evidence; the alternative would make sense if search results span many modules but only two files are relevant to the current bug. It does not directly fix the observed condition: several subagents have inspected separate subsystems.
B:A file map supports targeted exploration and prevents repeatedly flooding context with unrelated source files; the relevant precondition would be an agent cannot tell where major services live in a large monorepo. The question instead describes several subagents have inspected separate subsystems.
C:When the working context is compacted after many exploratory reads, this would be useful because useful compaction shortens history while keeping the information needed to continue safely and verify the eventual result; this item instead calls for a way to summarize discoveries into a coordinator-friendly artifact.
Question 217
A developer uses Claude Code to change a risky authentication module. The team does not know where authorization is implemented. Which response is most reliable?
- Use plan mode to inspect the codebase and propose steps without applying edits
- Present the plan and wait for authorization before enabling write operations
- Use read-only search and inspection to map the responsible files
- Perform the authorized edits and run the specified verification steps
Explanation
Correct Answer
C:An exploratory pass establishes architecture evidence; jumping into edits risks changing the wrong component. This matches the workflow goal to explore relevant code before proposing a change rather than introducing an unrelated action.
Incorrect Answers
A:Read-only planning supports review of dependencies and risk before the team authorizes source changes; the relevant precondition would be the engineer wants a proposed design and affected-file list before any modifications. The question instead describes the team does not know where authorization is implemented.
B:This measure fits a different trigger: a proposed refactor requires an explicit human go/no-go. In that setting, A design review requirement is not satisfied merely by generating a plan if the next stage can still commit changes autonomously; it does not resolve the present need to explore relevant code before proposing a change.
D:When the approved implementation has already been planned and reviewed, this would be useful because once scope and approval are established, indefinite re-planning adds little value; controlled execution produces testable results; this item instead calls for a way to explore relevant code before proposing a change.
Question 218
A product team decomposes complex Claude tasks into subtasks. An analysis must validate extracted figures before calculating a ratio. What should the team do to resolve this particular gap?
- Call Claude once with a precise classification instruction and output contract
- Extract, validate, calculate, then explain in that order
- Use an agent loop with step budgets and decisions conditioned on observations
- Collect sources concurrently and synthesize once all results pass validation
Explanation
Correct Answer
B:The arithmetic is only meaningful after source values pass type and completeness checks; dependencies determine stage order.
Incorrect Answers
A:A single well-scoped task avoids unnecessary agent coordination when no independent dependencies or tools are needed; the relevant precondition would be a customer email requires one classification label and no external data. The question instead describes an analysis must validate extracted figures before calculating a ratio.
C:This measure fits a different trigger: the number of investigative steps cannot be known until initial evidence arrives. In that setting, Adaptive tool choice is useful for uncertain exploration, provided the host constrains possible actions and termination; it does not resolve the present need to use sequential decomposition for dependent computation.
D:Independent lookups can run together, but writing should wait for the join to prevent omissions; the alternative would make sense if a report must combine three unrelated data sources before drafting. It does not directly fix the observed condition: an analysis must validate extracted figures before calculating a ratio.
Question 219
A customer-service agent uses hooks around tool operations. The application needs a durable record whenever a privileged tool is invoked. Which control is most directly relevant?
- Record the event at the tool-execution boundary using a hook or service log
- Remove or deny shell execution in the actual tool permissions or hook
- Inspect and reject paths outside the approved workspace in pre-tool logic
- Transform dates at a controlled post-tool boundary into a canonical format
Explanation
Correct Answer
A:Execution-time logging captures real calls and outcomes; model-generated explanations alone are not reliable audit evidence.
Incorrect Answers
B:This measure fits a different trigger: a prompt says 'never run shell commands,' but the tool remains available and callable. In that setting, Text guidance is probabilistic, whereas execution controls can reject a forbidden operation regardless of model output; it does not resolve the present need to audit actual privileged execution attempts.
C:When a dangerous path appears in file-write arguments before a tool runs, this would be useful because argument-level checks prevent an otherwise permitted write tool from escaping the authorized directory; this item instead calls for a way to audit actual privileged execution attempts.
D:A post-tool normalization step makes structured evidence consistent without hoping a later prompt corrects every variant; the alternative would make sense if a source-search tool returns inconsistent date formats that confuse downstream parsing. It does not directly fix the observed condition: the application needs a durable record whenever a privileged tool is invoked.
Question 220
An MCP service encounters errors while fetching customer records. A database loses connectivity during a lookup. How should the workflow be adjusted?
- Return a permission-denied outcome and require an authorized path
- Return a structured validation error identifying the affected field
- Classify the issue as transient unavailable and provide a safe diagnostic identifier
- Return a distinct not-found status without recommending blind retries
- Return a structured rate-limit error and retry-after guidance
Explanation
Correct Answer
C:Clear error categorization lets orchestration retry selectively while keeping internal stack traces out of model-facing content.
Incorrect Answers
A:This measure fits a different trigger: a tool call is rejected because the caller lacks access to the account. In that setting, A permission error cannot be fixed by repeating the same call; it needs identity correction or escalation without leaking account details; it does not resolve the present need to identify a recoverable service failure.
B:When an invalid date format fails argument validation before tool execution, this would be useful because a field-specific error supports a corrected retry, whereas generic failure text makes it hard to choose an appropriate next action; this item instead calls for a way to identify a recoverable service failure.
D:A genuinely missing resource is different from transient infrastructure failure; retrying unchanged identifiers adds load without resolving it; the alternative would make sense if a requested account number is well formed but does not exist. It does not directly fix the observed condition: a database loses connectivity during a lookup.
E:Throttling is transient and should be handled with bounded delay, not treated as an absent customer record; the relevant precondition would be a backend returns HTTP 429 with a retry window. The question instead describes a database loses connectivity during a lookup.