Question 90
An agentic research pipeline must propagate failures through several workers. A search specialist returns only two of four required evidence sources. How should the workflow be adjusted?
- Mark the calculation blocked and include the validation failure
- Refuse publication until mandatory worker outcomes are successful or formally waived
- Pass a typed timeout outcome instead of a successful empty payload
- Return partial-result status and the identifiers of missing sources
- Return per-source status with valid outputs and a partial-completion flag
Explanation
Correct Answer
D:Downstream synthesis must know that evidence collection is incomplete; treating a partial list as full coverage can distort conclusions.
Incorrect Answers
A:A computed value obtained from unverified inputs should not be promoted as evidence merely because a number was produced; it protects against the case where an analysis agent reports a numeric result but its input validation failed, rather than the failure described here: a search specialist returns only two of four required evidence sources.
B:This choice is relevant when the final publication stage needs all required source checks complete; a release gate should inspect explicit prerequisite state rather than trust an optimistic synthesis paragraph. Here, however, a search specialist returns only two of four required evidence sources.
C:It would be appropriate when a writing agent receives an upstream timeout labeled success; correct error semantics prevent downstream agents from mistaking unavailable data for evidence of absence. Here, a search specialist returns only two of four required evidence sources, so a different mechanism is required.
E:An all-or-nothing summary discards useful work, while an all-success label conceals missing coverage; partial status preserves both truths; the alternative problem occurs when one of three independent data sources fails while the others are usable. The present scenario instead requires the system to represent incomplete coverage explicitly.
Question 91
A codebase migration process must decide when to spawn specialists. Two services have unrelated APIs and can be analyzed separately. Which implementation decision is the best fit?
- Serialize updates or give workers isolated copies before concurrency
- Run schema discovery before generating or applying migration steps
- Delegate each service to a scoped worker and compare results afterward
- Use one owning reviewer or coordinator for the business-rule judgment
Explanation
Correct Answer
C:Independent boundaries permit concurrency without creating cross-service state conflicts. The relevant requirement is to partition independent analysis work, and this measure operates at that decision point.
Incorrect Answers
A:This choice is relevant when several workers share a mutable global configuration during analysis; parallel actions that change the same state can race; independence must include write effects, not merely different prompts. Here, however, two services have unrelated APIs and can be analyzed separately.
B:A migration plan based on guessed schema can break data, so dependency evidence must be present first; it protects against the case where a database migration depends on a schema inventory that does not yet exist, rather than the failure described here: two services have unrelated APIs and can be analyzed separately.
D:It would be appropriate when a change requires one authoritative interpretation of business rules; independent conflicting interpretations should not each drive a code change; central adjudication preserves consistency. Here, two services have unrelated APIs and can be analyzed separately, so a different mechanism is required.
Question 92
A coordinator controls which tools each research worker may call. A source-finding worker needs web search but cannot modify internal documents. What should the architect specify in the design?
- Expose read-only verification tools to the reviewer and reserve release actions for the owner
- Allow approved retrieval tools while withholding document-writing capabilities
- Let the coordinator select subsequent tools or specialists using completed status fields
- Give the writer the validated evidence artifact and no external search tool
Explanation
Correct Answer
B:Tool distribution should follow the worker's role; disallowing writes prevents discovery tasks from changing records.
Incorrect Answers
A:This choice is relevant when a reviewer needs source documents but must not publish final content; verification and publication are separate authority levels; a review worker can inspect evidence without being able to commit the final artifact. Here, however, a source-finding worker needs web search but cannot modify internal documents.
C:A routing decision depends on multiple worker results; it belongs at the point that sees the combined state; the alternative problem occurs when an orchestrator requires two worker outputs to choose the next branch. The present scenario instead requires the system to give the search worker a least-privilege tool set.
D:It would be appropriate when a drafting worker must produce prose using validated notes, not new unreviewed sources; removing discovery tools prevents the writing stage from introducing fresh unvetted material after validation. Here, a source-finding worker needs web search but cannot modify internal documents, so a different mechanism is required.
Question 93
An engineering organization maintains Claude Code commands and Skills. A project-specific release command should be reviewed with code changes. Which change most directly addresses this condition?
- Require actual deployment permission and approval outside the command text
- Keep the project's command or skill files under repository review
- Describe when the skill should be used and its expected workflow
- Make the long routine explicitly invocable rather than persistent root memory
Explanation
Correct Answer
B:A reviewed definition lets the team examine behavior changes and use the same release routine across contributors. For the stated case, the check is whether the system can version the shared workflow definition; this approach supplies that control.
Incorrect Answers
A:It would be appropriate when a command can deploy production, but a user can trigger it easily; a reusable command is an interface, not a security control; the deployment system must enforce who may perform the action. Here, a project-specific release command should be reviewed with code changes, so a different mechanism is required.
C:A clear purpose helps capability discovery; a generic name and vague summary can cause missed or incorrect invocation; it protects against the case where the team wants Claude to choose a documentation procedure when a related task appears, rather than the failure described here: a project-specific release command should be reviewed with code changes.
D:This choice is relevant when a custom command contains a long fixed prompt that every task automatically loads; optional procedures belong in reusable commands or Skills; always-loaded instructions should stay concise and broadly relevant. Here, however, a project-specific release command should be reviewed with code changes.
Question 94
A Claude Agent SDK application maintains long-running support sessions. A returning customer asks about a previously paused case with the same conversation state. What is the most defensible next step?
- Query the authoritative order service before another financial decision
- Use the saved session identity and recover relevant persisted messages
- Store session identifiers and required checkpoints in a durable backing store
- Create a fresh session scoped to the new customer's identity
Explanation
Correct Answer
B:Resumption preserves continuity for the same case, provided the application verifies authorization and current account state.
Incorrect Answers
A:Conversation memory is not a transactional system; live order status takes precedence over stale statements in a transcript; the alternative problem occurs when the resumed session references an order whose refund status may have changed. The present scenario instead requires the system to resume the intended prior conversation.
C:This choice is relevant when a long-running session needs to survive application restarts; recoverable state allows orchestration to continue after host failure, unlike in-memory variables lost on restart. Here, however, a returning customer asks about a previously paused case with the same conversation state.
D:It would be appropriate when a support case is closed, and a new unrelated customer starts a chat; a new identity requires isolated conversation state; reusing a previous customer's session can disclose private history. Here, a returning customer asks about a previously paused case with the same conversation state, so a different mechanism is required.
Question 95
A repository assistant uses several parallel agents to examine unfamiliar services. Two agents intend to edit the same config file concurrently. Which approach satisfies the stated constraint?
- State the expected test command and success condition in the delegated task
- Return a concise file-indexed summary and paths to relevant excerpts
- Limit task nesting and keep coordination responsibility at the parent
- Write the validated summary once and pass its identifier to downstream jobs
- Partition write ownership or serialize changes to the shared file
Explanation
Correct Answer
E:Parallel reads may be safe, but simultaneous edits to shared state invite races and difficult-to-merge patches. This matches the workflow goal to prevent conflicting parallel edits rather than introducing an unrelated action.
Incorrect Answers
A:It would be appropriate when a worker must verify that its proposed fix passes a particular unit test; a specialist with explicit verification criteria can return evidence of completion instead of an untested code suggestion. Here, two agents intend to edit the same config file concurrently, so a different mechanism is required.
B:A narrow handoff saves parent context while retaining references that can be reopened when deeper inspection is needed; it protects against the case where a search agent finishes with fifty raw file excerpts, rather than the failure described here: two agents intend to edit the same config file concurrently.
C:Unbounded subagent spawning increases cost and makes failure ownership obscure; a bounded hierarchy is more predictable; the alternative problem occurs when an analysis worker stalls after requesting another worker recursively. The present scenario instead requires the system to prevent conflicting parallel edits.
D:This choice is relevant when one specialist's findings are needed by three downstream specialists; a versioned artifact prevents three copies of the same evolving findings from diverging across worker contexts. Here, however, two agents intend to edit the same config file concurrently.
Question 96
A Claude-based coding assistant generates patch-review records. A line number in a valid review object exceeds the length of the referenced file. Which mechanism should be selected?
- Mark missing checks as blocked and do not assign a passing verdict
- Cross-check referenced symbols with source or build metadata
- Run permitted commands and report any unavailable or failing test conditions
- Check line ranges against the exact repository revision before accepting findings
Explanation
Correct Answer
D:Schema integers can still point outside actual files; source-aware validation prevents misleading review annotations.
Incorrect Answers
A:This choice is relevant when the model returns a plausible patch summary even though required static checks were skipped; a fluent summary cannot turn skipped checks into evidence; state should reflect actual execution results. Here, however, a line number in a valid review object exceeds the length of the referenced file.
B:Factual checking must follow formatting validation because a correctly typed nonexistent method is still an invalid finding; it protects against the case where a finding names an API method that has been removed, rather than the failure described here: a line number in a valid review object exceeds the length of the referenced file.
C:It would be appropriate when the assistant proposes tests that do not run in the current build; describing test commands does not prove they are executable or passing; CI evidence is required. Here, a line number in a valid review object exceeds the length of the referenced file, so a different mechanism is required.
Question 97
A research assistant uses the Messages API to coordinate evidence gathering. A response ends because the output-token limit was reached halfway through a structured answer. What should be implemented at the appropriate boundary?
- Wait for the first result and derive the second request from verified output
- Treat ordinary assistant text as content rather than an executable tool instruction
- Detect max_tokens and continue or retry within a controlled budget
- Validate each tool-result ID against a pending request before appending it
Explanation
Correct Answer
C:The stop reason distinguishes token exhaustion from a finished response; blindly accepting the fragment risks malformed or unsupported findings.
Incorrect Answers
A:This choice is relevant when two sequential searches must use the first response to form the second query; when the next arguments depend on earlier evidence, speculative parallel execution can issue an invalid or misleading query. Here, however, a response ends because the output-token limit was reached halfway through a structured answer.
B:Only explicit tool-use content requests host execution; natural-language statements do not authorize a network call; it protects against the case where an assistant text block describes a search but includes no tool_use block, rather than the failure described here: a response ends because the output-token limit was reached halfway through a structured answer.
D:It would be appropriate when a returned tool_use_id is absent from the outstanding-request map; correlation prevents an orphaned or stale result from contaminating the conversation and makes replay mismatches detectable. Here, a response ends because the output-token limit was reached halfway through a structured answer, so a different mechanism is required.
Question 98
A support workflow uses model-generated structured actions. A proposed address change contains a nonexistent postal code. Which option best meets the operational need?
- Consult current quota state before committing the mutation
- Include targeted validator diagnostics instead of a generic 'try again'
- Validate the record identity against the authenticated session
- Check postal-code validity with an authoritative service
Explanation
Correct Answer
D:A field can be a syntactically valid string while failing the external real-world requirement for a deliverable address.
Incorrect Answers
A:This choice is relevant when an action payload validates but would exceed a daily transaction limit; current cumulative limits depend on external state and require an authoritative execution-time check, not just a schema. Here, however, a proposed address change contains a nonexistent postal code.
B:It would be appropriate when the same invalid tool argument is regenerated after retry; specific error feedback helps repair a mismatched field, whereas an uninformative retry prompt may repeat the original value. Here, a proposed address change contains a nonexistent postal code, so a different mechanism is required.
C:Schema matching cannot establish that a valid-looking identifier belongs to the authorized customer; it protects against the case where a model response contains a correct enum but refers to the wrong customer ID, rather than the failure described here: a proposed address change contains a nonexistent postal code.
Question 99
An unfamiliar codebase requires careful changes through Claude Code. Claude starts inventing module names instead of inspecting the repository. What is the most appropriate design choice?
- Search the repository and read referenced files before proposing modifications
- Change the generator input and regenerate outputs using the supported workflow
- Capture the original reproducible failure and the verified result after the patch
- Mark the integration check blocked and provide the exact human-run requirement
Explanation
Correct Answer
A:File evidence prevents made-up dependencies and directs changes to the real code organization. This matches the workflow goal to ground the task in actual source structure rather than introducing an unrelated action.
Incorrect Answers
B:Derived artifacts are not authoritative; editing them directly is fragile and hides the persistent source of the behavior; the alternative problem occurs when the assistant modified generated files that will be overwritten by builds. The present scenario instead requires the system to ground the task in actual source structure.
C:It would be appropriate when the team wants proof that a bug is fixed, not just a plausible description; a before-and-after test provides stronger evidence of correction than a narrative assertion alone. Here, Claude starts inventing module names instead of inspecting the repository, so a different mechanism is required.
D:This choice is relevant when the acceptance test depends on a credential unavailable in the agent workspace; an unrun test must not be reported as passing; explicit blocked status lets the maintainer complete verification safely. Here, however, Claude starts inventing module names instead of inspecting the repository.
Question 100
An agent controller must coordinate recoverable workflows across interruptions. A tool result was durably saved with its request ID, but the host crashed before passing it back to Claude. Which response is most reliable?
- Reauthorize actions against the current permission configuration
- Create a clean conversation and explicitly scoped context for the new task
- Persist origin session, fork lineage, task ID, and artifact references
- Load checkpoint status and reuse completed result artifacts
- Replay the persisted, ID-correlated tool_result into the restored conversation
Explanation
Correct Answer
E:When the result is already durably recorded, the host can resume the exact tool-use/result exchange without repeating the side-effecting operation.
Incorrect Answers
A:It would be appropriate when a worker changed permissions in an earlier session, but today's role policy differs; session continuity is not continuing authorization; permissions may change independently of the saved transcript. Here, a tool result was durably saved with its request ID, but the host crashed before passing it back to Claude, so a different mechanism is required.
B:Starting fresh avoids accidentally promoting data from an unrelated case into a new workflow; the alternative problem occurs when a new agent task accidentally inherits old customer contact details. The present scenario instead requires the system to restore a completed tool result without running the tool twice.
C:This choice is relevant when engineers need to trace exactly where a resumed result originated; provenance across resumes and forks is essential for diagnosing which state or branch produced a disputed conclusion. Here, however, a tool result was durably saved with its request ID, but the host crashed before passing it back to Claude.
D:A checkpointed task graph supports partial recovery and avoids paying for identical completed work; it protects against the case where an interrupted research task has several finished specialist outputs, rather than the failure described here: a tool result was durably saved with its request ID, but the host crashed before passing it back to Claude.