Question 360
A hook script itself crashes during a mandatory policy check. What production behavior is safest?
- Check protected-path checks in a pre-tool permission handler
- Conduct a PostToolUse handler after successful file-edit tools
- Read result payload semantics in a post-tool validation hook
- Fail closed or suspend the protected action according to explicit policy
Explanation
Correct Answer
D:When the decision service is unavailable, a high-risk action should not silently proceed. The policy needs a defined error path and observable failure state.
Incorrect Answers
A:A programmatic check can compare the requested path to restricted locations before access occurs; this contributes to denying sensitive file reads before access without meeting the need for handling a failed mandatory policy check.
B:A post-tool callback can inspect the completed edit and trigger formatting or verification in response; that serves reacting to a completed editing tool, whereas the case requires handling a failed mandatory policy check.
C:Because transport success does not necessarily mean business success, the measure supports detecting business failure despite transport success rather than the needed handling a failed mandatory policy check.
Question 361
A refactor repeatedly fails with the same stack trace, and the agent proposes the same edit each time. What workflow change helps?
- Decouple reproducible regressions from known nondeterministic failures
- Include a focused regression test that reproduces the bug before or with the fix
- Implement the failing edge case to tests and implement the smallest correction
- Analyze the root cause before attempting another identical patch
Explanation
Correct Answer
D:Repeated unmodified retries do not add evidence. Inspecting the failure path, inputs and assumptions can reveal the missing dependency or incorrect hypothesis.
Incorrect Answers
A:A single intermittent result is weak evidence; this may support interpreting flaky test signals correctly but cannot alone meet the requirement for escaping a repetitive unsuccessful repair loop.
B:A targeted test captures the desired behavior and guards against future regressions; the stated problem requires escaping a repetitive unsuccessful repair loop rather than establishing a regression test for a repaired defect.
C:Treating review feedback as a concrete test turns a discovered omission into durable coverage; a method suited to incorporating edge-case feedback into tests does not meet the need for escaping a repetitive unsuccessful repair loop.
Question 362
A batch processor changes its output schema version without warning and downstream parsers fail. What governance step is useful?
- Confirm tenant authorization and business references against the system of record
- Version the result contract and test consumers before deploying schema changes
- Supplement schema validity with independent source-backed checks
- Tag required properties and disallow additional properties in the output schema
Explanation
Correct Answer
B:Structured outputs create a stable interface that downstream systems depend upon. Even valid new JSON can be incompatible with older consumer expectations.
Incorrect Answers
A:JSON types and patterns cannot establish whether an identifier is authorized for the caller; that relates to validating authorization beyond JSON shape, not the necessary managing downstream compatibility of structured outputs.
C:A schema constrains representation, not factual truth; the deciding condition is managing downstream compatibility of structured outputs, whereas this action concerns distinguishing schema conformance from factual accuracy.
D:Required fields and closed object shapes prevent silently missing or invented keys; the option would serve forbidding missing and unexpected JSON fields, not managing downstream compatibility of structured outputs as specified in the stem.
Question 363
A worker attempts a tool invocation not included in its declared permissions. How should the runtime react?
- Permit a privileged diagnostics component inspect logs and return sanitized findings
- Reject the unauthorized call and report a permission denial to orchestration
- Drop the email-sending tool while keeping read and draft capabilities
- Furnish the coordinator lookup access and confine refund mutations to the authorized specialist
Explanation
Correct Answer
B:A capability boundary must be enforced at execution time. Silently granting the requested tool would defeat the separation established in agent configuration.
Incorrect Answers
A:Dedicated diagnostics isolates sensitive infrastructure data while providing actionable summaries; in this scenario, segregating diagnostic access from regular agents is secondary to enforcing a denied capability at runtime.
C:When addressing separating drafting from outbound delivery permissions, tool distribution should distinguish generating a proposed communication from transmitting it; the present task instead calls for enforcing a denied capability at runtime.
D:Partitioning capabilities prevents the orchestration role from bypassing specialist approval; isolating privileged refund capability and enforcing a denied capability at runtime require different controls in this situation.
Question 364
One conditional rule imposes Python formatting on TypeScript changes. What editorial fix restores clarity?
- Publish Python formatting requirements in the always-loaded shared instruction file without excluding TypeScript paths
- Correct overlapping conditions and specify one unambiguous policy
- Employ a recursive path pattern matching Terraform files across modules
- Focus on tightening the rule's paths glob to target only migration files
- Split language-specific guidance into separate path-targeted rule files
Explanation
Correct Answer
E:Distinct rules with distinct globs avoid contradictory language conventions. A mixed rule creates noise and can lead the agent to select the wrong formatter.
Incorrect Answers
A:Global instructions would continue forcing Python formatting guidance onto unrelated TypeScript edits. Scope the rule to Python files or directories with a correct path selector.
B:Conflicting active instructions can make outcomes nondeterministic; the action supports preventing contradictory conditional instructions but falls short of separating language-specific rules in this case.
C:Recursive globs represent a class of files regardless of module depth; this can serve covering nested infrastructure files consistently, but leaves the requirement for separating language-specific rules unmet.
D:A narrower glob controls the rule's applicability; that addresses narrowing a conditional rule's file scope, whereas this case calls for separating language-specific rules.
Question 365
A coding task requires discovery, design and editing, and editing before discovery often breaks tests. What plan is preferable?
- Focus on grouping closely related low-cost tasks into a meaningful worker assignment
- Require data normalization before downstream analysis
- Sequence exploration, plan approval and implementation as dependent tasks
- Include concurrency limits and a work queue to the task scheduler
Explanation
Correct Answer
C:Staging the work lets decisions use verified repository context. Premature mutation can amplify incorrect assumptions and complicate rollback.
Incorrect Answers
A:Because excessive decomposition multiplies setup and context cost, the measure supports avoiding overfragmentation of delegated work rather than the needed decomposing dependent coding stages.
B:The task graph must express actual data dependencies; its benefits for reflecting real data dependencies in the task graph do not resolve the need for decomposing dependent coding stages.
D:Capacity-aware scheduling controls rate limits and tool contention; this contributes to scheduling within available execution capacity without meeting the need for decomposing dependent coding stages.
Question 366
A scheduled batch runs near account rate limits and many tasks remain queued. What operational design helps?
- Assign a unique custom_id to every request and match returned results by that identifier
- Monitor pending request limits and pace submissions by available capacity
- Retain outcomes by input ID and create a retry queue from non-success statuses
- Dispatch independent requests through the asynchronous Message Batches API
Explanation
Correct Answer
B:Batch queues remain subject to provider limits and account constraints. Capacity-aware scheduling reduces unexpected expiration or throttling.
Incorrect Answers
A:Batch completion order is not guaranteed; a method suited to correlating out-of-order batch results does not meet the need for respecting throughput limits when batching.
C:Structured reconciliation lets teams preserve successful results and focus intervention on failures; that may serve constructing targeted post-batch retries, but it is no substitute for respecting throughput limits when batching.
D:Batch processing targets high-volume work without low-latency requirements and can reduce cost; this may support selecting asynchronous processing for independent bulk work but cannot alone meet the requirement for respecting throughput limits when batching.
Question 367
A controller loses its in-memory state when a worker restarts. What architecture makes agent state resilient?
- Branch the existing session into an isolated alternative branch
- Audit the persisted operation ID against the payment system before retrying
- Record message order and essential tool-result linkage along with session metadata
- Save checkpoint or session data in durable storage outside the worker process
- Log and compare policy versions before accepting the fork's recommendation
Explanation
Correct Answer
D:External persistence allows another process to resume execution after a crash. Solely in-memory histories vanish with the worker and break workflow continuity.
Incorrect Answers
A:Forking supports experimentation from a shared starting point while separating subsequent messages; the deciding condition is surviving worker restarts with durable sessions, whereas this action concerns branching session state without overwriting the original.
B:A durable transcript does not guarantee whether the external effect committed; the option would serve reconciling external actions after session interruption, not surviving worker restarts with durable sessions as specified in the stem.
C:Resumable state includes structured execution history, not merely the final natural-language summary; this addresses checkpointing structured agent progress, while the question is about surviving worker restarts with durable sessions.
E:Although a session branch can retain a different configuration snapshot, the choice addresses detecting configuration drift across forks and not the requested surviving worker restarts with durable sessions.
Question 368
A refactor must identify all relevant matches before changes, and inspection of existing files may not use a shell. Which TWO built-in-tool workflow choices satisfy these limits? Choose TWO.
- Inspect each existing source file with Read rather than shell-based cat commands
- Use Glob alone to locate every occurrence of a function name inside file contents
- Use Bash to print all source files and search their contents before editing
- Grep for content references, Read each affected file, then make reviewed Edit operations
Explanation
Correct Answers
A:Choosing direct file inspection over shell execution adds a necessary safeguard: reading file content through the dedicated tool avoids an unrestricted shell surface; matching the tool to the operation supports least privilege and clearer audit traces.
D:A search-first workflow reveals scope and avoids blind global replacement. Focused edits can be tested incrementally and preserve unrelated source.
Incorrect Answers
B:Glob matches path patterns; it does not search text inside files. Content references should be found with Grep and then inspected with Read, which does not invoke a shell.
C:A shell-based inspection directly violates the stated no-shell constraint. Built-in Grep and Read provide the relevant discovery and inspection capabilities.
Question 369
An agent should ordinarily decide whether it needs a CRM tool instead of being forced to call one on every turn. Which tool-selection strategy fits?
- Permit automatic tool selection based on the available tool definitions
- Handle and execute both tool calls, returning their respective results
- Handle the output as truncated and continue or retry with an adequate token budget
- Produce the completed assistant message to the caller
Explanation
Correct Answer
A:Automatic selection lets Claude choose between answering and requesting a tool when appropriate. Forced invocation would make even simple conversational acknowledgments trigger needless external work.
Incorrect Answers
B:A single model response can request multiple client tools; the host should process each authorized call and retain its unique tool-use identifier so results cannot be crossed; that addresses handling parallel client-tool calls, whereas this case calls for choosing optional rather than mandatory tool use.
C:A max_tokens stop means the generation reached the response budget, not that the report passed validation; this concerns detecting token-limit truncation, leaving the requirement for choosing optional rather than mandatory tool use outstanding.
D:The end_turn reason indicates the model finished its turn; the orchestration loop can surface the produced answer rather than manufacturing another tool step; recognizing a completed model turn and choosing optional rather than mandatory tool use are distinct responsibilities under these constraints.
Question 370
An unfamiliar repository contains a complex checkout flow and the requested modification has hidden side effects. What should happen before editing?
- Pinpoint specific tests and acceptance criteria alongside the edit steps
- Explore dependencies and produce a change plan before mutation
- Share the plan for approval before allowing write-capable execution
- Describe deployment order, checkpoints and rollback triggers before execution
Explanation
Correct Answer
B:Discovery identifies relevant modules, tests and side effects. Plan-first work reduces the chance that the agent patches an isolated line while breaking the wider transaction.
Incorrect Answers
A:A useful plan describes not just changes but the evidence that will establish success; supporting embedding verification into the implementation plan cannot replace the required mapping dependencies before modifying unfamiliar code.
C:Sensitive migrations benefit from separating design review from mutation; that serves requiring plan approval for a migration, whereas the case requires mapping dependencies before modifying unfamiliar code.
D:Because multi-service updates have dependencies and failure domains that cannot be improvised safely, the measure supports preparing release sequencing and rollback rather than the needed mapping dependencies before modifying unfamiliar code.