Question 170
A team needs to retrieve batch results several months after processing without storing them itself. What retention issue must it address?
- Maintain each request's custom_id and associated error classification
- Persist results to controlled storage within the documented retrieval period
- Persist outcomes by input ID and create a retry queue from non-success statuses
- Select asynchronous batching and budget for deferred results
Explanation
Correct Answer
B:Batch result URLs are not an archival store. The application must download and retain required records before provider retention windows elapse.
Incorrect Answers
A:Stable item identifiers connect failures to inputs and retry history; supporting linking each batch error to its source record cannot replace the required preserving batch output beyond service retention.
C:Structured reconciliation lets teams preserve successful results and focus intervention on failures; that serves constructing targeted post-batch retries, whereas the case requires preserving batch output beyond service retention.
D:Because batch processing can reduce unit cost at the expense of immediate completion, the measure supports trading response latency for bulk-processing cost rather than the needed preserving batch output beyond service retention.
Question 171
A reviewer wants a skill to run in its own context rather than contaminate the main coding conversation. Which skill behavior is relevant?
- Limit skill tool permissions to the necessary read-oriented operations
- Configure isolated or forked context for the specialized skill
- Eliminate embedded secrets and access approved credentials through secure runtime configuration
- Employ project-relative paths and document required environment inputs
Explanation
Correct Answer
B:Context isolation can keep reviewer instructions and intermediate analysis apart from the parent task. It supports independent evaluation and limits accumulated noise.
Incorrect Answers
A:A skill's allowed tool surface should match its declared job; that may serve defining least-privilege tools for a skill, but it is no substitute for isolating a specialized skill's context.
C:Skill contents may be loaded into model context and shared in repositories; the stated problem requires isolating a specialized skill's context rather than securing reusable skill contents.
D:Portable skills avoid machine-specific assumptions and resolve resources from a predictable project context; a method suited to removing machine-specific assumptions from a skill does not meet the need for isolating a specialized skill's context.
Question 172
A tool error message reveals a private backend connection string to the agent. What error design prevents leakage?
- Identify partial completion explicitly and identify which data remains unavailable
- Repeat time-sensitive reads after recovery and replace stale failure state
- Redact secrets in public tool errors and retain details in protected logs
- Follow bounded retry-after guidance in the retry scheduler
Explanation
Correct Answer
C:Agent-visible diagnostics should be actionable without exposing credentials. Separate internal correlation IDs can support operators while keeping sensitive infrastructure details private.
Incorrect Answers
A:A partial response must not masquerade as complete evidence; this addresses communicating incomplete tool output, while the question is about sanitizing error information returned to models.
B:Cached failure artifacts are not live evidence; that relates to refreshing state after a transient service failure, not the necessary sanitizing error information returned to models.
D:Rate-limit responses can indicate when another attempt is reasonable; the deciding condition is sanitizing error information returned to models, whereas this action concerns honoring rate-limit backoff information.
Question 173
The agent is explicitly operating in a read-only assessment environment. Which interaction should it avoid?
- Introduce the reviewed patch within its stated scope and verify the security regression
- Do not transition from planning or inspection into write-capable execution
- Carry out the narrow approved edit directly and run the relevant check
- Seek routine approval and automatically enable file writes once a stakeholder agrees
Explanation
Correct Answer
B:An assessment boundary prohibits side effects regardless of how confident the proposed fix appears. The deliverable should remain a reviewed plan or findings.
Incorrect Answers
A:Replanning a settled narrow remediation can delay correction and introduce scope drift; the unsatisfied requirement is honoring a read-only operating constraint, not avoiding needless replanning of an approved patch.
C:When considering implementing a tiny well-scoped change, a bounded, well-specified low-risk change often needs no lengthy planning phase; the requirement for honoring a read-only operating constraint remains unmet.
D:Approval does not remove an environment's explicit read-only operating restriction. Assessment must remain non-mutating unless the operating context itself is formally changed.
Question 174
A pull-request reviewer must run unattended in CI but may only read source and report issues, not commit changes. Which TWO implementation decisions satisfy these constraints? Choose TWO.
- Give the CI reviewer read-only repository credentials and no write-capable tools
- Invoke Claude Code with its non-interactive -p mode in the CI job
- Confine retrieval and review to the changed files and their affected dependencies
- Create proposed changes in an isolated workspace or pull request for review
Explanation
Correct Answers
A:Alongside invoking Claude Code in a noninteractive pipeline, the workflow needs enforcing read-only CI review permissions: CI runners often have elevated tokens; A read-only tool surface and minimal permissions prevent a review agent from making unrequested source changes.
B:The print-style command supports scripting without a chat session. CI should provide explicit prompts, scoped tools and machine-readable results for the pipeline to consume.
Incorrect Answers
C:A diff-aware prompt and bounded context reduce cost while preserving relevant evidence; the action supports constraining CI review to changed scope but falls short of invoking Claude Code in a noninteractive pipeline in this case.
D:Generated tests may be incorrect or brittle; this can serve isolating generated CI code changes, but leaves the requirement for invoking Claude Code in a noninteractive pipeline unmet.
Question 175
A support coordinator needs to search CRM and then ask a separate refund specialist for approval. How should privileges be divided?
- Condition publish authority on validated workflow state instead of mere tool availability
- Give the coordinator lookup access and confine refund mutations to the authorized specialist
- Publish separated tool sets or filtered capabilities according to worker identity
- Qualify or rename tools to identify their owning integration and purpose
Explanation
Correct Answer
B:Partitioning capabilities prevents the orchestration role from bypassing specialist approval. The parent can request an action without holding all execution privileges itself.
Incorrect Answers
A:Tool existence is not business authorization for every stage; that serves making a sensitive tool contingent on workflow state, whereas the case requires isolating privileged refund capability.
C:Different trust levels deserve different callable surfaces; its benefits for partitioning sensitive and ordinary MCP capabilities do not resolve the need for isolating privileged refund capability.
D:Unique, descriptive identifiers help the agent and dispatcher target the intended backend; supporting resolving tool name collisions cannot replace the required isolating privileged refund capability.
Question 176
A research subagent is asked to summarize a repository but has no access to the relevant files. What should the parent provide?
- Delegate with the precise file paths or retrieved evidence required for the task
- Start a reviewer with an isolated prompt and the source requirements only
- Launch the two workers in separate contexts and compare at the parent level
- Confine both delegated instructions and available data connectors to the assigned case
Explanation
Correct Answer
A:Subagents have their own context and should be given task-critical inputs explicitly. Assuming all hidden parent conversation details appear automatically can produce unsupported summaries.
Incorrect Answers
B:A clean review context reduces anchoring on the first draft; a method suited to maintaining independence for critique does not meet the need for passing task-specific context into a subagent.
C:For obtaining independent subagent hypotheses, isolated contexts create genuinely independent assessments; however, this requirement concerns passing task-specific context into a subagent.
D:Context minimization and access controls complement each other; that may serve narrowing a worker's data-access boundary, but it is no substitute for passing task-specific context into a subagent.
Question 177
A customer returns tomorrow to continue a partially researched case. What state strategy avoids starting from scratch?
- Copy into an isolated branch from the same validated baseline and keep branch histories separate
- Match the persisted operation ID against the payment system before retrying
- Persist a session identifier and resume its authorized conversation state
- Verify again time-sensitive evidence before continuing the resumed workflow
Explanation
Correct Answer
C:Session continuity reconnects the agent to relevant history and progress. Without durable session tracking, follow-up turns may omit prior checks and repeat tool work.
Incorrect Answers
A:Although a common starting context makes alternatives comparable, the choice addresses comparing alternatives from a common session baseline and not the requested resuming a saved agent session.
B:A durable transcript does not guarantee whether the external effect committed; the deciding condition is resuming a saved agent session, whereas this action concerns reconciling external actions after session interruption.
D:Session retention preserves past context, not the current truth of external systems; this addresses refreshing stale evidence during resume, while the question is about resuming a saved agent session.
Question 178
A coordinator keeps spawning an additional worker whenever a report lacks a citation, despite already having the source document. Which fix is most direct?
- Employ a coordinator that assigns bounded tasks and combines worker outputs
- Monitor each branch's completion state and retry only the failed dependency
- Hand off trivial deterministic steps directly and reserve specialists for nontrivial reasoning
- Add a targeted citation repair pass before further broad delegation
Explanation
Correct Answer
D:Missing source attribution is a bounded post-processing defect. Re-running the whole research tree is more expensive and can introduce new disagreement without fixing the specific citation gap.
Incorrect Answers
A:When considering centralized multi-agent coordination, a central orchestrator can define evidence requirements, route work by specialty and reconcile returned results; the requirement for targeted repair instead of repeated broad delegation remains unmet.
B:When addressing recovering a failed worker without rerunning completed branches, durable per-worker status lets the orchestrator resume a specific failed task; the present task instead calls for targeted repair instead of repeated broad delegation.
C:Agent delegation has token, latency and observability costs; avoiding needless specialist agent invocations and targeted repair instead of repeated broad delegation require different controls in this situation.
Question 179
An extraction pipeline needs high precision on financial values but can tolerate slower processing. Which pattern supports it?
- Add a second independent validation pass against the source document
- Create a labeled test set including true positives, negatives and borderline cases
- Compare two generated financial outputs to each other without opening the original document
- Keep stable validated facts and limit later passes to approved presentation edits
- Direct the reviewer to deliver evidenced findings for approved implementation
Explanation
Correct Answer
A:A dedicated verification step can compare amounts and field provenance before acceptance. Extra latency can be justified when financial errors are costly.
Incorrect Answers
B:Calibration requires examples with known outcomes, not only successful demonstrations; this can serve measuring review precision against ground truth, but leaves the requirement for trading review latency for financial precision unmet.
C:Agreement between model outputs does not establish financial accuracy. The independent pass must compare amounts, fields and calculations with the source file.
D:Separating factual content from stylistic changes reduces accidental meaning drift; this concerns preserving verified facts during editing, leaving the requirement for trading review latency for financial precision outstanding.
E:Reviewer and implementer responsibilities differ; separating review conclusions from edit authority and trading review latency for financial precision are distinct responsibilities under these constraints.
Question 180
A document extraction job contains 8,000 unrelated forms. Which partitioning enables scalable throughput?
- Define parallel prerequisite branches with a join gate before completion
- Decouple read-only verification from the authorized state-changing step
- Begin by completing mandatory high-value deliverables before optional expansions
- Divide inputs into independent batches with stable document IDs
Explanation
Correct Answer
D:Independent document chunks can run concurrently and recover separately. Unique IDs preserve attribution when outputs arrive out of order.
Incorrect Answers
A:A fan-out and join pattern preserves concurrency while enforcing that no required verification branch is skipped; its benefits for joining mandatory parallel checks do not resolve the need for partitioning bulk documents into traceable units.
B:Distinct stages expose eligibility decisions and permit human approval before mutations; supporting separating validation from side-effecting tasks cannot replace the required partitioning bulk documents into traceable units.
C:A critical-path plan ensures that the required outcome exists even if time runs short; this contributes to prioritizing the critical reporting path without meeting the need for partitioning bulk documents into traceable units.