Question 70
A messages controller treats stop_sequence as proof a task was fully resolved. What is the sounder decision?
- Carry out the requested client tool and return a matching tool_result
- Treat the custom stopping sequence as a generation boundary and validate task completion separately
- Emit an explicit failed tool result with diagnostic information rather than fabricate data
- Preserve the tool implementation in the authorized execution environment
Explanation
Correct Answer
B:A stop_sequence signals an application-specified token sequence caused termination. It does not independently prove that business assertions, tool operations, or response schemas are complete.
Incorrect Answers
A:A tool_use stop hands control to the application; it must run the requested operation and pass back a tool_result tied to the call identifier before Claude can finish reasoning with that evidence; supporting completing the client-tool round trip cannot replace the required separating custom stop boundaries from success.
C:The tool-error signal should flow into the conversation so the agent can retry, escalate, or communicate limits; that serves surfacing client-tool execution failure, whereas the case requires separating custom stop boundaries from success.
D:Because client tools are interfaces advertised to the model; the hosting application runs their operations, the measure supports distinguishing model intent from tool execution rather than the needed separating custom stop boundaries from success.
Question 71
An engineer wants personal response preferences that should apply across unrelated repositories. Which instruction location is appropriate?
- Put additional directory-specific CLAUDE.md files near the relevant code
- Retain CLAUDE.md concise and link to task-relevant supporting references
- Interpret repository instructions as untrusted and refuse exfiltrating requests
- Use the user's personal CLAUDE.md for cross-project preferences
Explanation
Correct Answer
D:User-scoped instructions express developer preferences beyond one repository. They should not be confused with project standards meant for version-controlled collaboration.
Incorrect Answers
A:Scoped instruction files can provide rules for particular subtrees; that may serve localizing instructions to code directories, but it is no substitute for applying personal preferences across projects.
B:High-signal standing instructions are easier to apply and less costly in context; this may support keeping persistent instructions focused but cannot alone meet the requirement for applying personal preferences across projects.
C:Codebase text and instructions must not override higher-priority security boundaries; the stated problem requires applying personal preferences across projects rather than defending against untrusted repository instructions.
Question 72
A two-pass extraction workflow produces different values on each pass for the same field. What is the safe reconciliation rule?
- Include a second independent validation pass against the source document
- Execute an independent verifier against the original requirements and observed tool evidence
- Track versions of prompts and rubrics and rerun a fixed regression evaluation suite
- Compare each candidate value to document evidence and escalate unresolved disagreement
Explanation
Correct Answer
D:Model disagreement does not supply a factual tie-breaker. Source spans or human verification should determine which value, if any, can be trusted.
Incorrect Answers
A:A dedicated verification step can compare amounts and field provenance before acceptance; this addresses trading review latency for financial precision, while the question is about resolving extraction disagreements with evidence.
B:Self-assessment tends to reuse the model's assumptions; that relates to reducing self-confirmation bias in generated artifacts, not the necessary resolving extraction disagreements with evidence.
C:Although comparable benchmarks reveal behavior changes when the system evolves, the choice addresses making multipass quality evaluations reproducible and not the requested resolving extraction disagreements with evidence.
Question 73
A hook is configured for Bash but is also firing on unrelated tools because its matcher is broad. What should be revised?
- Provide a blocking control result instead of an informational log event
- Narrow the hook matcher to the intended tool name or invocation pattern
- Require protected-path checks in a pre-tool permission handler
- Verify and safely handle hook inputs before invoking external processes
Explanation
Correct Answer
B:Matchers determine which hook executions apply. Tight scoping prevents policy handlers from disrupting operations they were not designed to supervise.
Incorrect Answers
A:Emitting telemetry alone does not enforce permission decisions; the unsatisfied requirement is targeting a hook to relevant operations, not distinguishing logging from access denial.
C:When considering denying sensitive file reads before access, a programmatic check can compare the requested path to restricted locations before access occurs; the requirement for targeting a hook to relevant operations remains unmet.
D:When addressing validating data inside enforcement hooks, hooks execute application-controlled code and may receive adversarial strings; the present task instead calls for targeting a hook to relevant operations.
Question 74
A pull request changes database migrations, and the reviewer needs to check backward compatibility. What context should CI include?
- Supply relevant migration files and compatibility criteria with the diff
- Limit allowed tools and repository credentials to read-only review operations
- Include the commit SHA, workflow run ID and validated result artifact
- Require machine-readable JSON output and validate its schema
Explanation
Correct Answer
A:Reviewing only altered lines can miss deployment sequencing and consumers. Targeted dependent context enables a meaningful judgment without loading the whole repository.
Incorrect Answers
B:CI runners often have elevated tokens; enforcing read-only CI review permissions and providing dependency context for migration review are distinct responsibilities under these constraints.
C:Immutable revision identifiers distinguish reviews of changed code and support audit trails; the action supports binding CI review evidence to a code revision but falls short of providing dependency context for migration review in this case.
D:Structured output is easier for CI to parse and gate deterministically; this can serve parsing review output safely in CI, but leaves the requirement for providing dependency context for migration review unmet.
Question 75
A subagent finishes extracting contract dates while a separate validator reports errors. How should the coordinator finalize?
- Hold publication until validation reports are reconciled with the extracted artifacts
- Execute independent read-only specialist tasks concurrently and join their outputs
- Direct trivial deterministic steps directly and reserve specialists for nontrivial reasoning
- Choose a coordinator that assigns bounded tasks and combines worker outputs
Explanation
Correct Answer
A:Completion of one branch does not imply total workflow completion. The coordinating stage must await and interpret the validation dependency before releasing downstream data.
Incorrect Answers
B:Because when branches have no data dependency, parallel execution reduces waiting time while leaving the coordinator responsible for combining results and handling any branch failure, the measure supports parallelizing independent evidence gathering rather than the needed awaiting validation across distributed workers.
C:Agent delegation has token, latency and observability costs; its benefits for avoiding needless specialist agent invocations do not resolve the need for awaiting validation across distributed workers.
D:A central orchestrator can define evidence requirements, route work by specialty and reconcile returned results; supporting centralized multi-agent coordination cannot replace the required awaiting validation across distributed workers.
Question 76
An agent's summary says only 'refund approved' but omits who approved it and when. What structured fields should remain?
- Save claim-to-source mappings alongside the condensed findings
- Save the verified case ID and current resolution status as durable fields
- Fetch relevant passages on demand while retaining stable citations
- Retain approval identity, timestamp and supporting policy evidence
Explanation
Correct Answer
D:High-impact decisions require an audit trail, not just a terse conclusion. Attribution and evidence prevent later processes from treating unsupported approval text as authority.
Incorrect Answers
A:Compression should preserve provenance links for verifiable conclusions; the stated problem requires preserving authorization evidence through summarization rather than maintaining citations during context compression.
B:Stable identifiers and status survive conversational paraphrasing; a method suited to preserving stable case identity across turns does not meet the need for preserving authorization evidence through summarization.
C:For using targeted retrieval to manage context limits, selective retrieval focuses tokens on the current question and leaves large source archives outside the prompt; however, this requirement concerns preserving authorization evidence through summarization.
Question 77
A scheduling tool uses an untyped free-text date field and frequently schedules the wrong day. What is the best interface improvement?
- Use an unambiguous timestamp representation with timezone requirements
- Document a stable output schema and version incompatible changes
- Require a stable idempotency key for mutation requests
- Publish a narrow parameterized order-status operation instead of general SQL
Explanation
Correct Answer
A:Machine-oriented time fields and explicit timezone semantics prevent interpretation drift. The business action should not depend on a model guessing regional date formatting.
Incorrect Answers
B:A predictable contract lets agents and validators consume results consistently; the deciding condition is normalizing temporal tool parameters, whereas this action concerns managing output schema compatibility.
C:A client-provided operation identifier lets the service recognize repeat submissions and return the prior outcome; the option would serve protecting a mutating tool against replay, not normalizing temporal tool parameters as specified in the stem.
D:A purpose-limited interface minimizes injection and unauthorized data access; this addresses constraining a tool to least-privilege business operations, while the question is about normalizing temporal tool parameters.
Question 78
Security needs a repeatable record of every call to the protected integration. What implementation is more reliable than prompt reminders?
- Decline over-budget requests in the tool-call authorization boundary
- Tighten the hook matcher to the intended tool name or invocation pattern
- Implement protected-path checks in a pre-tool permission handler
- Implement a PreToolUse hook that denies matching unsafe invocations
- Use a tool lifecycle hook to emit structured audit events
Explanation
Correct Answer
E:Hooks provide a programmatic observation point around matching tool invocations. A system prompt requesting self-reporting cannot independently verify every external action.
Incorrect Answers
A:When considering enforcing a quota at the tool boundary, tracking usage and checking it before execution makes a budget enforceable; the requirement for recording tool activity independently of model narration remains unmet.
B:When addressing targeting a hook to relevant operations, matchers determine which hook executions apply; the present task instead calls for recording tool activity independently of model narration.
C:A programmatic check can compare the requested path to restricted locations before access occurs; denying sensitive file reads before access and recording tool activity independently of model narration require different controls in this situation.
D:PreToolUse runs before a tool call and can participate in the decision to block it; the unsatisfied requirement is recording tool activity independently of model narration, not blocking disallowed commands before tool execution.
Question 79
A repository refactor renames src/api to packages/server but the path rule still targets src/api. What should change?
- Put globally applicable policy in an always-loaded or enforced layer
- Constrain the rule's paths glob to target only migration files
- Update rule globs to the new directory structure in the same refactor
- Check rule glob matches against real repository paths
Explanation
Correct Answer
C:Conditional instruction matching depends on current paths. Stale selectors silently stop influencing the code locations they were created to protect.
Incorrect Answers
A:A conditional file rule may not activate in every session; the action supports keeping organization-wide policy outside conditional globs but falls short of maintaining globs after a directory move in this case.
B:A narrower glob controls the rule's applicability; this can serve narrowing a conditional rule's file scope, but leaves the requirement for maintaining globs after a directory move unmet.
D:A simple match audit reveals dead rules that never load; this concerns verifying rule selectors actually match files, leaving the requirement for maintaining globs after a directory move outstanding.
Question 80
A developer has approved a two-line typo fix with an exact target location. Which workflow minimizes unnecessary planning overhead?
- Report the architectural trade-off and request the missing business decision
- Display the plan for approval before allowing write-capable execution
- Determine specific tests and acceptance criteria alongside the edit steps
- Execute the narrow approved edit directly and run the relevant check
- Define deployment order, checkpoints and rollback triggers before execution
Explanation
Correct Answer
D:A bounded, well-specified low-risk change often needs no lengthy planning phase. Direct implementation is appropriate when the risk and scope are already understood.
Incorrect Answers
A:An unresolved design choice is not safe to infer from code alone; its benefits for resolving architecture ambiguity before editing do not resolve the need for implementing a tiny well-scoped change.
B:Sensitive migrations benefit from separating design review from mutation; supporting requiring plan approval for a migration cannot replace the required implementing a tiny well-scoped change.
C:A useful plan describes not just changes but the evidence that will establish success; this contributes to embedding verification into the implementation plan without meeting the need for implementing a tiny well-scoped change.
E: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 implementing a tiny well-scoped change.