Question 170
A transcription feature must expose confidence limitations to operators rather than invent certainty. What should the output model include?
- Represent ambiguous spans and review-required statuses explicitly.
- Validate business rules and user authority in the payment service before executing the write.
- For this workflow, escape or sanitize untrusted output according to its display context.
- Evaluate outputs on the same evaluation suite and roll out version changes deliberately.
Explanation
Correct Answer
A:A structured uncertainty path gives operations a safe alternative to presenting unclear content as verified fact.
Incorrect Answers
B:The model's recommendation is an input, not a permission grant; the execution boundary must independently reject unauthorized actions; testing uncertainty-aware transcription output would still fail after improving model-to-payment boundary.
C:Model and tool content can contain scripts or markup; safe rendering is an application responsibility independent of the prompt; the scenario's controlling requirement is uncertainty-aware transcription output, not the related multi-criterion integration concern.
D:Even related model releases can differ in instruction following and tool choices; region parity requires testing; this implements cross-region model behavior parity rather than the demanded uncertainty-aware transcription output.
Question 171
A model-driven search UI needs the last five user turns but not the previous five hundred. How should the client construct context?
- Curate relevant recent turns and a summary of durable earlier constraints.
- Hold operational traces separate from the curated user-facing answer.
- Supply stable document IDs, versions, and location metadata with retrieved excerpts.
- Put in place defined retention limits, consent and access controls to conversation records.
- At the application layer, scope stored conversation history to the authenticated session and tenant.
Explanation
Correct Answer
A:Explicit context assembly controls costs and stale instructions; the API does not autonomously decide which history is authoritative.
Incorrect Answers
B:Debugging data is valuable to operators, but the product should render a purpose-built result rather than indiscriminately expose internal steps; it follows separation of traces from UI constraints but leaves selective conversation assembly unspecified.
C:Provenance makes citations and later audits traceable to the exact revision rather than an undifferentiated text blob; even if versioned retrieval provenance succeeds, selective conversation assembly still needs separate verification.
D:More stored context is not inherently better; privacy obligations and data minimization require purposeful retention; when the goal is selective conversation assembly, optimizing conversation retention governance misses that operational boundary.
E:The application chooses which messages to resend; incorrect session keys become a direct privacy and correctness defect; testing selective conversation assembly would still fail after improving conversation-session identity.
Question 172
An application needs a JSON field with an account ID that the model might guess. What design prevents invented identifiers?
- Resolve account IDs through trusted lookup and validate any generated reference against it.
- Set retrieval as untrusted task data and block tool authority from crossing that boundary.
- Audit minimal pseudonymous operational metadata and keep content in approved secure storage.
- Assess outputs on the same evaluation suite and roll out version changes deliberately.
Explanation
Correct Answer
A:Language models can produce syntactically plausible but nonexistent IDs; authoritative resolution must occur outside generation.
Incorrect Answers
B:Documents can supply facts, not governing instructions; privilege decisions must remain in trusted application logic; when the goal is verified identifier resolution, optimizing document-instruction trust boundary misses that operational boundary.
C:Diagnostics and product analytics have different data needs; leaking raw conversation material into telemetry broadens exposure; for this request, verified identifier resolution is the missing piece, not privacy-preserving analytics.
D:Even related model releases can differ in instruction following and tool choices; region parity requires testing; a remedy for cross-region model behavior parity is not a substitute for verified identifier resolution.
Question 173
A service exposes a plugin ecosystem to enterprise users. What security boundary must installation enforce?
- Inspect again authorization before loading history or calling privileged tools.
- Require reviewed permissions and scoped access for each connector before activation.
- Furnish only the necessary authorized account fields for each request.
- Document concise content sections and accessibility constraints, then test them in the client UI.
Explanation
Correct Answer
B:Plugins may introduce new external data and actions; install-time authorization prevents implicit privilege expansion.
Incorrect Answers
A:Long-lived chat IDs do not preserve access rights; current identity and permissions must gate each request; the scenario's controlling requirement is plugin permission review, not the related per-turn authorization concern.
C:Reliable context can be loaded from authoritative systems while minimizing sensitive disclosure and token bloat; for this request, plugin permission review is the missing piece, not minimal authoritative account context.
D:Model instructions should align with display requirements, but interface rendering must verify readability and semantics; this may help accessible response design but cannot by itself deliver plugin permission review.
Question 174
An organization wants to support both read-only search and high-impact account deletion. What API design reduces risk?
- State ambiguous spans and review-required statuses explicitly.
- Enforce reviewed permissions and scoped access for each connector before activation.
- Preserve user-provided content in a separate user or quoted data field.
- Expose separate tools with distinct permissions and explicit approval for deletion.
Explanation
Correct Answer
D:Capability partitioning makes read-only interaction safer and turns irreversible actions into auditable, independently authorized operations.
Incorrect Answers
A:A structured uncertainty path gives operations a safe alternative to presenting unclear content as verified fact; this targets uncertainty-aware transcription output, while the system must demonstrate capability-separated tool interfaces.
B:Plugins may introduce new external data and actions; install-time authorization prevents implicit privilege expansion; the capability-separated tool interfaces criterion differs from plugin permission review; the gap persists.
C:Combining untrusted input with the highest-priority prompt surface risks instruction confusion and makes auditing difficult; a remedy for separation of user data and policy is not a substitute for capability-separated tool interfaces.
Question 175
A product automatically persists every model conversation indefinitely 'in case it helps later.' What should its storage policy do?
- Document a constrained status enum and validate it at the application boundary.
- Model ambiguous spans and review-required statuses explicitly.
- Log reviewer decisions with provenance and feed them into a versioned evaluation dataset.
- Apply defined retention limits, consent and access controls to conversation records.
Explanation
Correct Answer
D:More stored context is not inherently better; privacy obligations and data minimization require purposeful retention.
Incorrect Answers
A:An explicit schema makes downstream branching predictable; free-form labels require fragile string interpretation; this implements finite machine-readable statuses rather than the demanded conversation retention governance.
B:A structured uncertainty path gives operations a safe alternative to presenting unclear content as verified fact; a remedy for uncertainty-aware transcription output is not a substitute for conversation retention governance.
C:Human adjudication supplies labeled evidence; keeping labels and source examples traceable makes quality improvements measurable; it changes reviewer feedback into evals; it does not prove conversation retention governance in this case.
Question 176
Two developers get different Claude Code behavior on the same repository because they have different personal instructions. What should be committed for shared project guidance?
- Examine which instruction files load for the current working path and resolve conflicts explicitly.
- Release schema and consumer changes with compatibility tests under a coordinated release.
- Place team-wide rules in the repository's CLAUDE.md and review changes.
- As an alternative, manage prompt templates as versioned deployment artifacts.
- Lock a tested model version where the platform provides versioned identifiers.
Explanation
Correct Answer
C:Versioned project instructions are visible to collaborators, whereas personal configuration can vary across machines.
Incorrect Answers
A:File location and instruction hierarchy affect applicable context; copying more directions into chat can mask the root conflict; this targets scoped instruction hierarchy, while the system must demonstrate shared project instruction source.
B:Generated output contracts span both sides of the interface; configuration drift can break consumers even when API calls succeed; the shared project instruction source criterion differs from schema release coordination; the gap persists.
D:Reviewed, traceable templates connect observed behavior to a release and allow reversal of regressions; changing prompt template change tracking does not establish shared project instruction source in this application.
E:An unpinned alias can change behavior without an application deployment; version pinning enables controlled evaluation and rollback; a remedy for stable production model identity is not a substitute for shared project instruction source.
Question 177
A production request unexpectedly changes quality when a model alias advances. What configuration practice would reduce this risk?
- Pin a tested model version where the platform provides versioned identifiers.
- Put team-wide rules in the repository's CLAUDE.md and review changes.
- Mandate an independent diff review, scoped test results, and human approval for sensitive changes.
- Isolate versioned static instructions from typed runtime variables.
Explanation
Correct Answer
A:An unpinned alias can change behavior without an application deployment; version pinning enables controlled evaluation and rollback.
Incorrect Answers
B:Versioned project instructions are visible to collaborators, whereas personal configuration can vary across machines; the consequence is improved shared project instruction source, not verified stable production model identity.
C:Generated code remains subject to ordinary repository controls; reviewing the actual file changes and tests prevents an assistant's confident summary from granting merge authority; because human review gate for AI-generated changes is its concern, stable production model identity remains outside its effect here.
D:Controlled templates make prompt revisions auditable, and validated variables reduce accidental changes to governing instructions; this may help static versus dynamic prompt values but cannot by itself deliver stable production model identity.
Question 178
A team changes prompt text manually in the production console without recording it. Which control improves reproducibility?
- Manage prompt templates as versioned deployment artifacts.
- Maintain revisions of or pin the connector dependency and run tool-contract regression tests.
- Log model identifier, prompt revision, tool schema version, and safe request correlation IDs.
- Insist on an independent diff review, scoped test results, and human approval for sensitive changes.
Explanation
Correct Answer
A:Reviewed, traceable templates connect observed behavior to a release and allow reversal of regressions. Version records should link the template, the test results, the approval decision, and the deployment where it was active.
Incorrect Answers
B:Tool schemas are part of the application interface; uncontrolled plugin updates can break dispatch or model selection; the prompt template change tracking criterion differs from connector version compatibility; the gap persists.
C:Behavior depends on a combination of versioned inputs; isolated text logs may not identify which configuration was active; testing prompt template change tracking would still fail after improving inference incident provenance.
D:Generated code remains subject to ordinary repository controls; reviewing the actual file changes and tests prevents an assistant's confident summary from granting merge authority; even if human review gate for AI-generated changes succeeds, prompt template change tracking still needs separate verification.
Question 179
A Claude Code repository needs different default permissions for local experimentation and CI. What should configuration separate?
- Specify the server command or endpoint and its needed scoped credentials in the appropriate project settings.
- Record both version identifiers and compare them on the same evaluation set before switching traffic.
- Review which instruction files load for the current working path and resolve conflicts explicitly.
- Use environment-appropriate settings and restrict the CI permission mode.
Explanation
Correct Answer
D:Automation runs without the same human supervision as local sessions, so its authority should be explicitly narrower.
Incorrect Answers
A:MCP clients need a connection definition and authentication path; a tool name alone does not create a working integration; this targets project MCP connection configuration, while the system must demonstrate environment-specific Claude Code permissions.
B:Behavior changes across model versions are deployment changes; a tested rollout prevents silent production regressions; when the goal is environment-specific Claude Code permissions, optimizing multi-criterion integration misses that operational boundary.
C:File location and instruction hierarchy affect applicable context; copying more directions into chat can mask the root conflict; the scenario's controlling requirement is environment-specific Claude Code permissions, not the related scoped instruction hierarchy concern.
Question 180
A secret needed by a custom tool is written directly in settings.json checked into Git. What should change?
- Reference a secure external credential source instead of committing secret values.
- Verify environment-specific endpoint mappings and credentials before deployment.
- Position team-wide rules in the repository's CLAUDE.md and review changes.
- Place the security requirement in a higher-priority managed policy and actual tool permissions.
Explanation
Correct Answer
A:Configuration files can be reviewed and copied; secret lifecycle belongs in a credential manager or protected environment.
Incorrect Answers
B:The correct tool schema can still cause damage when its endpoint is misconfigured; environment segregation must include targets and identity; the effect concerns environment endpoint isolation; the success condition is secret externalization.
C:Versioned project instructions are visible to collaborators, whereas personal configuration can vary across machines; even if shared project instruction source succeeds, secret externalization still needs separate verification.
D:Advisory instructions can conflict; enforceable organizational controls must not be overridden by user-local customization; where secret externalization matters, a control for nonoverridable organizational policy leaves that test unmet.