Question 290 A codebase has several modules named `client.py`, and the agent proposes editing one based on an ambiguous search hit. What is required? Assume all modules with the same basename are identical. Edit the first match returned by a search tool. Resolve the complete file path and inspect imports and callers to identify the actual module. Rewrite every matching module to be safe. Explanation Correct Answer C:Same-named files can serve different services and scopes. Incorrect Answers A:Identical names do not imply identical behavior. B:Search order does not determine ownership. D:That expands change scope unpredictably.
Question 291 A synthesis worker reports accurate claims but has no URLs attached after the researcher’s handoff. What should be carried forward? Only a list of unassociated homepage URLs. Structured claim-to-source links with dates or document locations. Only the researcher’s final confidence score. A copy of the coordinator prompt without findings. Explanation Correct Answer B:Provenance must survive a worker boundary for defensible citations. Incorrect Answers A:Unlinked sources do not show which evidence supports which claim. C:Confidence cannot recreate the original evidence. D:Instructions alone cannot support final statements.
Question 292 A backend-only rule is written with a wildcard that also matches frontend files. What should the team verify? That every file in the repo shares the same programming language. Only the number of lines in the rule file. The glob pattern against representative paths before relying on it. That the rule title uses uppercase letters. Explanation Correct Answer C:A mis-scoped pattern can introduce irrelevant or conflicting instructions. Incorrect Answers A:The scenario explicitly distinguishes backend and frontend. B:Length does not establish which files it matches. D:Naming style does not correct a path glob.
Question 293 A clinical note extractor is required to identify prescribed drugs, but it treats a patient’s family history as a prescription. What prompt criterion is missing? Guess prescriptions from diagnosis alone. Ignore sections that mention treatment entirely. Require attribution to the patient’s current treatment plan, not merely mention of a drug. Extract every medical term in the note as a medication. Explanation Correct Answer C:Role and temporal context distinguish prescriptions from incidental mentions. Incorrect Answers A:Inference without record evidence can be unsafe. B:That would omit actual prescriptions. D:Broad extraction does not satisfy the requested entity definition.
Question 294 A long support thread includes a product version, exact error code, attempted repairs and a current blocker. Which summary is most restart-safe? The same repair command repeated as the first step every session. A statement that the case is resolved despite the open blocker. A generic overview of what the product usually does. A concise state table of version, error, attempted steps with outcomes and the next diagnostic action. Explanation Correct Answer D:Preserving evidence and ruled-out hypotheses prevents repeated troubleshooting. Incorrect Answers A:Prior failed attempts should influence continuation. B:That misrepresents progress. C:It does not capture this customer’s observed state.
Question 295 A long-running agent needs to show that an unsafe command was denied by policy. What should its interception result include? A fabricated successful execution transcript. Only a chat message telling the user to trust the system. An explicit denial status and reason, correlated to the attempted call. An empty tool result with no status. Explanation Correct Answer C:The model and auditors must distinguish a policy rejection from a tool outage or success. Incorrect Answers A:False success can trigger further unsafe actions. B:A factual denial record is more reliable than a claim of safety. D:Ambiguity makes recovery and auditing difficult.
Question 296 A workspace contains a 50 MB generated log and a tiny config file. The agent needs only config values. Which approach is proportional? Write an invented config file and inspect it. Read every log line before opening the config. Read the specific config instead of dumping the entire log into context. Run destructive cleanup on the log to make it smaller. Explanation Correct Answer C:Targeted reads avoid unnecessary token and attention costs. Incorrect Answers A:Synthetic data do not reveal the real current settings. B:Large unrelated data cannot improve a simple config lookup. D:The request does not authorize log deletion.
Question 297 A skill contains environment-specific secrets in its checked-in prompt body. What is the proper remediation? Mask the secret only in the final assistant answer. Change the skill name to make it less discoverable. Tell users not to read the secret lines. Remove and rotate secrets; obtain credentials through secure runtime mechanisms. Explanation Correct Answer D:A skill may be shared or version-controlled and should not embed live tokens. Incorrect Answers A:The committed skill still exposes it. B:Renaming does not revoke a leaked credential. C:Disclosure is not prevented by instructions.
Question 298 A team discovers that its JSON schema uses features the current structured-output API does not support. What should happen? Turn every validation failure into a fabricated successful record. Delete application validation because the API has schema mode. Assume all JSON Schema draft features are supported automatically. Adapt the sent schema to supported constructs and validate any extra constraints in application code. Explanation Correct Answer D:The integration must honor API limitations without abandoning needed business checks. Incorrect Answers A:False data are worse than explicit contract failure. B:Business rules may exceed the supported schema subset. C:API subset limitations can reject an otherwise valid general schema.
Question 299 A content moderation agent outputs an internal risk score, but a downstream publish tool only sees the final text. How should the safety decision be passed? Run the score calculation only after publication. Pass a validated risk decision as explicit workflow state checked by the publishing step. Have publication infer risk from the number of words. Assume that silence from the model means safe publication. Explanation Correct Answer B:A typed gate avoids parsing free-form assistant text to infer approval. Incorrect Answers A:Post-hoc scoring cannot prevent an inappropriate release. C:Length does not establish a policy judgment. D:Missing a decision is not affirmative approval.
Question 300 A CRM MCP server is healthy, but the caller supplied a negative invoice total outside allowed validation. Which category is appropriate? A successful empty invoice. A nonretryable invalid-argument failure with the violated constraint. A retryable network timeout. A permission-granted status with no data. Explanation Correct Answer B:The input must be changed; retrying unchanged values is not a transient recovery. Incorrect Answers A:A malformed request is not evidence of an invoice with zero lines. C:The server is reachable and has rejected invalid business input. D:Validation is unrelated to authorization success.
