Question 30 A rules file uses paths: ["tests/**"], but generated tests live in packages/service/tests. What should be reviewed? Whether the test files have more than 100 lines. Whether the glob covers nested test directories, then extend it to the intended paths. Whether the agent’s greeting mentions tests. Whether production database tokens are committed. Explanation Correct Answer B:Path patterns must match the actual repository layout. Incorrect Answers A:File length does not determine path matching. C:Greeting content does not load the proper scoped rule. D:Adding secrets is not a glob correction.
Question 31 Two engineers resume the same mutable agent session simultaneously and apply conflicting file edits. What is the architecture concern? Concurrent writers require coordination or isolated session forks. Both edits are guaranteed correct if generated by the same model. Session tokens automatically merge conflicting filesystem edits. The model’s vocabulary will become smaller. Explanation Correct Answer A:A single mutable session cannot safely assume serialized edits by independent actors. Incorrect Answers B:Shared model identity cannot prevent conflicting changes. C:A conversation identifier does not provide transactional conflict resolution. D:Word choice is not the primary hazard of concurrent state mutation.
Question 32 A document extractor finds a contract end date but cannot identify which page contains it. What provenance feature should be added? Return a confidence percentage instead of the page. Add a page number chosen at random. Record document ID, page or span and extracted value together. Omit the document ID to reduce output length. Explanation Correct Answer C:Location-level evidence lets reviewers trace each assertion. Incorrect Answers A:Confidence does not locate supporting text. B:That would fabricate provenance. D:The value would lose source association.
Question 33 A retrieval tool can return a document excerpt and a source URI, but the assistant cites only the excerpt as if it were self-authenticating. What output field should be emphasized? A generic “verified” flag generated by the retrieval model. An arbitrary confidence emoji with no source. Only the byte count of the excerpt. A stable source identifier or URI tied to the returned excerpt. Explanation Correct Answer D:Claims can then be traced to the actual supporting document. Incorrect Answers A:A self-declared label is not a traceable citation. B:A symbol provides no provenance. C:Length cannot identify evidence origin.
Question 34 A privacy reviewer often flags personal information even when the data are anonymous synthetic fixtures. What instruction could improve specificity? Define the identification standard and require evidence that data are real or linkable. Treat every string of digits as a real identity number. Declare all filenames containing test safe by default. Ignore all data handling because fixtures exist. Explanation Correct Answer A:The model must distinguish protected personal data from test values without linkage. Incorrect Answers B:A digit pattern alone is not conclusive. C:A name does not prove actual anonymization. D:Production PII may still appear in other files.
Question 35 A content pipeline runs classify → redact → publish. A classification tool returns “unknown” for an unusual document. What should orchestration do? Route unknown documents to review instead of publishing under a guessed classification. Ask the model to invent a likely category without recording uncertainty. Treat unknown as public so the pipeline continues. Skip redaction and let the publishing service decide. Explanation Correct Answer A:An unclassified document must not bypass a required privacy gate. Incorrect Answers B:Guessing conceals the missing evidence and defeats the review policy. C:Failing open could disclose sensitive content. D:The final service should not inherit an unresolved upstream protection decision.
Question 36 A code-generation agent fixes the reported bug but introduces an unrelated formatting change across 100 files. What should its next refinement optimize? Expand the patch to every directory for consistency. Skip tests because the functional bug is fixed. Delete version history so reviewers cannot see churn. Minimize the diff to the requested behavior and preserve unaffected files. Explanation Correct Answer D:Scope discipline improves reviewability and reduces regression risk. Incorrect Answers A:Broader changes magnify the unnecessary side effects. B:Unrelated changes may have introduced regressions. C:Hiding diffs undermines accountability.
Question 37 A developer asks Claude Code to find filenames matching **/*payment*.ts before opening files. Which built-in capability is the best fit? Glob to discover paths matching the filename pattern. Edit to replace code in an unknown file. Write to create a new file with that name. Bash to delete unmatched files. Explanation Correct Answer A:Glob targets path patterns without reading every file’s content. Incorrect Answers B:No target file has been identified for editing. C:Creating a file does not discover existing matches. D:Deletion is unnecessary and destructive for file discovery.
Question 38 A PR-review prompt should minimize false positives while preserving meaningful defects. Which TWO requirements most help? The model should ignore code context to reduce token costs. Every finding must identify a relevant changed code path and plausible failure mechanism. All findings must be labeled critical to attract attention. Every changed line should produce a comment. The review must state assumptions and avoid certainty when a critical dependency is not visible. Explanation Correct Answers B:Evidence-based scope filters unrelated observations. E:Explicit uncertainty prevents unsupported high-severity claims. Incorrect Answers A:Context is needed to judge whether the pattern is truly problematic. C:Misleading severity prevents prioritization. D:A quota encourages noise regardless of risk.
Question 39 Three research workers share one rate-limited external API. The coordinator starts all at once and receives throttling errors. What should scheduling account for? Ignoring 429 responses as successful empty results. A shared concurrency budget or rate limiter across those subagent tool calls. Increasing model temperature for each worker. Sending extra requests to outrun the limiter. Explanation Correct Answer B:Parallel tasks remain useful but must respect the shared backend capacity. Incorrect Answers A:A throttle is a retryable failure, not proof of no data. C:Sampling variability does not increase the external API quota. D:More requests can amplify throttling.
Question 40 An agent summarizes a disputed technical incident from incomplete records. Which TWO practices best preserve reliability? Convert all inferences into settled facts for a simpler story. Label each conclusion as confirmed, inferred or unresolved with supporting evidence. Suppress contradictory records that make the summary less concise. Record timestamps and source systems for critical indicators. Attach plausible but unvisited URLs to make the report appear sourced. Explanation Correct Answers B:Readers can distinguish observation from interpretation. D:Provenance allows independent reconstruction and checks for conflicting records. Incorrect Answers A:That overstates evidence. C:Relevant conflict must be addressed. E:Unverified citations are misleading.
