Question 270 An agent is fixing an API authorization regression in a monorepo. Which TWO context choices are most useful? Read the endpoint’s route, middleware and service path rather than every package. Ignore repository-specific conventions and tests. Inspect the relevant permission tests and recent diff for the failing behavior. Use a similarly named file from an unrelated project. Load every generated dependency into the prompt. Explanation Correct Answers A:The targeted call chain exposes enforcement boundaries. C:Tests and changed lines ground a reproducible fix. Incorrect Answers B:That risks incompatible edits. D:Cross-project context is not evidence for this repository. E:That consumes context with low-value data.
Question 271 A payroll tool returns HTTP 200 and an MCP content field saying “budget exceeded” with isError=true. How should the caller interpret it? Suppress the error before the model sees it. Repeat the payment without checking whether limits changed. Treat it as a failed business action despite successful transport. Treat any 200 response as a completed salary payment. Explanation Correct Answer C:Tool-level semantics can indicate failure independent of HTTP success. Incorrect Answers A:The model needs the failure to choose a safe next step. B:Blind retries do not resolve a budget restriction. D:Transport success does not override an explicit tool error.
Question 272 A financial assistant calls a payment tool twice after a network timeout, causing duplicate charges. Which loop safeguard directly protects the business operation? Retry more aggressively with exponential backoff alone. Hide duplicate-payment errors from the model. Increase the tool description to 1,000 words. Give the payment command an idempotency key and reconcile outcome before retrying. Explanation Correct Answer D:An uncertain response must not cause an irreversible action to be repeated blindly. Incorrect Answers A:Backoff changes timing but cannot prevent duplicate writes. B:Suppressing the error makes reconciliation harder and leaves customers charged twice. C:Better wording cannot guarantee exactly-once payment under network uncertainty.
Question 273 A slash command can be invoked but returns inconsistent output headings across developers. What should be standardized? A rule that commands must never use parameters. The model’s greeting length only. The skill or command prompt template, input parameters, and required output structure. Each developer’s random punctuation style. Explanation Correct Answer C:A shared invocation contract produces consistent deliverables across sessions. Incorrect Answers A:Useful commands may need explicit target inputs. B:Greeting text does not define release-note sections. D:Uncontrolled variation makes automation harder.
Question 274 A model returns syntactically valid JSON containing a supplier ID that does not exist. Which validation layer catches this before an invoice is created? Trust the value because the JSON parser accepted the object. Check the supplier ID against the authoritative supplier registry before executing the write. Change the ID to a random known supplier. Ask the model to express confidence as a percentage. Explanation Correct Answer B:Referential integrity is a business invariant beyond JSON syntax. Incorrect Answers A:Parsing proves shape, not existence of an entity. C:Substitution could misdirect the payment. D:A confidence number is not registry verification.
Question 275 A developer adds an MCP endpoint that can export confidential data but lets every workspace access it. Which integration control is needed? Remove all logging for the endpoint. Restrict server permissions and client scope to authorized projects and users. Publish the endpoint’s access token in a global config. Assume developers will not invoke dangerous tools. Explanation Correct Answer B:An MCP connection is a capability boundary, not merely a convenience plugin. Incorrect Answers A:Lack of telemetry makes unexpected exports harder to detect. C:An exposed token broadens unauthorized access. D:Available capabilities can be called accidentally or maliciously.
Question 276 An extraction worker returns a customer identifier without the source document reference, and a pricing agent uses it to quote a contract. What would improve the chain? Send the quote without identifying the customer. Require the identifier to carry a source reference and validate customer identity before pricing. Assign a random identifier if evidence is missing. Trust identifiers generated by any worker automatically. Explanation Correct Answer B:Provenance prevents an ungrounded intermediate value from becoming a financial assertion. Incorrect Answers A:That obscures the target of a consequential business action. C:Guessing can misassociate accounts. D:Inter-agent handoff cannot certify a fact.
Question 277 A team stores a code-review skill inside a private developer folder, so colleagues cannot invoke it. What should be changed? Tell every developer to copy the author’s whole home directory. Ask Claude to invent a new skill every time. Require admins to paste the entire contents into chat. Place the reviewed skill in a project-shared location with agreed invocation and permissions. Explanation Correct Answer D:Reusable team workflows must be discoverable from the shared repository. Incorrect Answers A:That would expose private settings and does not offer stable versioning. B:Repeated generation defeats consistency. C:Manual copying is unnecessary overhead.
Question 278 Several MCP tools return dates in Unix seconds, ISO strings, and local time without an offset. What should a normalization hook guarantee? Convert unambiguous values into a canonical timezone-aware representation and flag ambiguous values. Strip all timezone offsets before passing dates to the model. Assume every numeric timestamp is milliseconds. Ask the final report writer to guess how each vendor formats dates. Explanation Correct Answer A:Normalization needs both consistent units and explicit handling of missing timezone information. Incorrect Answers B:Removing offsets makes ordering across systems ambiguous. C:That may incorrectly scale sources that supply seconds. D:Late guessing cannot reliably undo reasoning based on inconsistent timestamps.
Question 279 A team wants every developer in one repository to use a shared local MCP test server configuration, but not affect unrelated projects. Where should the checked-in configuration live? Only in a prose README that the client never reads. In the project-scoped .mcp.json with reviewed server definitions. In the production secrets file committed beside source code. In each user’s unrelated global preferences with no version control. Explanation Correct Answer B:Repository-scoped MCP configuration can be shared within the project. Incorrect Answers A:Documentation alone does not register an MCP server. C:Credentials must not be published to the repository. D:Private per-user files do not provide a repeatable team project setup.
Question 280 A repository contains 1,500 files, and a model is losing the relevant symbol after reading massive generated bundles. What should it exclude? All code directly defining the target symbol. All tests even when debugging a failing test. The current git diff because it is newer. Generated dependencies and unrelated build artifacts unless the task specifically requires them. Explanation Correct Answer D:Filtering noise preserves budget for authoritative source code. Incorrect Answers A:That removes the core task evidence. B:Tests can be essential evidence. C:The diff often identifies the actual changed behavior.
