Question 190 A skill is intended to read repository files but not execute shell commands. Which metadata is most relevant? A promise that the model will never be tempted to run Bash. A request for unrestricted production credentials. An allowed-tools declaration limited to read-oriented capabilities. An arbitrary font name in the skill file. Explanation Correct Answer C:Tool scoping lets the skill operate with least privilege. Incorrect Answers A:A prompt promise is weaker than an enforceable capability boundary. B:The read task does not require them. D:Presentation choices do not constrain tool use.
Question 191 A browser-run assistant has long tool traces but only a small amount of current task state matters. What is the best durable checkpoint? Store only the most recent assistant greeting. Store a list of guesses about future tool results. Store only the elapsed execution time. Store the objective, verified completed work, open tasks, relevant file IDs and constraints. Explanation Correct Answer D:A checkpoint can resume precisely without replaying every old tool output. Incorrect Answers A:A greeting lacks task state. B:Guessed future outcomes do not preserve verified progress. C:Duration does not identify what work remains.
Question 192 A compliance assistant flags an invoice as missing a signature even though the applicable process accepts digitally signed records. How should criteria be revised? Delete the signature requirement from all documents. Specify acceptable evidence types for the governing process, including valid digital signatures. Require wet ink on every document regardless of policy. Ask for a confident yes/no without citing accepted methods. Explanation Correct Answer B:The prompt should test the actual compliance rule rather than a paper-only assumption. Incorrect Answers A:Some workflows genuinely require verified authorization. C:That imposes an unsupported restriction. D:A binary assertion does not establish compliance.
Question 193 A user authorizes a data-cleaning job but does not authorize deletion of original source files. An agent discovers that deletion would save disk space. What is appropriate? Mark the files deleted even though they were kept. Delete originals automatically after deriving cleaned outputs. Ask a subordinate model to authorize deletion on the user’s behalf. Preserve the source files and request separate approval before destructive cleanup. Explanation Correct Answer D:Efficiency does not expand the authorized scope. Incorrect Answers A:A false report undermines trust. B:The user did not authorize destruction. C:Delegation does not grant new user consent.
Question 194 A PR review agent can read the patch but has permission to merge into main. What tool boundary is appropriate? Put the main-branch deployment token into the review prompt. Disable code review logs. Allow every review finding to trigger an automatic merge. Give the reviewer read-only repository access and separate authorized merge capability. Explanation Correct Answer D:Review does not itself imply permission to change protected branches. Incorrect Answers A:Prompt access to secrets is unnecessary and risky. B:An audit trail helps investigate unexpected suggestions. C:A model assessment should not bypass release controls.
Question 195 A customer asks, “Assess the risks of moving our operations to cloud.” The coordinator assigns only storage security and access control. What is missing? A coverage plan that includes operational, financial, compliance, resilience, and security dimensions as relevant. A lower temperature in the storage worker. An instruction to the final writer to invent the other dimensions. More detail about access control while ignoring resilience. Explanation Correct Answer A:Broad questions require explicit breadth criteria before specialist dispatch. Incorrect Answers B:Sampling settings cannot recover omitted dimensions. C:Missing work cannot be replaced by speculation. D:Deeper narrow analysis does not solve the scope omission.
Question 196 An MCP service is moved from local stdio transport to a hosted remote endpoint. What architecture change must be verified? The client registration, authentication, transport details, and trust of the remote endpoint. A blanket instruction to ignore all TLS errors. The model’s output word limit. Only the color theme in Claude Code. Explanation Correct Answer A:A hosted service requires reachable connection settings and identity checks. Incorrect Answers B:Bypassing transport security undermines the new hosted integration. C:Word limits are unrelated to transport or authentication. D:Theme settings do not establish an MCP connection.
Question 197 Two independent reviewers disagree about whether a new API route bypasses authorization. What is the best adjudication? Count their words and trust the longer answer. Ask the first reviewer to overwrite the second’s notes. Publish both labels as definitive simultaneously. Trace the route to its actual authorization middleware and run a focused negative-permission test. Explanation Correct Answer D:Executable evidence resolves the disputed access boundary. Incorrect Answers A:Verbosity does not prove an access check exists. B:Suppressing dissent does not evaluate the code. C:Contradictory verdicts require resolution.
Question 198 A source says a product feature “may be available in some regions” but the assistant writes “available everywhere.” What is the sound correction? Choose one unsupported country and claim it was tested. Remove the source and keep the categorical claim. Preserve the source’s qualified regional scope and state that availability outside it is unverified. Leave the universal assertion because it is easier to read. Explanation Correct Answer C:Claims must not exceed the evidence’s geographic certainty. Incorrect Answers A:That invents an availability check. B:That hides a material limitation. D:Clarity does not justify overclaiming.
Question 199 A developer wants to try a risky alternative refactor while keeping a known-good agent conversation intact. Which session operation fits? Delete the original branch before experimenting. Resume and rewrite the original conversation in place. Fork the existing session into an isolated alternative branch. Reuse the original session ID for two simultaneous conflicting writes. Explanation Correct Answer C:A fork preserves the source trajectory while permitting experiments without overwriting it. Incorrect Answers A:That removes the baseline needed for comparison and rollback. B:This may destroy the known-good decision path. D:Concurrent mutation risks state collisions.
Question 200 An agent’s tool list contains two tools with the same side effect but different billing permissions. How should it avoid privilege confusion? Rotate the display order daily and hope the model learns. Describe both as unrestricted billing updates. Always select the more privileged tool. Name and document the authorization boundary explicitly, and expose only the appropriate tier to each agent. Explanation Correct Answer D:Tool scope and descriptions must distinguish restricted from privileged operations. Incorrect Answers A:Tool ordering is not a permission mechanism. B:Misleading descriptions erase the security difference. C:Privilege should be minimized to what the operation needs.
