Question 210
A staging configuration accidentally points to production write tools. What deployment control addresses this?
- Assuming test users cannot trigger real writes.
- Changing only the page banner to “staging.”
- Explicit environment bindings verified before enabling write execution.
- Using smaller test inputs against the same production tools.
Explanation
Correct Answer
C:The environment label must correspond to actual endpoints and credentials, especially for mutations. Use environment-specific identities and endpoint validation so staging cannot reach production mutation capabilities merely because a URL was copied.
Incorrect Answers
A:Actual permissions and endpoints determine effects, not the user’s informal role label.
B:A banner does not redirect backend operations away from production.
D:Small changes can still modify real production records.
Question 211
Who changed a safety-relevant setting and when? Which record best answers this operational question?
- An authenticated configuration-change audit linked to old and new values.
- Only the file's latest modification timestamp.
- A generated explanation guessing the administrator.
- Only the current setting value.
Explanation
Correct Answer
A:An append-only or appropriately protected change history should connect the actor, timestamp, previous value, and new value to the configuration event.
Incorrect Answers
B:A timestamp alone does not identify the actor, previous value, or exact change responsible for the current setting.
C:A plausible identity guess is not an audit record.
D:The final value does not identify its author or prior state.
Question 212
A feature rollout assigns 10% of users to a new prompt. The same user must remain in one variant across requests while the rollout configuration is unchanged. Which assignment method fits?
- Use a stable user-based bucket and the configured rollout threshold.
- Switch a user to the new prompt whenever a response is slow.
- Assign the new prompt to every tenth request globally.
- Draw a fresh random value independently for every request.
Explanation
Correct Answer
A:A deterministic bucket preserves a user's cohort across requests while allowing the threshold to control exposure. Per-request randomness would mix prompt variants within the same user's workflow.
Incorrect Answers
B:Response latency is not the configured cohort rule. This introduces outcome-dependent assignment and breaks the unchanged-rollout guarantee.
C:Request ordering does not provide stable membership for an individual user and can change under concurrent traffic.
D:The long-run share may approach 10%, but a single user can alternate between variants. That violates the stated consistency requirement.
Question 213
A CI job must ask Claude Code to perform a noninteractive task and return output. Which interface is appropriate?
- A plain text file that is never passed to Claude Code.
- Claude Code’s documented print/headless invocation, such as claude -p.
- The Messages API’s model field set to a shell command.
- Only an interactive menu awaiting manual input.
Explanation
Correct Answer
B:Headless execution supports programmatic use without depending on an interactive terminal conversation. The automation should also choose a supported output format and check the process result so noninteractive execution remains machine-consumable.
Incorrect Answers
A:An unused file does not invoke the agent.
C:API model selection is not a Claude Code CLI invocation mechanism.
D:A waiting terminal prompt prevents unattended CI completion.
Question 214
A program consumes incremental Claude Code CLI events. What output configuration should it select?
- Assume CLI events are identical to raw API SSE frames.
- The shortest available human summary only.
- Screen-scraped terminal colors.
- The CLI’s supported structured streaming output format.
Explanation
Correct Answer
D:Machine-readable events provide a defined parsing contract; terminal prose is designed primarily for people. Parse the documented event schema, including completion and error signals, instead of depending on terminal decoration that may change between versions.
Incorrect Answers
A:Different interfaces can wrap events differently; use the CLI’s documented format.
B:A final summary omits the incremental events the program needs.
C:Presentation escapes are not a reliable semantic event interface.
Question 215
A repository has specialized CLAUDE.md instructions in a subdirectory. What should users understand about scope?
- A nested file applies automatically to every unrelated repository.
- Every nested instruction necessarily replaces all root instructions.
- Nested instructions can apply when Claude works with files in that subtree.
- The filename itself grants extra filesystem privileges.
Explanation
Correct Answer
C:Directory-specific guidance supports local conventions without placing every detail into the root instruction file.
Incorrect Answers
A:Its directory scope does not establish global authority.
B:Instruction loading can combine scopes rather than erase broader guidance.
D:Instruction files guide behavior; they do not independently confer operating-system permissions.
Question 216
Only frontend files should receive a particular style rule. What organization mechanism fits Claude Code?
- Path-scoped project rules with the appropriate file matching configuration.
- A global instruction covering every project unconditionally.
- Renaming backend files as frontend files.
- Putting the rule only in an unread build artifact.
Explanation
Correct Answer
A:Targeted rules limit irrelevant instruction loading while applying conventions where they matter. Match rules to the relevant paths through the supported rule metadata; file organization alone does not enforce which guidance is loaded.
Incorrect Answers
B:That broadens the rule beyond the stated frontend scope.
C:Changing labels does not provide a sensible scope mechanism.
D:A file must be part of the supported instruction mechanism to guide behavior dependably.
Question 217
Several Claude Code sessions exist. A developer must resume one particular earlier task. What should be used?
- Start a fresh session and assume old state appears automatically.
- Always continue the latest session regardless of identity.
- Copy only the previous terminal window title.
- The documented resume mechanism identifying that session.
Explanation
Correct Answer
D:Session identity matters when multiple conversations exist, because a general continue command may select a different recent task.
Incorrect Answers
A:New context does not guarantee recovery of the selected conversation.
B:Latest and intended are not necessarily the same task.
C:A display title is not necessarily the supported session identifier or full state.
Question 218
A developer wants Claude Code to inspect a repository and propose changes before implementation. Which workflow fits?
- Grant unrestricted edits and review only after deployment.
- Skip repository inspection and infer implementation from filenames.
- Ask for a plan after deleting the relevant source files.
- Use its planning workflow and review the proposed approach before permitting edits.
Explanation
Correct Answer
D:Planning separates investigation from execution so the developer can assess the intended changes first. The planning checkpoint is useful only if implementation remains gated until the proposed changes have been reviewed under the requested workflow.
Incorrect Answers
A:That performs changes before the requested planning checkpoint.
B:Filenames alone may not establish dependencies or current behavior.
C:Removing evidence prevents useful investigation and is unrelated to planning.
Question 219
A project defines a Claude Code Skill for release review. How can a developer deliberately request it?
- Set an environment variable whose value happens to match the Skill name.
- Assume installing it means it must execute on every turn.
- Invoke its documented slash-command name or explicitly request the skill.
- Change the model’s API key to the skill name.
Explanation
Correct Answer
C:Naming the intended workflow helps deliberate selection; installation merely makes the capability available rather than requiring it on every task.
Incorrect Answers
A:A matching string in environment configuration is not the documented user invocation interface for that workflow.
B:Availability does not imply unconditional invocation for unrelated tasks.
D:Credentials authenticate access; they do not select task instructions.
Question 220
How does Claude Code auto memory differ from a deliberately authored project CLAUDE.md?
- CLAUDE.md can never be changed by its maintainers.
- Auto memory is a guaranteed permission-enforcement engine.
- Auto memory is identical to retraining the model’s weights.
- Auto memory records learned context, while CLAUDE.md contains maintained instructions.
Explanation
Correct Answer
D:The distinction concerns authorship and purpose; both need review when stale or inappropriate information affects behavior.
Incorrect Answers
A:The project can update its instruction file through its normal review process.
B:Remembered context is not a substitute for execution controls.
C:Stored contextual notes are not a model-training operation.