Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 4

Question 30

A team wants repeatable Claude Code workflows for contributors. Developers repeatedly invoke a fixed changelog-generation routine by name. Which approach satisfies the stated constraint?

  1. Design the command to accept and validate a module argument
  2. Put the command in the user's personal Claude Code command scope
  3. Package the instructions and supporting resources as a Claude Code Skill
  4. Create a project-scoped custom command with clear inputs and instructions
  5. Keep a concise skill overview and reference supplemental files as needed

Explanation

Correct Answer

D:

A named command makes an explicit routine easy to invoke consistently without embedding the same instructions in every chat.

Incorrect Answers

A:

Parameterized workflows remain reusable while tying file operations to a caller-specified scope instead of assuming a default; the alternative problem occurs when a command needs a target module supplied by the developer. The present scenario instead requires the system to package a deliberate reusable command.

B:

It would be appropriate when an individual uses a private note-cleanup command in many projects; user-level commands support individual workflow shortcuts without adding unapproved tasks to team configuration. Here, developers repeatedly invoke a fixed changelog-generation routine by name, so a different mechanism is required.

C:

Skills can encapsulate specialized procedures with descriptions that support appropriate discovery and reuse; it protects against the case where a repository has a capability that Claude should discover when task context matches, rather than the failure described here: developers repeatedly invoke a fixed changelog-generation routine by name.

E:

This choice is relevant when a skill includes several bulky examples rarely needed for selection; progressive disclosure preserves context while still allowing detailed procedures to be consulted for the actual task. Here, however, developers repeatedly invoke a fixed changelog-generation routine by name.

Question 31

A release team operates Claude Code as part of automated change analysis. The pipeline prints a detailed narrative but downstream jobs need a verdict field. Which mechanism should be selected?

  1. Emit a fixed verdict schema and reject unparseable output
  2. Enforce fixed review instructions and disable unauthorized tools in the runner
  3. Compare findings against adjudicated examples and track precision and recall
  4. Re-run build and tests from the committed revision in a clean environment

Explanation

Correct Answer

A:

A deterministic parser must be able to identify pass, needs-review, and blocked states without interpreting conversational language.

Incorrect Answers

B:

This choice is relevant when a pull request contains deliberately malicious prompt text in documentation; untrusted files can attempt to override the reviewer's goals, so executable permissions and instruction boundaries must remain separate. Here, however, the pipeline prints a detailed narrative but downstream jobs need a verdict field.

C:

CI model output can drift; longitudinal labeled evaluation measures whether the review remains useful rather than merely verbose; the alternative problem occurs when the model-generated review itself must be measured for false positives over time. The present scenario instead requires the system to convert the output to a validated structured record.

D:

Verification from a clean checkout detects hidden local state and makes results reproducible across runners; it protects against the case where an agent proposes a code change that CI cannot reproduce in a fresh checkout, rather than the failure described here: the pipeline prints a detailed narrative but downstream jobs need a verdict field.

Question 32

A monorepo uses different coding guidelines for backend and frontend files. TypeScript files under web/ require browser-specific linting rules. What should be implemented at the appropriate boundary?

  1. Scope the testing guidance to Python service paths
  2. Correct the rule's glob or path targeting to match actual edited files
  3. Create a path-scoped rule targeting the web source patterns
  4. Narrow the patterns or align their formatter requirements

Explanation

Correct Answer

C:

Path-specific guidance appears for relevant work instead of affecting backend files that use different tooling. For the stated case, the check is whether the system can apply instructions based on file location; this approach supplies that control.

Incorrect Answers

A:

A service-specific rule prevents Claude from running a frontend command on an unrelated backend change; it protects against the case where python services require an alternate unit-test command, rather than the failure described here: typeScript files under web/ require browser-specific linting rules.

B:

Path-scoped rules depend on matching paths; the model cannot follow instructions that the configuration never selects; the alternative problem occurs when a rule never loads because its path pattern misses nested packages. The present scenario instead requires the system to apply instructions based on file location.

D:

This choice is relevant when two path rules overlap and recommend incompatible formatters; overlapping rule sets should lead to one predictable tool choice; unresolved conflicts create patch churn. Here, however, typeScript files under web/ require browser-specific linting rules.

Question 33

A multi-agent research controller launches short-lived specialists. A finance analyst must not obtain the searcher's CRM write tool. Which option best meets the operational need?

  1. Use a credential broker or application-side authorization rather than giving secrets in prompts
  2. Pass a curated task brief and the approved evidence only
  3. Assign each worker a purpose-specific allowed-tool set
  4. Require status, result, and unresolved-work fields in each worker handoff

Explanation

Correct Answer

C:

Tool permissions should follow delegated duties, not the maximum privileges of the parent or another worker. This matches the workflow goal to separate privileges per specialist rather than introducing an unrelated action.

Incorrect Answers

A:

This choice is relevant when an autonomous worker encounters an unexpected request for credentials; secrets placed in worker context are vulnerable to disclosure in logs or outputs; privileged access belongs in controlled service boundaries. Here, however, a finance analyst must not obtain the searcher's CRM write tool.

B:

Specialized workers benefit from an explicit information boundary; inherited unrelated history can confuse task scope and expose sensitive data; it protects against the case where a writing specialist receives the entire private search history, including unrelated leads, rather than the failure described here: a finance analyst must not obtain the searcher's CRM write tool.

D:

A typed completion status allows the parent to distinguish finished evidence collection from a partial exploratory result; the alternative problem occurs when the coordinator cannot tell whether a worker finished or merely started searching. The present scenario instead requires the system to separate privileges per specialist.

Question 34

A support team exposes several similar billing tools through an MCP server. Claude often confuses a read-only balance lookup with a credit-issuance action. What is the most appropriate design choice?

  1. Declare meaningful JSON Schema types, required fields, and relevant constraints
  2. Return a concise typed result with the required fields and stable meanings
  3. Describe the active-account limitation and expected error behavior in the tool contract
  4. Write descriptions stating purpose, side effects, eligibility, and safe-use boundaries

Explanation

Correct Answer

D:

Tool descriptions guide model selection; explicit mutation semantics reduce accidental use of a credit action for an informational request.

Incorrect Answers

A:

Clear argument schemas improve generated calls and let the host reject malformed or missing identifiers before sensitive execution; it protects against the case where a refund tool accepts order IDs but its schema describes an untyped free-form string, rather than the failure described here: Claude often confuses a read-only balance lookup with a credit-issuance action.

B:

This choice is relevant when a tool returns a large raw object although callers need just three fields; a narrow result format saves context and keeps downstream reasoning focused on validated data rather than incidental service metadata. Here, however, Claude often confuses a read-only balance lookup with a credit-issuance action.

C:

It would be appropriate when an account lookup can handle only active accounts, which is absent from its description; a precondition affects whether the model should choose the tool; hiding it leads to avoidable failures and misleading answers. Here, Claude often confuses a read-only balance lookup with a credit-issuance action, so a different mechanism is required.

Question 35

A document-processing architecture decides between online calls and batch jobs. A nightly backlog consists of independent invoices due by the next morning. Which response is most reliable?

  1. Compute field-level error metrics by input cohort and workflow version
  2. Invoke the model synchronously with bounded timeout and fallback
  3. Use batch requests with durable tracking and completion monitoring
  4. Mark uncertain items for separate human adjudication while preserving successful outputs
  5. Maintain stable item IDs and a durable submission ledger

Explanation

Correct Answer

C:

The deadline tolerates deferred execution and requests are independent, making batch orchestration a suitable operational pattern.

Incorrect Answers

A:

This choice is relevant when the team wants to know whether extraction quality changes over several nightly runs; aggregate success status does not prove accuracy; scored examples reveal changes in extraction behavior over time. Here, however, a nightly backlog consists of independent invoices due by the next morning.

B:

A transaction depends on the output now; asynchronous batch scheduling would not satisfy the immediate completion dependency; it protects against the case where a cashier needs a tax calculation before finishing the same transaction, rather than the failure described here: a nightly backlog consists of independent invoices due by the next morning.

D:

Partial automation can still be useful, but the pipeline must not pretend every item met the same confidence threshold; the alternative problem occurs when a large job contains some documents that require human review. The present scenario instead requires the system to schedule a noninteractive bulk workload.

E:

It would be appropriate when a batch worker may retry failed items after network interruption; a replay-safe manifest distinguishes outstanding items from completed or already submitted work and prevents accidental duplicate processing. Here, a nightly backlog consists of independent invoices due by the next morning, so a different mechanism is required.

Question 36

An organization wants scoped Claude Code instructions without huge root prompts. Database migration files need extra review beyond normal code. What should the team do to resolve this particular gap?

  1. Define a rule for generated paths that points to the generator command
  2. Create a dedicated rule for migration directories and file patterns
  3. Target only the legacy module rather than the repository root
  4. Enforce secret scanning in CI for the affected files

Explanation

Correct Answer

B:

Targeted rules add critical checks when touching migrations without burdening unrelated UI tasks. The relevant requirement is to attach migration-specific controls to relevant paths, and this measure operates at that decision point.

Incorrect Answers

A:

When a generated file is matched, Claude should change the generator input rather than apply a patch that will be overwritten; it protects against the case where generated source files should not be edited by hand, rather than the failure described here: database migration files need extra review beyond normal code.

C:

A narrow rule preserves compatibility requirements without imposing obsolete practices on active packages; the alternative problem occurs when a rule is needed only for a deprecated module kept for compatibility. The present scenario instead requires the system to attach migration-specific controls to relevant paths.

D:

This choice is relevant when developers accidentally commit secrets in test-fixture paths; path-specific instructions may remind Claude, but only a scanner or policy gate reliably blocks forbidden content before merge. Here, however, database migration files need extra review beyond normal code.

Question 37

An engineering company integrates Claude Code into a secure CI workflow. The job must comment on a PR but may not modify the branch. Which control is most directly relevant?

  1. Use schema-checked JSON findings with severity and source locations
  2. Gate success on actual test and lint exit statuses, not the assistant's assertion
  3. Grant only the API permission required for review comments and withhold repository-write capability
  4. Use separate restricted runners or permission scopes for untrusted PR analysis

Explanation

Correct Answer

C:

Least privilege lets CI publish feedback without creating commits or changing protected source files. This matches the workflow goal to use narrowly scoped write authority rather than introducing an unrelated action.

Incorrect Answers

A:

A validated contract lets downstream code render and deduplicate findings without trusting prose extraction; it protects against the case where the pipeline requires a stable format for severity, path, and line number, rather than the failure described here: the job must comment on a PR but may not modify the branch.

B:

A generated summary is evidence to inspect, not a replacement for deterministic pipeline results; the alternative problem occurs when a build reports a Claude-generated pass despite skipped unit tests. The present scenario instead requires the system to use narrowly scoped write authority.

D:

This choice is relevant when the reviewer has access to production deployment credentials through the runner image; a model that reads untrusted code should not share a security boundary with credentials capable of changing production. Here, however, the job must comment on a PR but may not modify the branch.

Question 38

An internal compliance agent handles ambiguous requests. A new regulation is mentioned but the maintained policy library has not been updated. How should the workflow be adjusted?

  1. Record the verified conditions and perform the permitted action
  2. Seek the necessary proof before making the decision
  3. Record both versions and request adjudication from the policy owner
  4. Request compliance-owner guidance instead of inferring a rule

Explanation

Correct Answer

D:

Regulatory ambiguity cannot be repaired by confident paraphrase; the accountable team must supply the applicable policy.

Incorrect Answers

A:

This choice is relevant when a reviewer confirms that all documented conditions are satisfied; reliable systems should distinguish genuine blockers from resolved ones and avoid unnecessary repeated escalation. Here, however, a new regulation is mentioned but the maintained policy library has not been updated.

B:

Eligibility text is not enough when required supporting facts are absent; clarification can close the gap without inventing evidence; the alternative problem occurs when a document explicitly allows an action, but the request lacks required evidence. The present scenario instead requires the system to flag the missing authoritative interpretation.

C:

Silent selection of whichever date looks newest can be wrong when different scopes or transition rules apply; it protects against the case where two approved policy documents contain contradictory effective dates, rather than the failure described here: a new regulation is mentioned but the maintained policy library has not been updated.

Question 39

A customer-service assistant may hand tasks to billing, returns, and policy specialists. Each specialist should see only the customer details it needs. Which implementation decision is the best fit?

  1. Check the current case state and ownership before starting an escalated job
  2. Assign one authorized mutation owner and separate advisory from execution roles
  3. Make identity verification an explicit prerequisite in the coordinator workflow
  4. Pass narrowly scoped task context and redact irrelevant customer information

Explanation

Correct Answer

D:

A smaller context packet reduces leakage and helps a specialist focus on the authorized decision instead of broad account history.

Incorrect Answers

A:

This choice is relevant when an escalation specialist is assigned a case that has already been resolved; state validation at handoff keeps asynchronous specialists from acting on superseded requests or producing redundant outcomes. Here, however, each specialist should see only the customer details it needs.

B:

A clear action boundary prevents double credits; two autonomous agents should not both be able to commit the same transaction; it protects against the case where billing and returns specialists could both initiate credits on the same order, rather than the failure described here: each specialist should see only the customer details it needs.

C:

A worker should not process a customer-specific return against an unverified identity; graph ordering enforces the prerequisite; the alternative problem occurs when a returns worker needs a decision made by the identity-verification worker. The present scenario instead requires the system to minimize private data exposed to subagents.

Question 40

A document classifier uses Claude to tag contracts with legal categories. A particular clause belongs to two related categories but reviewers expect one primary label. What should the architect specify in the design?

  1. Replace obsolete examples and rerun evaluation on current cases
  2. Include labeled examples that explain how to break ties
  3. Show benign text that resembles the target but does not satisfy its criteria
  4. Choose examples from different layouts and writing styles while keeping the rubric stable
  5. Select diverse boundary cases based on observed error patterns

Explanation

Correct Answer

B:

Examples should show which category wins at overlapping boundaries rather than leaving Claude to infer an unstated hierarchy.

Incorrect Answers

A:

Few-shot content can encode policy mistakes as strongly as explicit instructions; stale exemplars should be treated as configuration debt; the alternative problem occurs when a prompt example contains outdated policy thresholds. The present scenario instead requires the system to demonstrate category precedence.

C:

It would be appropriate when reviewers identify recurring false positives from harmless boilerplate; negative cases calibrate the decision boundary more directly than adding more obvious positive examples. Here, a particular clause belongs to two related categories but reviewers expect one primary label, so a different mechanism is required.

D:

Varied examples prevent the model from associating a label with incidental formatting instead of the clause meaning; it protects against the case where the training examples use only one document style, but production inputs vary, rather than the failure described here: a particular clause belongs to two related categories but reviewers expect one primary label.

E:

This choice is relevant when the team has hundreds of redundant examples in a single prompt; a smaller curated set often gives clearer task guidance and avoids burying the decision rule under duplicated demonstrations. Here, however, a particular clause belongs to two related categories but reviewers expect one primary label.

Claude Certified Architect - Foundations (CCAR-F) (CCA-F) Practice Tests