Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 12

Question 110

A workflow has hooks that log errors but does not stop a failed authorization check. What change is required?

  1. Specify ordered responsibilities and nonconflicting hook contracts
  2. Return a blocking control result instead of an informational log event
  3. Execute a PostToolUse handler after successful file-edit tools
  4. Attach the validation to the agent-turn or stop lifecycle event
  5. Enforce a PreToolUse hook that denies matching unsafe invocations

Explanation

Correct Answer

B:

Emitting telemetry alone does not enforce permission decisions. The hook must use the runtime's supported deny mechanism at the correct stage.

Incorrect Answers

A:

Clear handler scope and explicit failure precedence prevent accidental interactions; this contributes to resolving conflicting lifecycle handlers without meeting the need for distinguishing logging from access denial.

C:

Because a post-tool callback can inspect the completed edit and trigger formatting or verification in response, the measure supports reacting to a completed editing tool rather than the needed distinguishing logging from access denial.

D:

End-of-turn hooks can trigger aggregate checks that would be wasteful after every individual edit; its benefits for validating at the end of agent activity do not resolve the need for distinguishing logging from access denial.

E:

PreToolUse runs before a tool call and can participate in the decision to block it; supporting blocking disallowed commands before tool execution cannot replace the required distinguishing logging from access denial.

Question 111

A delegated incident analyst should inspect only authorized customer records and must disclose which document versions support its conclusions. Which TWO controls fit this review? Choose TWO.

  1. Include inspected document identifiers, revisions and cited evidence in the delegated result
  2. Run the two workers in separate contexts and compare at the parent level
  3. Return only source titles without immutable IDs, versions, or inspected passages
  4. Forward the full case transcript to every specialist irrespective of authorization
  5. Limit the investigator's instructions and connectors to the assigned, authorized customer case
  6. Require each analyst to say 'completed' but omit source versions and record-access constraints

Explanation

Correct Answers

A:

This choice addresses validating the evidentiary basis of a subagent answer: source and version metadata distinguish inspected facts from memory-based guesses; the parent can then decide whether missing evidence warrants another tool call.

E:

Context minimization and access controls complement each other. A narrow prompt without restricted connectors still permits irrelevant private data retrieval.

Incorrect Answers

B:

Isolated contexts create genuinely independent assessments; the stated problem requires narrowing a worker's data-access boundary rather than obtaining independent subagent hypotheses.

C:

A title alone cannot establish which revision was used or whether the worker read the authorized record. The contract must require inspected evidence and document versions.

D:

Sending all records expands the sensitive-data exposure. The delegated analyst must receive only case-authorized material plus enough source lineage to substantiate the conclusion.

F:

A status field alone cannot prove a review used authorized records or traceable document versions. Both evidence provenance and scoped access are explicit requirements.

Question 112

A stakeholder wants to review a migration strategy before database files change. Which operating approach fits?

  1. Use the reviewed patch within its stated scope and verify the security regression
  2. Return to investigation and revise the plan around the newly discovered dependency
  3. Present the plan for approval before allowing write-capable execution
  4. Return the architectural trade-off and request the missing business decision

Explanation

Correct Answer

C:

Sensitive migrations benefit from separating design review from mutation. Plan-only work prevents an assistant from implementing unapproved data changes.

Incorrect Answers

A:

Replanning a settled narrow remediation can delay correction and introduce scope drift; that relates to avoiding needless replanning of an approved patch, not the necessary requiring plan approval for a migration.

B:

Although planning is iterative when evidence changes, the choice addresses replanning when new dependency evidence appears and not the requested requiring plan approval for a migration.

D:

An unresolved design choice is not safe to infer from code alone; the option would serve resolving architecture ambiguity before editing, not requiring plan approval for a migration as specified in the stem.

Question 113

A support workflow mixes verification and mutation in one large opaque tool. What improvement helps auditability?

  1. Stage exploration, plan approval and implementation as dependent tasks
  2. Separate the issue into bounded states with explicit exit and escalation criteria
  3. Separate read-only verification from the authorized state-changing step
  4. Establish measurable deliverables and pass-fail conditions for each subtask
  5. Build parallel prerequisite branches with a join gate before completion

Explanation

Correct Answer

C:

Distinct stages expose eligibility decisions and permit human approval before mutations. A monolithic operation hides policy checks and is harder to recover safely.

Incorrect Answers

A:

Staging the work lets decisions use verified repository context; in this scenario, decomposing dependent coding stages is secondary to separating validation from side-effecting tasks.

B:

When considering defining bounded task exits, bounded tasks define what evidence and action terminate work; the requirement for separating validation from side-effecting tasks remains unmet.

D:

A task boundary is useful when the parent can decide whether its output satisfies requirements; specifying task-level acceptance criteria and separating validation from side-effecting tasks require different controls in this situation.

E:

A fan-out and join pattern preserves concurrency while enforcing that no required verification branch is skipped; the unsatisfied requirement is separating validation from side-effecting tasks, not joining mandatory parallel checks.

Question 114

A customer service tool cannot access a tenant due to insufficient permissions. What should the MCP result convey?

  1. Observe bounded retry-after guidance in the retry scheduler
  2. Differentiate a legitimate not-found result from a tool execution error
  3. Remove secrets in public tool errors and retain details in protected logs
  4. Return an explicit authorization failure rather than a misleading not-found response

Explanation

Correct Answer

D:

Permission denials differ from nonexistent resources. Clear forbidden status enables escalation and avoids having the agent guess a missing customer record.

Incorrect Answers

A:

Rate-limit responses can indicate when another attempt is reasonable; honoring rate-limit backoff information and distinguishing access denial from missing data are distinct responsibilities under these constraints.

B:

A successful search with zero matches is not necessarily an infrastructure failure; the action supports separating no-match outcomes from failures but falls short of distinguishing access denial from missing data in this case.

C:

Agent-visible diagnostics should be actionable without exposing credentials; this can serve sanitizing error information returned to models, but leaves the requirement for distinguishing access denial from missing data unmet.

Question 115

Several workers report success but use different field names for the same order identifier. What delegation rule prevents integration failures?

  1. Mandate structured task results with IDs, status and completion evidence
  2. Hand off with the precise file paths or retrieved evidence required for the task
  3. Define one shared input and output field contract before spawning workers
  4. Add an explicit output contract and completion criteria in the delegated prompt
  5. Provide stable source identifiers alongside evidence extracts and claims

Explanation

Correct Answer

C:

Consistent schemas make aggregation mechanical and reduce misbound identifiers. Letting every agent invent identifiers defeats otherwise correct specialization.

Incorrect Answers

A:

Because stable identifiers and machine-readable outcomes let the parent distinguish completed assignments from new work, the measure supports normalizing delegated task results rather than the needed enforcing a shared worker data contract.

B:

Subagents have their own context and should be given task-critical inputs explicitly; its benefits for passing task-specific context into a subagent do not resolve the need for enforcing a shared worker data contract.

D:

A subagent needs a bounded objective, format and acceptance test; this contributes to making subagent completion testable without meeting the need for enforcing a shared worker data contract.

E:

Citation-safe merge requires referential metadata; that serves preserving source pointers through delegation, whereas the case requires enforcing a shared worker data contract.

Question 116

An analysis worker needs only search_documents, but the global registry exposes shell and file-write tools. What should configuration enforce?

  1. Resolve each tool's selection criteria and record type in descriptions
  2. Exclude the email-sending tool while keeping read and draft capabilities
  3. Include a pre-execution authorization and query-validation layer at the tool dispatcher
  4. Restrict publish authority on validated workflow state instead of mere tool availability
  5. Apply a scoped tool allowlist for the worker invocation

Explanation

Correct Answer

E:

A narrow allowance reduces both accidental mutation and prompt-injection opportunities. Registering every capability globally undermines useful worker separation.

Incorrect Answers

A:

Tool routing quality depends on distinct metadata; a method suited to disambiguating adjacent retrieval tools does not meet the need for applying per-worker tool allowlists.

B:

For separating drafting from outbound delivery permissions, tool distribution should distinguish generating a proposed communication from transmitting it; however, this requirement concerns applying per-worker tool allowlists.

C:

Model-generated arguments are proposals; that may serve validating privileged tool calls before execution, but it is no substitute for applying per-worker tool allowlists.

D:

Tool existence is not business authorization for every stage; this may support making a sensitive tool contingent on workflow state but cannot alone meet the requirement for applying per-worker tool allowlists.

Question 117

A refund flow must confirm eligibility before running the payment reversal. What should implement the ordering?

  1. Direct to identity clarification instead of advancing to account-specific tools
  2. Build in an executable policy gate before the final approval or refund call
  3. Direct the case to an explicit policy-resolution or human-review state
  4. Make eligibility verification a required predecessor in the workflow graph

Explanation

Correct Answer

D:

An explicit dependency ensures the mutation cannot begin before policy checks succeed. A verbal instruction alone is less reliable than a controller gate.

Incorrect Answers

A:

Identity-dependent steps require verified identifiers; the deciding condition is enforcing refund prerequisites before mutation, whereas this action concerns blocking state transitions on missing identity.

B:

A deterministic guard can compare amounts with configured limits; the option would serve enforcing financial limits in a workflow, not enforcing refund prerequisites before mutation as specified in the stem.

C:

Ambiguity requires a governed branch instead of a silent best guess; this addresses modeling ambiguity as an escalation branch, while the question is about enforcing refund prerequisites before mutation.

Question 118

A document extraction fails because one field is not parseable, although the raw text contains it. What retry strategy is reasonable?

  1. Verify extracted items against their source spans or document evidence
  2. Retry with targeted feedback identifying the invalid field and expected format
  3. Label failure types and retry only plausibly transient conditions
  4. Hold back or mark the amount uncertain pending human verification

Explanation

Correct Answer

B:

Feedback-based repair gives the model a specific correctable defect. Repeating the same unchanged prompt often reproduces the same structural mistake.

Incorrect Answers

A:

When addressing matching extracted entities to source evidence, per-field provenance validation detects additions not supported by the original content; the present task instead calls for targeting a retry at a correctable parse defect.

C:

Authentication and permission errors require configuration or credential repair; the unsatisfied requirement is targeting a retry at a correctable parse defect, not distinguishing retryable and nonretryable failures.

D:

High-impact numeric ambiguity deserves explicit uncertainty; in this scenario, calibrating abstention for unreadable amounts is secondary to targeting a retry at a correctable parse defect.

Question 119

Two tools named update_record and modify_record are being confused when a support agent handles refunds. What interface change reduces misrouting?

  1. Explain the target resource, permitted query scope and result semantics
  2. Give each tool a distinct action-focused name and usage description
  3. Release a stable output schema and version incompatible changes
  4. Record explicit success, not-found and error result shapes

Explanation

Correct Answer

B:

Discriminating names and descriptions specify the business intent and side effects of each operation. Semantic overlap in tool definitions makes the model's selection less predictable.

Incorrect Answers

A:

Effective tool descriptions explain when the operation should be used and what it returns; this can serve exposing the resource scope of retrieval tools, but leaves the requirement for differentiating similar tool responsibilities unmet.

C:

A predictable contract lets agents and validators consume results consistently; this concerns managing output schema compatibility, leaving the requirement for differentiating similar tool responsibilities outstanding.

D:

A clear output contract lets downstream agents tell absent records from service faults; specifying the semantics of empty tool results and differentiating similar tool responsibilities are distinct responsibilities under these constraints.

Question 120

A monorepo's shared instructions are bloated, individual packages need different conventions, and a framework upgrade made some rules obsolete. Which THREE maintenance actions correct these issues? Choose THREE.

  1. Move package-only conventions into instructions scoped to those package directories
  2. Revise the outdated framework rules and test them after the upgrade
  3. Trim the shared CLAUDE.md and link out to details needed only for specific tasks
  4. Settle the conflicting scope and establish one authoritative test command

Explanation

Correct Answers

A:

Selecting package-specific guidance matters in this scenario: directory-local guidance gives specialized expectations without imposing them on unrelated packages; the file's location communicates the intended subtree.

B:

This choice addresses maintaining instruction accuracy after upgrades: standing guidance can become stale as dependencies and conventions evolve; keeping it in the code-review process prevents outdated patterns from propagating.

C:

High-signal standing instructions are easier to apply and less costly in context. Bulk background should be retrieved as needed rather than always appended to coding sessions.

Incorrect Answers

D:

Because overlapping instructions can create inconsistent tool behavior, the measure supports eliminating conflicting project instructions rather than the needed keeping persistent instructions focused.

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