Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 21

Question 200

A server supplies a reusable user-invoked workflow template through its prompt capability. What should a client avoid assuming?

  1. That a template can help structure repeated tasks
  2. That template inputs may need validation
  3. That the template grants permission to execute protected tools
  4. That the user can review a rendered template

Explanation

Correct Answer

C:

A prompt can guide a workflow but does not confer access rights. Actual tool operations still require the application's and server's authorization checks.

Incorrect Answers

A:

Reusable instructions can be useful for consistency. The limitation concerns authority, not the value of a well-designed template.

B:

Inputs can influence rendered instructions and should be handled according to the contract. Validation remains relevant for reusable prompts.

D:

Review can be appropriate before use. The template's existence does not prevent a client from making its content visible and understandable.

Question 201

A backend needs one fixed operation from an existing well-documented REST API. It has no need for dynamic tool discovery across hosts. What is a reasonable baseline?

  1. Build an MCP layer because every model integration requires it
  2. Ask the model to invent an undocumented endpoint
  3. Expose every internal endpoint to avoid future adapter work
  4. A direct API adapter with a validated narrow contract
  5. Replace the API with browser automation immediately

Explanation

Correct Answer

D:

Protocol choice should serve the integration requirements. A direct adapter may be simpler when discovery and cross-host capability exposure add no demonstrated value.

Incorrect Answers

A:

MCP can be useful, but it is not mandatory for every external operation. Unneeded protocol layers add maintenance and failure surfaces.

B:

Generated guesses do not establish a working or supported API. Implementation must follow the actual service interface.

C:

Broad exposure increases complexity and authority beyond the current need. The integration should start with the required operation and explicit contract.

E:

A stable purpose-built interface already exists. Browser control may add fragility without solving a missing capability in this scenario.

Question 202

An agent runs a CLI whose human-readable output changes with locale and terminal width. The application parses fixed character positions. What should be preferred if supported?

  1. Removing all whitespace before parsing positions
  2. A documented machine-readable output format
  3. A model guess at the meaning of shifted columns
  4. A wider terminal to preserve today's layout

Explanation

Correct Answer

B:

Structured output provides stable fields independent of presentation. The integration should still validate versions, errors, and required data before using the result.

Incorrect Answers

A:

Whitespace removal changes the positions and can merge fields. It does not create explicit data boundaries or stable semantics.

C:

Interpretation can fail silently on unexpected output. A supported structured format is more reliable than guessing column semantics.

D:

Width can change again and locale still affects wording. Presentation settings are not a robust substitute for a machine contract.

Question 203

A CLI tool accepts a user-provided filename. The implementation concatenates it into a shell command. What is the safer integration design?

  1. Ask the model whether the filename looks harmless
  2. Pass validated arguments through an argument-vector interface
  3. Remove spaces but allow all other metacharacters
  4. Concatenate validated-looking values into a shell command string

Explanation

Correct Answer

B:

Separating executable and arguments avoids interpreting user text as shell syntax. Path and operation permissions still need validation for the intended task.

Incorrect Answers

A:

Appearance is not reliable validation. Execution should use deterministic argument handling and explicit access constraints.

C:

Space removal is not a complete defense against shell interpretation. A structured invocation avoids depending on an incomplete character blacklist.

D:

Visual plausibility does not prevent shell interpretation of metacharacters. An argument-vector boundary and appropriate validation avoid treating data as command syntax.

Question 204

An MCP configuration points to a newly supplied remote server URL. Before enabling it for sensitive work, what must be assessed?

  1. Only the number of tools the server advertises
  2. Only a successful tool-list response from the endpoint
  3. Server identity, permitted data flows, and exposed capabilities
  4. Only an encrypted connection, without verifying the intended server identity

Explanation

Correct Answer

C:

Connecting creates a new trust and data-processing relationship. The host must understand who receives information and what operations become available.

Incorrect Answers

A:

Tool count says little about consequence or trustworthiness. One powerful operation can matter more than many narrow read-only tools.

B:

Discovery establishes that an endpoint responds with capabilities, not that it is the intended trusted server. Identity and permitted data flows still need verification.

D:

Encryption to an unintended endpoint can still disclose data to the wrong party. Trust assessment must establish who controls the actual destination.

Question 205

One autonomous service delegates a research task to another autonomous service. What must their integration define beyond a list of callable functions?

  1. Task ownership, progress, result semantics, and cancellation behavior
  2. Only the format of the second agent's final answer
  3. Only a shared maximum answer length
  4. A guarantee that both agents use the same model

Explanation

Correct Answer

A:

Delegated work has a lifecycle and accountability boundary. The services need to agree how accepted tasks progress, fail, complete, and return evidence.

Incorrect Answers

B:

Final formatting does not define ownership, progress, or cancellation while work is underway. The caller needs a lifecycle contract for the delegated operation.

C:

Output length is a small formatting constraint. It does not cover acceptance, failure, progress, or ownership of the delegated operation.

D:

Model identity does not define coordination semantics. Different or identical models still need a reliable task protocol.

Question 206

A client assumes every server supports a particular optional protocol feature. Some connections fail during use. What should change?

  1. Discover supported capabilities and handle unsupported features explicitly
  2. Treat every server as the newest implementation
  3. Retry the unsupported operation indefinitely
  4. Hide feature errors and report successful execution

Explanation

Correct Answer

A:

Optional features require agreement between participants. The client should select compatible behavior rather than assume support from protocol membership alone.

Incorrect Answers

B:

Version assumptions do not establish deployed capability. Servers can support different revisions or optional feature subsets.

C:

An unsupported capability is not necessarily a transient outage. Repetition cannot add a missing feature to the server.

D:

Concealing failure makes downstream state unreliable. The application must represent unsupported behavior and choose a valid alternative or stop.

Question 207

A remote server changes its tool list while a client retains an old cached catalog. What should the client support?

  1. Assume the first catalog remains valid forever
  2. Treat removed tools as successful no-ops so existing plans can continue
  3. A supported refresh or change-notification mechanism
  4. Delete all saved user work whenever a tool changes

Explanation

Correct Answer

C:

Discovery results can become stale. Refreshing through the protocol's supported mechanism keeps selection aligned with the capabilities actually exposed.

Incorrect Answers

A:

A permanent assumption can cause calls to removed tools or miss updated contracts. Capability discovery needs a lifecycle policy.

B:

A successful no-op hides that the planned capability no longer exists. The client should refresh the catalog and explicitly handle unavailable operations rather than fabricate successful execution.

D:

Catalog refresh does not inherently require loss of unrelated task state. The application should update capability knowledge and assess affected operations.

Question 208

A service receives a webhook claiming that an analysis job completed. What should happen before its contents update trusted job state?

  1. Verify authenticity and validate the event against the job contract
  2. Ask Claude whether the message resembles previous callbacks
  3. Trust the event because it arrived at the correct URL
  4. Accept any payload containing a known job identifier

Explanation

Correct Answer

A:

External callbacks are untrusted until checked. Signature verification and event validation help establish that the update came from the expected source and applies to the job.

Incorrect Answers

B:

Similarity is not cryptographic verification. The integration should validate the service's actual authentication and event semantics.

C:

Knowing the endpoint does not prove sender identity. An attacker or mistaken system may still submit data to that address.

D:

A job ID is a correlation value, not necessarily a credential. Possessing it alone should not grant authority to change trusted status.

Question 209

A valid signed completion event is delivered several times. Its handler would otherwise issue the same reward repeatedly. What additional control is needed?

  1. Reject all signed events after the first day
  2. Change the reward amount on every retry
  3. Treat every valid signature as a new completion
  4. Track event identity and make handling idempotent

Explanation

Correct Answer

D:

Authenticity does not establish novelty. Deduplication or idempotent processing prevents a repeated legitimate event from causing repeated business effects.

Incorrect Answers

A:

A broad time cutoff can discard valid later events. Replay control should follow the event contract rather than an arbitrary shutdown rule.

B:

Varying the consequence does not prevent duplication. It makes repeated processing less predictable while preserving the underlying defect.

C:

The same signed message can be delivered again. Signature validity proves origin and integrity under the contract, not that processing has never occurred.

Question 210

A client receives different failures from external calls. Which cases generally warrant considering a bounded retry after checking the service contract? Choose TWO.

  1. An authorization denial for the current identity
  2. A transient service-unavailable response
  3. A validation error for a missing required field
  4. A rate-limit response with retry guidance
  5. A request for an unsupported operation version

Explanation

Correct Answers

B:

Temporary unavailability may resolve, so a bounded retry can be appropriate. The policy still needs deadlines and protection against repeating unsafe side effects.

D:

The service explicitly signals a capacity condition and often a delay. Respecting that guidance can support recovery without an immediate retry storm.

Incorrect Answers

A:

Repeated requests do not grant missing permission. The identity or approved access arrangement must be corrected before the operation can succeed.

C:

An unchanged malformed request is likely to fail again. The caller must correct the payload rather than retry it blindly.

E:

An unsupported contract is not repaired by waiting. The client must use a supported interface or handle the incompatibility explicitly.

Claude Certified Architect - Professional (CCAR-P) Practice Tests