Anthropic CCAR-F: Claude Certified Architect - Foundations

Last checked November 10, 2026 against the official Anthropic exam page.

1. Question 1Agentic Architecture & Orchestration

   A developer builds a loop around the Messages API. A response arrives with stop_reason set to 'tool_use'. What should the application do next?

   1. ARun the requested tool, then send a tool_result block that references the tool_use id in the next user message
   2. BShow the text blocks to the user and close the session, because the model has finished its turn
   3. CRestart the conversation from the system prompt and place the raw tool output inside the system field so the model reads it first
   4. DRemove the tool_use block from the history and resend the same request with a higher temperature to get a different response
   Show answer and explanationHide answer

   Correct answer: A

   A stop_reason of 'tool_use' means the model is waiting for a tool result, so the application runs the tool and returns a tool_result tied to the tool_use id, then calls the model again. Closing the session ends the loop too early, and restarting with tool output in the system field or changing temperature does not complete the exchange correctly.

2. Question 2Claude Code Configuration & Workflows

   A monorepo has a frontend folder and a database folder that follow different coding conventions. The team wants each set of conventions applied only when Claude is editing files in the matching folder. Which mechanism fits best?

   1. AOne long root CLAUDE.md that lists every convention for every folder, loaded in full on every single request
   2. BA separate custom slash command for each folder that developers must run before touching any file in that folder
   3. CShell environment variables exported on each developer's machine that switch conventions whenever they change into a folder
   4. DRule files under .claude/rules, each scoped with a path pattern so that it loads only for files in the matching folder
   Show answer and explanationHide answer

   Correct answer: D

   Path-scoped rule files load conventions only when Claude works in the matching area, which keeps context focused. A single large CLAUDE.md loads everything every time, slash commands depend on people remembering to run them, and shell variables on a developer's machine are not read as convention rules.

3. Question 3Prompt Engineering & Structured Output

   A code review prompt reports many style nitpicks as bugs, which buries the real defects in noise. Which prompt change most directly reduces these false positives?

   1. ATell the model to be thorough and to report every possible concern it can find anywhere in the diff, including personal style preferences
   2. BRaise the output token limit so the model has room to explain each finding at greater length in the review it produces
   3. CDefine explicit criteria for a reportable bug, including severity levels and examples of non-issues the model should skip
   4. DAsk the model to rate its confidence from one to one hundred on each finding, then post all findings regardless of the score
   Show answer and explanationHide answer

   Correct answer: C

   Explicit, testable criteria with severity definitions and named non-issues tell the model what to exclude, which cuts false positives. Asking for thoroughness, longer output, or confidence scores that are still all posted does not narrow what counts as a bug.

4. Question 4Tool Design & MCP Integration

   Two tools, lookup_customer and search_customers, both accept a free-text name. The agent keeps choosing search_customers even when it already has an exact customer ID. What is the best fix?

   1. AAdd the word 'important' to the description of lookup_customer so the model prioritizes it over the other tool
   2. BDelete search_customers from the agent's toolset entirely, even though some workflows still need partial-name lookups for customers
   3. CRewrite both descriptions to state what each tool returns, which input it expects, and when it should not be used
   4. DRaise the max_tokens limit so the model has more room to reason about which of the two tools is the correct choice
   Show answer and explanationHide answer

   Correct answer: C

   Clear descriptions that state what each tool returns, the input it expects, and when to avoid it give the model the signals it needs to route correctly. Emphasis words and a higher token limit do not resolve the ambiguity, and removing a tool that some workflows still need breaks legitimate use.

5. Question 5Context Management & Reliability

   During a long support chat, the customer's order number and the refund amount agreed earlier begin to drift in the model's replies. Which mitigation is most effective?

   1. AAsk the model to restate the figures in its own conversational words, so that replies sound more natural to the customer
   2. BShorten the system prompt to the bare minimum so that more of the context window remains free for the conversation history
   3. CKeep critical facts such as the order number and agreed amount in a structured block that is supplied again on every turn
   4. DTurn off the order lookup tool entirely, so the model stops receiving tool results that add noise and bulk to the context window over time
   Show answer and explanationHide answer

   Correct answer: C

   Pinning critical facts in a structured block that is re-supplied each turn keeps them exact instead of relying on paraphrase or recall across a long conversation. Asking for restated figures, trimming the system prompt, or removing the lookup tool does not preserve the facts reliably.