Claude Certified Developer — Foundations
53 questions with instant scoring, a per-section breakdown, an explanation for every answer, and a list of exactly which lessons to revise for the ones you miss. Pass mark: 72%. Free to take — a free account unlocks the full answer review, gap analysis and certificate.
Sample Claude Certified Developer — Foundations practice questions
6 free sample questions from the practice bank, with the correct answer and the reasoning.
Question 1 · Agents and Workflows
A team is building an automated pipeline that processes incoming documents in a fixed sequence: classify, extract, validate, and route. Steps never vary. Which execution pattern is most appropriate?
- A.Single agent with dynamic tool selection each step
- B.Fully autonomous agent that decides its own approach
- C.Fixed workflow with predefined, deterministic steps Correct answer
- D.Agent swarm distributing steps across parallel instances
Why: When steps and their order are known in advance, fixed workflows are the right choice — more predictable, cheaper, faster, and easier to monitor than agents. Agents should be reserved for genuinely dynamic decision-making.
Question 2 · Applications and Integration
A production Claude application must handle up to 500 simultaneous API requests during business hours. Which architectural capability is most critical?
- A.Streaming enabled on all requests
- B.Prompt caching for all system prompts
- C.Extended thinking for complex requests
- D.Horizontal scaling across multiple API client instances Correct answer
Why: Handling 500 concurrent requests requires horizontal scaling — distributing load across multiple application instances. Streaming improves perceived latency, prompt caching reduces cost, and extended thinking adds capability — none address concurrency capacity.
Question 3 · Claude Code
A team of 10 developers share a repository and want project-specific Claude Code instructions (build commands, architecture notes, conventions) available to everyone without per-user configuration. Where should they put these?
- A.In each developer's ~/.claude/settings.json
- B.In a CLAUDE.md file at the project root, committed to the repository Correct answer
- C.In a local .env file on each developer's machine
- D.In a Slack message pinned to the team channel
Why: CLAUDE.md is read at every Claude Code session start and is committed to the repository, making it available to every developer cloning the repo. Per-user settings require each developer to configure manually. .env files are for secrets, not instructions.
Question 4 · Eval, Testing, and Debugging
A multi-step agent produces incorrect final outputs in about 15% of production runs, but inputs appear valid. What is the most effective first debugging step?
- A.Analyse distributed traces to identify at which step the output first diverges from expected behaviour Correct answer
- B.Increase the model tier from Sonnet to Opus for all steps
- C.Add more few-shot examples to the system prompt
- D.Reduce the number of tools available to the agent
Why: Tracing records every model turn, tool call, and result. Analysing traces narrows the problem from 'the agent fails' to 'step 3's tool result is malformed'. Upgrading the model or changing the prompt are changes to test, not debugging steps.
Question 5 · Model Selection and Optimization
A data scientist runs the same Claude prompt 50 times at temperature=0 and observes minor variations in three outputs. They conclude there is a bug. What is the correct interpretation?
- A.temperature=0 should produce perfectly identical outputs; this indicates a model defect
- B.The data scientist must use a seed parameter alongside temperature=0 for consistency
- C.Minor variation at temperature=0 is expected due to hardware floating-point differences Correct answer
- D.The prompt must be redesigned because temperature=0 is not functioning
Why: temperature=0 produces near-deterministic (not perfectly deterministic) outputs. Minor variation can occur due to hardware floating-point arithmetic differences across server instances. Claude does not expose a seed parameter. This is a documented characteristic, not an error.
Question 6 · Prompt and Context Engineering
A developer wants Claude to maintain a specific persona and never break character across every turn of a conversation. Where should this instruction be placed?
- A.At the end of each user message
- B.In the system prompt, which applies persistently to all turns Correct answer
- C.As a few-shot example at the start of the conversation
- D.In a PostToolUse hook that appends the instruction after each turn
Why: The system prompt is the operator-level instruction that persists for the entire conversation — the correct location for persistent rules, personas, and constraints. Per-turn prefixes add noise. Few-shot examples demonstrate style, not enforce persistent behaviour. PostToolUse hooks are for post-tool actions.
What this practice exam covers
53 questions in 8 sections.
- Agents and Workflows
- Applications and Integration
- Claude Code
- Eval, Testing, and Debugging
- Model Selection and Optimization
- Prompt and Context Engineering
- Security and Safety
- Tools and MCPs
More practice for this exam
Questions about this practice exam
- Are these the official Claude Certified Developer — Foundations exam questions?
- No. They are practice questions written against the published exam domains. This site is not affiliated with or endorsed by Anthropic.
- How many practice questions are there?
- 53 questions in 8 sections.
- Do I need an account to practise?
- No. You can take the practice questions without an account and your scores save in your browser. A free account adds the full answer review, gap analysis and a completion certificate.
- Do the questions explain the answers?
- Yes. Every question comes with an explanation of the correct answer.
Practice questions written against the published exam domains. Not official exam questions. Not affiliated with or endorsed by Anthropic.