CCA-F Exam Questions — 60 Free Practice Questions

Question 1 of 60

D1beginnerconcept

In an agentic loop with Claude, what stop_reason value indicates the model wants to execute a tool?

Question 2 of 60

D1intermediatescenario

What is the recommended approach for managing tool access in this multi-capability agent?

Question 3 of 60

D1advancedquickfire

Which enforcement mechanism CANNOT be bypassed by the model during an agentic workflow?

Question 4 of 60

D2beginnerconcept

When a tool execution fails, what should you include in the tool result to properly signal the error to Claude?

Question 5 of 60

D2intermediatescenario

Where should you configure the MCP servers for team-wide sharing?

Question 6 of 60

D2intermediatequickfire

In the Model Context Protocol, what is the key difference between tools and resources?

Question 7 of 60

D3beginnerconcept

What is the correct hierarchy for CLAUDE.md file loading in Claude Code?

Question 8 of 60

D3intermediatescenario

Which flag is essential for running Claude Code in a CI/CD environment?

Question 9 of 60

D3advancedquickfire

In Claude Code hooks configuration, what does an exit code of 2 from a PreToolUse hook cause?

Question 10 of 60

D4beginnerconcept

What is the most reliable method to get structured JSON output from Claude?

Question 11 of 60

D4intermediatescenario

What is the most cost-effective approach for this bulk processing task?

Question 12 of 60

D4intermediatequickfire

Why should optional fields in tool schemas be marked as NOT required rather than using default values?

Question 13 of 60

D5beginnerconcept

What is the 'lost in the middle' effect in the context of LLM prompts?

Question 14 of 60

D5intermediatescenario

What is the best approach to manage this tool result bloat?

Question 15 of 60

D5advancedquickfire

Why is progressive summarization considered a RISK rather than a pure benefit in long-running agentic conversations?

Question 16 of 60

D1beginnerconcept

What does stop_reason: 'end_turn' indicate in a Claude API response within an agentic loop?

Question 17 of 60

D1intermediateconcept

What is the primary advantage of using the Task tool to spawn subagents compared to making multiple sequential tool calls in the same agent?

Question 18 of 60

D1intermediatescenario

What context does the subagent receive when spawned via the Task tool?

Question 19 of 60

D1beginnerquickfire

Which tool in Claude Code is used to spawn an independent subagent to complete a subtask?

Question 20 of 60

D1advancedconcept

When should you use programmatic enforcement (e.g., hooks) vs prompt-based enforcement for a constraint in an agentic workflow?

Question 21 of 60

D1advancedscenario

What is the correct way to launch all three subagents in parallel rather than sequentially?

Question 22 of 60

D1intermediatequickfire

In a coordinator-subagent architecture, which component is responsible for decomposing the high-level goal into subtasks?

Question 23 of 60

D1advancedscenario

What is the most reliable implementation of this approval requirement?

Question 24 of 60

D1intermediateconcept

What is the purpose of fork_session in an agentic workflow?

Question 25 of 60

D1beginnerscenario

What should your loop do next?

Question 26 of 60

D1advancedquickfire

When building a multi-agent system, what happens if you pass a subagent's entire conversation history back to the coordinator after task completion?

Question 27 of 60

D1intermediateconcept

In a well-designed agentic system, when should a subagent escalate back to the coordinator rather than attempting to self-recover?

Question 28 of 60

D1advancedscenario

What is the most appropriate system response to this detection?

Question 29 of 60

D2beginnerconcept

According to MCP best practices, what is the PRIMARY mechanism Claude uses to decide which tool to call?

Question 30 of 60

D2intermediateconcept

The official exam guide gives an example of tool overload degrading tool-selection reliability. Which comparison does it use?

Question 31 of 60

D2intermediatescenario

How should the tool result be structured to communicate this outcome accurately?

Question 32 of 60

D2beginnerquickfire

What does setting tool_choice: 'any' do in a Claude API call?

Question 33 of 60

D2advancedscenario

What MCP feature would most reduce this exploratory overhead?

Question 34 of 60

D2advancedconcept

Why is it important to distinguish between 'access failure' and 'empty result' in a tool's return value?

Question 35 of 60

D2intermediatequickfire

In MCP tool schema design, what is the effect of providing detailed descriptions for each parameter in addition to the tool-level description?

Question 36 of 60

D2advancedscenario

What is the correct tool_choice configuration for this requirement?

Question 37 of 60

D3beginnerconcept

What is the purpose of the .claude/rules/ directory with glob patterns in a Claude Code project?

Question 38 of 60

D3intermediateconcept

Where are .claude/commands/ stored, and who has access to them?

Question 39 of 60

D3intermediatescenario

What is the best Claude Code feature to standardize this workflow?

Question 40 of 60

D3beginnerquickfire

In Claude Code, what does 'Plan mode' allow you to do before Claude executes any changes?

Question 41 of 60

D3advancedscenario

Which Claude Code feature provides full context isolation for a sub-agent task?

Question 42 of 60

D3intermediateconcept

What is the Skills system in Claude Code used for?

Question 43 of 60

D3advancedquickfire

Which Claude Code hook type runs BEFORE a tool call and can be used to block the tool from executing?

Question 44 of 60

D3beginnerscenario

What does the developer get automatically from the repository's .claude/ folder?

Question 45 of 60

D3advancedconcept

When using the -p (headless) flag for Claude Code in CI/CD, what must you also specify to parse Claude's output programmatically?

Question 46 of 60

D4beginnerconcept

Why are few-shot examples more effective than verbal instructions alone for enforcing a specific output format?

Question 47 of 60

D4intermediateconcept

Why can Claude not reliably self-review its own reasoning for logical errors?

Question 48 of 60

D4intermediatescenario

How should you design the tool schema to prevent hallucinated severity values?

Question 49 of 60

D4beginnerquickfire

What is one important limitation of the Batch API compared to synchronous API calls?

Question 50 of 60

D4advancedscenario

What is the fundamental weakness of this self-review approach?

Question 51 of 60

D4intermediateconcept

What is the key benefit of using explicit, measurable criteria in prompts instead of vague quality descriptors?

Question 52 of 60

D4advancedquickfire

When implementing a validation loop for Claude's output, what distinguishes an effective loop from one that adds cost without benefit?

Question 53 of 60

D4beginnerscenario

Which approach gives you the most reliable, parseable output?

Question 54 of 60

D4advancedconcept

You need both high output consistency AND domain-specific format enforcement in a single Claude interaction. What combination of techniques achieves this most effectively?

Question 55 of 60

D5beginnerconcept

Why is Claude's self-reported confidence level considered poorly calibrated?

Question 56 of 60

D5intermediatescenario

What should a well-designed agentic system do at this point?

Question 57 of 60

D5intermediatequickfire

When should an agentic system escalate to a human operator due to a 'policy gap'?

Question 58 of 60

D5advancedconcept

What is the recommended strategy for preserving critical information in a long-running agentic conversation that may undergo summarization?

Question 59 of 60

D5advancedscenario

What constitutes best practice for structured error propagation in this scenario?

Question 60 of 60

D5beginnerquickfire

Practice MCQs

In an agentic loop with Claude, what stop_reason value indicates the model wants to execute a tool?

What is the recommended approach for managing tool access in this multi-capability agent?

Which enforcement mechanism CANNOT be bypassed by the model during an agentic workflow?

When a tool execution fails, what should you include in the tool result to properly signal the error to Claude?

Where should you configure the MCP servers for team-wide sharing?

In the Model Context Protocol, what is the key difference between tools and resources?

What is the correct hierarchy for CLAUDE.md file loading in Claude Code?

Which flag is essential for running Claude Code in a CI/CD environment?

In Claude Code hooks configuration, what does an exit code of 2 from a PreToolUse hook cause?

What is the most reliable method to get structured JSON output from Claude?

What is the most cost-effective approach for this bulk processing task?

Why should optional fields in tool schemas be marked as NOT required rather than using default values?

What is the 'lost in the middle' effect in the context of LLM prompts?

What is the best approach to manage this tool result bloat?

Why is progressive summarization considered a RISK rather than a pure benefit in long-running agentic conversations?

What does stop_reason: 'end_turn' indicate in a Claude API response within an agentic loop?

What is the primary advantage of using the Task tool to spawn subagents compared to making multiple sequential tool calls in the same agent?

What context does the subagent receive when spawned via the Task tool?

Which tool in Claude Code is used to spawn an independent subagent to complete a subtask?

When should you use programmatic enforcement (e.g., hooks) vs prompt-based enforcement for a constraint in an agentic workflow?

What is the correct way to launch all three subagents in parallel rather than sequentially?

In a coordinator-subagent architecture, which component is responsible for decomposing the high-level goal into subtasks?

What is the most reliable implementation of this approval requirement?

What is the purpose of fork_session in an agentic workflow?

What should your loop do next?

When building a multi-agent system, what happens if you pass a subagent's entire conversation history back to the coordinator after task completion?

In a well-designed agentic system, when should a subagent escalate back to the coordinator rather than attempting to self-recover?

What is the most appropriate system response to this detection?

According to MCP best practices, what is the PRIMARY mechanism Claude uses to decide which tool to call?

The official exam guide gives an example of tool overload degrading tool-selection reliability. Which comparison does it use?

How should the tool result be structured to communicate this outcome accurately?

What does setting tool_choice: 'any' do in a Claude API call?

What MCP feature would most reduce this exploratory overhead?

Why is it important to distinguish between 'access failure' and 'empty result' in a tool's return value?

In MCP tool schema design, what is the effect of providing detailed descriptions for each parameter in addition to the tool-level description?

What is the correct tool_choice configuration for this requirement?

What is the purpose of the .claude/rules/ directory with glob patterns in a Claude Code project?

Where are .claude/commands/ stored, and who has access to them?

What is the best Claude Code feature to standardize this workflow?

In Claude Code, what does 'Plan mode' allow you to do before Claude executes any changes?

Which Claude Code feature provides full context isolation for a sub-agent task?

What is the Skills system in Claude Code used for?

Which Claude Code hook type runs BEFORE a tool call and can be used to block the tool from executing?

What does the developer get automatically from the repository's .claude/ folder?

When using the -p (headless) flag for Claude Code in CI/CD, what must you also specify to parse Claude's output programmatically?

Why are few-shot examples more effective than verbal instructions alone for enforcing a specific output format?

Why can Claude not reliably self-review its own reasoning for logical errors?

How should you design the tool schema to prevent hallucinated severity values?

What is one important limitation of the Batch API compared to synchronous API calls?

What is the fundamental weakness of this self-review approach?

What is the key benefit of using explicit, measurable criteria in prompts instead of vague quality descriptors?

When implementing a validation loop for Claude's output, what distinguishes an effective loop from one that adds cost without benefit?

Which approach gives you the most reliable, parseable output?

You need both high output consistency AND domain-specific format enforcement in a single Claude interaction. What combination of techniques achieves this most effectively?

Why is Claude's self-reported confidence level considered poorly calibrated?

What should a well-designed agentic system do at this point?

When should an agentic system escalate to a human operator due to a 'policy gap'?

What is the recommended strategy for preserving critical information in a long-running agentic conversation that may undergo summarization?

What constitutes best practice for structured error propagation in this scenario?

When designing prompts for long documents, where should the most critical instructions be placed relative to the document content?