CCARF Practice Paper
Paper 4
Additional practice set · 37 questions. Use flashcard mode for instant feedback or exam mode to simulate the real test.
Start Paper 4 in the app
Questions in Paper 4
Q1D5 Context Management & Reliability — A multi-agent system has subagents for web search, academic database, and industry reports. The academic database subagent hits a 403 Forbid…
Q2D2 Tool Design & MCP Integration — A synthesis agent frequently returns control to the coordinator for simple fact verification, adding 2–3 round trips per task and 40% latenc…
Q3D4 Prompt Engineering & Structured Output — The moderation system classifies hate speech accurately for explicit slurs but misses coded language and dog-whistle terms that human modera…
Q4D5 Context Management & Reliability — A customer support agent handles a multi-issue session. After several turns, the agent refers to 'your recent refund request' instead of the…
Q5D5 Context Management & Reliability — A research subagent queries an academic journal database for articles on renewable energy policy. The database responds successfully but ret…
Q6D1 Agentic Architecture & Orchestration — The moderation system handles appeals where users contest a moderation decision. Currently, the same agent that made the original decision r…
Q7D4 Prompt Engineering & Structured Output — A moderation team introduces a severity scale ('low', 'medium', 'high', 'critical'). The same post (a user threatening to 'destroy' a compet…
Q8D5 Context Management & Reliability — A customer says: 'This is ridiculous, I have been waiting 20 minutes. Just connect me to a real person.' The agent has access to the custome…
Q9D2 Tool Design & MCP Integration — A developer is investigating a production bug. They know the error message 'InvalidStateTransition' is logged somewhere in the codebase, but…
Q10D5 Context Management & Reliability — A multi-agent research system has three subagents (financial filings, news, technical white papers) and a synthesis agent. Each subagent ret…
Q11D3 Claude Code Configuration & Workflows — A fintech company requires that all API response payloads are normalised to snake_case before being logged, and that any file write to the s…
Q12D4 Prompt Engineering & Structured Output — The moderation system uses tool_use with a JSON schema for classification output. All fields including 'sub_category' and 'target_demographi…
Q13D1 Agentic Architecture & Orchestration — A multi-agent research system must process a customer's request that involves three sequential stages: data collection, analysis, and report…
Q14D2 Tool Design & MCP Integration — Production logs show an agent frequently calls get_customer when users ask about orders (e.g. 'check my order #12345'), instead of calling l…
Q15D1 Agentic Architecture & Orchestration — In a hub-and-spoke research system, a coordinator manages several search and analysis subagents. Which responsibilities sit with the coordin…
Q16D4 Prompt Engineering & Structured Output — Your CI/CD pipeline processes code reviews using the Message Batches API. After a batch completes, you need to match each review result back…
Q17D4 Prompt Engineering & Structured Output — The moderation system validates that every classification includes a 'reasoning' field explaining the decision. When validation fails (empty…
Q18D3 Claude Code Configuration & Workflows — Project-level `.claude/CLAUDE.md` says 'use 4-space indentation matching the existing codebase.' A senior architect has 'use 2-space indenta…
Q19D5 Context Management & Reliability — A customer support agent searches for a customer named 'Sarah Johnson' and the lookup tool returns three matching accounts with the same nam…
Q20D2 Tool Design & MCP Integration — A research subagent's database query fails partway through an investigation. Which pieces of information belong in the structured error cont…
Q21D2 Tool Design & MCP Integration — A tool returns an empty array after a customer lookup. The agent retries 3 times, then escalates to a human agent. Analysis shows the custom…
Q22D4 Prompt Engineering & Structured Output — The moderation system's classification schema has a 'category' field defined as a free-text string. Auditors find 47 different category valu…
Q23D1 Agentic Architecture & Orchestration — A consulting firm needs an agent to add a comprehensive test suite to a large legacy codebase. The codebase has no existing tests, unclear d…
Q24D4 Prompt Engineering & Structured Output — A moderation classification tool includes a `detected_patterns` array where the model lists specific patterns it identified (e.g. 'repeated …
Q25D2 Tool Design & MCP Integration — A team's shared `.mcp.json` configures a PostgreSQL MCP server. A developer adds a personal staging database MCP server to their `~/.claude.…
Q26D5 Context Management & Reliability — A multi-agent research system finds conflicting data on a pharmaceutical compound's efficacy. A 2022 peer-reviewed clinical trial reports 78…
Q27D4 Prompt Engineering & Structured Output — Your CI/CD pipeline uses a system prompt with a tool named 'check_security' and also includes instructions that say 'check the security of e…
Q28D3 Claude Code Configuration & Workflows — A team has a /deploy-check skill that verifies deployment readiness. It needs to: (1) read configuration files across the repo, (2) run bash…
Q29D1 Agentic Architecture & Orchestration — An agent occasionally processes international transfers without required compliance checks. The compliance team requires 100% enforcement of…
Q30D1 Agentic Architecture & Orchestration — The moderation system auto-removes posts classified as policy violations. Production logs reveal that 3% of auto-removed posts were legitima…
Q31D4 Prompt Engineering & Structured Output — The moderation system's prompt instructs Claude to 'be conservative when moderating and err on the side of caution.' Reviewers find that inn…
Q32D4 Prompt Engineering & Structured Output — The moderation team asks a single Claude session to classify a post, then immediately asks the same session to independently review its own …
Q33D5 Context Management & Reliability — A customer support agent uses a token budget of 200k tokens. The system prompt consumes 8k tokens, conversation history takes 120k tokens, a…
Q34D1 Agentic Architecture & Orchestration — A developer resumes a Claude Code session after modifying 3 files in a 50-file codebase. The agent gives contradictory advice about the modi…
Q35D1 Agentic Architecture & Orchestration — The content moderation system uses a hub-and-spoke architecture with a coordinator that routes posts to specialist subagents: a text classif…
Q36D3 Claude Code Configuration & Workflows — A CI/CD pipeline runs three Claude Code steps in sequence: (1) generate a changelog from git commits, (2) review the changelog for accuracy,…
Q37D5 Context Management & Reliability — A web search subagent in a multi-agent research system times out while researching a complex topic. You need to design how this failure info…