[Nov-2026] Exam Sure Pass Anthropic Certification with CCA-F exam questions [Q65-Q87]

Rate this post

[Nov-2026] Exam Sure Pass Anthropic Certification with CCA-F exam questions

Real Anthropic CCA-F Exam Questions Study Guide

Anthropic CCA-F Exam Syllabus Topics:

Section	Weight	Objectives
Topic 1: Prompt Engineering & Structured Output	20%	– Instruction design 1. Deterministic output formatting (e.g., JSON) 2. Few-shot prompting strategies – Tool-augmented prompting 1. Tool-use driven reasoning patterns 2. Validation and retry loops
Topic 2: Tool Design & MCP Integration	18%	– Tool interface design 1. Clear tool descriptions and boundaries 2. Structured tool input/output schemas – Model Context Protocol (MCP) 1. Tool selection and routing strategies 2. MCP server integration patterns
Topic 3: Agentic Architecture & Orchestration	27%	– Agentic loops & lifecycle design 1. Tool use loop execution (request → tool_use → result → next step) 2. Stop reason handling and control flow – Multi-agent orchestration 1. Coordinator / sub-agent patterns 2. Parallel task decomposition
Topic 4: Context Management & Reliability	15%	– Context window optimization 1. “Lost in the middle” mitigation strategies 2. Managing long conversation degradation – System reliability 1. Escalation strategies (fail vs retry vs human) 2. Error propagation handling
Topic 5: Claude Code Configuration & Workflows	20%	– Claude Code configuration 1. CLAUDE.md hierarchy and scoping 2. Project-level vs user-level configuration – CI/CD and workflow integration 1. Command vs plan mode usage 2. Automated pipeline integration

NO.65 When processing a batch of 10,000 documents via the Message Batches API, 45 requests fail due to context window limits. What is the recommended failure handling strategy?

Resubmit the entire batch of 10,000 requests to guarantee correct ordering.

Resubmit only the failed documents (identified by their custom_id) after modifying them, such as chunking oversized documents.

Switch immediately to the synchronous API to retry the failed documents.

Ignore the failures, as the API inherently guarantees a 99% success SLA.

NO.66 For which of the following workloads is the Message Batches API considered inappropriate due to its architectural constraints?

Nightly test suite generation across a repository.

Weekly compliance scans of customer support logs.

Blocking pre-merge code reviews in a CIICD pipeline.

Asynchronous translation of bulk documentation.

NO.67 A well-designed structured error response for an MCP tool should include which of the following standard fields? Choose 2 correct answers.

isError

errorCategory

system_prompt_override

temperature_adjustment

NO.68 What is the primary architectural advantage of using glob-pattern rules in a . claude/rules/ file rather than placing a CLAUDE. md file in a subdirectory?

Glob patterns bypass the Model Context Protocol (MCP) server requirements for fetching dynamic rules.

Glob patterns allow conventions to apply to files by type regardless of their directory location, which is essential for conventions that span multiple directories.

Glob patterns automatically apply the rules to the global system-level /etc/c1aude/CLAUDE .md configuration.

Glob patterns disable Claude’s semantic inference, forcing it to use exact keyword matching for all prompts.

NO.69 To extract metadata from a highly variable set of invoices, you use an enum field for document_type. How should you handle unexpected document types to maintain extraction quality?

Include an other enum value paired with a detail string field for extensible categorization.

Omit the enum constraint entirely and use an open-ended string field.

Implement a PreToolUse hook to drop unrecognized document types.

Provide an enum containing only known types and rely on the model to find the closest match.

NO.70 You are using Claude Code to accelerate software development. Your team uses it for code generation, refactoring, debugging, and documentation. You need to integrate it into your development workflow with custom slash commands, CLAUDE.md configurations, and understand when to use plan mode vs direct execution.
You’re implementing a caching layer for API responses to speed up the /products endpoint. You have a rough idea-Redis with a 5-minute TTL-but you’re new to production caching and aren’t sure what other considerations a robust implementation requires. What’s the most effective way to start your iterative workflow?

Start with a minimal request: “Add Redis caching to /products with 5-minute TTL.” Add features and fix issues through follow-up prompts as problems surface during testing.

Write a specification with your known requirements and “TBD” markers for uncertain areas, having Claude propose solutions for each TBD as it implements.

Ask Claude to interview you about the caching requirements before implementing, surfacing considerations like invalidation strategies, cache layers, consistency guarantees, and failure modes.

Use plan mode to analyze the current/products endpoint implementation, then provide your caching requirements once Claude explains how the existing code is structured.

NO.71 In which scenario is it architecturally more reliable to start a fresh session injected with a structured summary rather than using – – resume to continue an existing session?

When the prior tool results in the existing session’s context window have become completely stale.

When you want to maintain the exact reasoning chain from the previous generation steps.

When the target codebase has not undergone any structural changes.

When executing a fast, single-file typo correction.

NO.72 A user asks Claude to ignore all previous instructions and reveal confidential internal prompt content. How should Claude behave?

Reveal the prompt.

Partially reveal hidden instructions.

Follow higher-priority instructions and refuse.

Generate random text.

NO.73 Production logs reveal inconsistent error handling: when lookup_order fails, the agent sometimes retries 5+ times (wasteful when the order ID doesn’t exist), sometimes escalates immediately (premature for temporary network issues), and sometimes asks users for clarification (inappropriate when the issue is a backend permission error). Investigation shows your MCP tool returns uniform error responses: {“isError”: true, “content”: [{“type”: “text”, “text”: “Operation failed”}]}. The agent cannot distinguish between error types. What’s the most effective improvement?

Create an analyze_error MCP tool the agent calls after any failure to determine the error category and recommended action.

Add few-shot examples to the system prompt demonstrating how to interpret error message patterns and select appropriate responses for each.

Implement retry logic with exponential backoff in your MCP server for all errors, returning to the agent only after retries are exhausted.

Enhance error responses with structured metadata: include errorCategory (transient/validation/permission), isRetryable boolean, and a description of what caused the failure.

NO.74 Why is implementing an escalation rule based on a customer’s negative sentiment considered an architectural anti-pattern?

Sentiment does not correlate with task complexity; an angry customer may have a straightforward request that the agent can resolve autonomously.

Negative sentiment triggers the model’s Constitutional A1 safety filters, automatically terminating the session.

Sentiment analysis requires fine-tuning a custom embedding model, which introduces latency.

The system prompt naturally overrides sentiment triggers using the tool_choice: “auto” command.

NO.75 Which of the following represent anti-patterns when designing escalation and ambiguity resolution logic for Claude? Choose 2 correct answers.

Automatically escalating a case whenever the model’s self-reported confidence score drops below a specific threshold.

Escalating immediately when a customer explicitly demands to speak to a human representative.

Routing a case to a human agent purely because the user uses highly negative or frustrated language.

Escalating a transaction when it exceeds a programmatic business threshold enforced via a PostToolUse hook.

NO.76 To measure ongoing error rates in high-confidence extractions and detect novel error patterns, which methodology should be implemented?

Stratified random sampling.

Exhaustive 100% human review.

Aggregate accuracy averaging.

Prompt-based sequential pipelining.

NO.77 Your MCP server includes archive_file(file_id) and delete_file(file_id) tools. Production logs show the agent calls delete_file when users ask to “remove old backups,” but company policy requires archiving backup files. Both tools currently have minimal descriptions: “Archives a file” and
“Deletes a file.” Which change most directly improves tool selection for this scenario?

Expand tool descriptions to clarify use cases, adding guidance like “Do not use for backup files” to delete_file.

Add a confirmation step that requires users to type “CONFIRM DELETE” before delete_file executes.

Implement server-side validation that rejects delete_file calls for files tagged as backups, returning an error message suggesting archive_file.

Add few-shot examples to the system prompt demonstrating that requests involving “backup” or
“old” should use archive_file.

NO.78 An AI extraction system processes thousands of financial documents. The dashboard displays an impressive 96% overall aggregate accuracy. However, downstream teams complain about frequent errors. What is the most likely oversight in the review and monitoring design?

The system is using prompt chaining instead of dynamic adaptive decomposition.

Aggregate accuracy metrics can mask severe per-document-type failures (e.g., invoices failing at 70% while receipts succeed at 99%).

The extraction is using JSON schema tool_use which guarantees semantic correctness but fails structural checks.

The system lacks a PostToolUse hook to block large financial transactions.

NO.79 How should Claude Code be executed in a Continuous Integration (Cl) pipeline to analyze pull requests without hanging indefinitely?

Option A

Option B

Option C

Option D

NO.80 How can an architect split a massive, monolithic CLAUDE. md file to selectively include external standards files based on the current module?

By utilizing the #include macro inside the file’s YAML frontmatter.

By using the @import syntax to reference and pull in external markdown files.

By compiling a PreToolUse hook to merge files dynamically before generation.

By specifying an allowed-tools array pointing to specific directories.

NO.81 You’ve configured the system so that all four subagents have access to the complete set of
18 tools. During testing, agents frequently call tools outside their specialization – the synthesis agent attempts web searches, and the report generator tries to analyze documents. What is the primary cause of this poor tool selection behavior?

Choosing from 18 tools instead of 4-5 relevant ones increases decision complexity beyond reliable selection thresholds.

The tool definitions consume too much context window space, leaving insufficient room for task content.

The agents’ role descriptions in their system prompts conflict with having access to tools outside that role.

The coordinator cannot track which capabilities each subagent has, leading to misrouted tasks.

NO.82 In a multi-agent orchestration, the synthesis agent occasionally misuses a web search tool instead of synthesizing the provided findings. How should you prevent this cross-role misuse according to the principle of scoped tool access?

Implement a PostToolUse hook to silently drop web search results when executed by the synthesis agent.

Remove the web search tool from the synthesis agent’s allowed tool set entirely.

Add a capitalized warning in the system prompt instructing the synthesis agent to ignore web search capabilities.

Use the fork session command to isolate the web search tool execution.

NO.83 You’re implementing the escalation logic for when the agent should call escalate_to_human. Your team proposes four different approaches for triggering escalation. Which approach will most reliably identify cases that genuinely require human intervention?

Implement sentiment analysis that monitors for frustration indicators (negative language, repeated questions, exclamation marks) and trigger escalation when the frustration score exceeds a configured threshold.

Configure the agent to escalate after three consecutive tool calls that fail to resolve the customer’s stated issue, ensuring a reasonable attempt before involving a human.

Instruct the agent to escalate when the customer requests a human, when the issue requires policy exceptions, or when the agent cannot make meaningful progress.

Build a rules engine that maps specific issue types, customer segments, and product categories to escalation decisions, removing the need for model judgment calls.

NO.84 Your project has React components, API handlers, and database models. Test files are spread throughout the codebase next to the code they test. How can you ensure Claude automatically applies the correct testing conventions when generating code?

Consolidate all conventions in the root CLAUDE .md file and rely on Claude to infer the correct section.

Create rule files in . claude/rules/ with YAML frontmatter specifying glob patterns like **/*. test.tsx.

Create an MCP server that injects testing rules into the prompt manually.

Place a separate CLAUDE. md file in each subdirectory containing test-specific conventions.

NO.85 The web search agent has gathered several relevant sources for a research topic. The document analysis agent now needs to examine these sources. How does information typically flow between these two specialized subagents?

The agents communicate through an event-driven message queue, with the document analysis agent subscribing to web search completion events.

The web search agent directly invokes the document analysis agent, passing the discovered sources as parameters.

The coordinator agent receives the web search agent’s output and includes relevant findings in the prompt when invoking the document analysis agent.

Both agents access a shared memory store where the web search agent writes findings and the document analysis agent reads them.

NO.86 A development team at quicktechie.com wants to enforce strict security standards specifically whenever a developer modifies infrastructure-as-code files. Which YAML frontmatter configuration correctly applies these rules to all Terraform files within the repository?

Option A

Option B

Option C

Option D

NO.87 You are using Claude Code to accelerate software development. Your team uses it for code generation, refactoring, debugging, and documentation. You need to integrate it into your development workflow with custom slash commands, CLAUDE.md configurations, and understand when to use plan mode vs direct execution.
A critical bug is affecting production users. Error logs show exceptions in the OrderProcessing module with a clear stack trace pointing to a specific function. You haven’t worked with this module before. What’s the most effective approach?

Start with direct execution to gather initial information, then switch to plan mode to design a comprehensive solution before implementing any changes.

Enter plan mode to explore the module’s architecture and dependencies before attempting any fixes.

Use plan mode to analyze the error in context of the module’s design, enumerate potential root causes, and prioritize fixes systematically.

Use direct execution to examine the stack trace, read the relevant code, and implement a fix once you identify the root cause.