Anthropic CCDV-F

Anthropic CCDV-F 試験シラバストピック:

Anthropic Claude Certified Developer-Foundations 認定 CCDV-F 試験問題：

問題 #1
A teammate is reviewing the team's threat model for a Claude application and has asked you to identify the categories of AI-specific threats that the model should cover. The teammate has already listed traditional web application threats and wants to know what additional categories apply to a Claude application.
Which AI-specific threat categories would you add?

A. Supply chain attacks on the Claude SDK because the SDK itself is the only point of vulnerability that a Claude application introduces beyond traditional web application threats.
B. Cross-site scripting and SQL injection, because these traditional web application threats apply with equal weight to any application that uses Claude in any way.
C. Prompt injection, data leakage from prompts or context, jailbreak attempts, and unsafe model output that bypasses application controls.
D. Network-level denial of service and physical infrastructure attacks, because these categories cover the threats most likely to affect any Claude application in production.

問題 #2
The team is debating whether to integrate with the Claude API directly or through a third-party abstraction layer that supports multiple LLM providers. The team has identified that all current and projected use cases run on Claude, no internal customer has requested LLM portability, and the team's product roadmap does not mention multi-provider support over the next two years. The third-party abstraction would add roughly 15 percent overhead in code complexity and introduce one additional dependency.
Which integration approach would you recommend?

A. The third-party abstraction layer, on the grounds that multi-provider support is valuable for any application as a matter of long-term flexibility across vendors.
B. Direct integration with Claude through its SDK, because no multi-provider need exists and abstraction would add complexity that does not pay off.
C. Both integration paths in parallel, where the application uses each path on different runs to compare which performs better in production over time.
D. A custom multi-provider abstraction layer the team builds in-house so that the team controls every part of the abstraction the application uses for its API calls.

問題 #3
Your Claude application processes 50-page legal contracts and produces summaries with citation references back to the source. The team is debating whether to send each contract whole or split it into smaller pieces.
The contracts fit within Claude's context window. Initial testing shows that whole-document processing produces summaries with stronger cross-section reasoning but occasionally drifts on citation accuracy in later sections. Chunked processing produces stronger citation accuracy per chunk but loses cross-section reasoning.
The team has not decided which property matters more.
How would you guide the team's decision?

A. Examine the use cases where cross-section reasoning failures occur and assess whether whole- document processing meets the application's accuracy requirements across a representative sample of contracts.
B. Review the citation accuracy results from chunked processing and determine whether the loss of cross- section reasoning produces summaries that still meet the application's quality bar.
C. Identify which property matters more for the application's actual use case and let that decision drive the chunking approach, then validate the choice against representative contracts.
D. Assess the cost and latency implications of both approaches against the application's performance requirements before recommending which processing strategy to adopt.

問題 #4
A team has deployed a multi-agent system in which a primary agent decomposes user requests and delegates subtasks to three specialized subagents: one for data retrieval, one for analysis, and one for report generation.
In production, the team observes that subagents are making redundant tool calls, occasionally exceeding token budgets, and sometimes producing outputs that contradict each other - all of which the primary agent passes along without catching.
What is the most appropriate way to address these failures?

A. Strengthen the primary agent's management layer to enforce per-subagent tool budgets, validate outputs against a defined schema before passing them forward, and establish explicit handoff contracts between stages.
B. Collapse the three subagents into a single large-context model call that handles retrieval, analysis, and generation in one pass.
C. Add retry logic to each subagent so that when a tool call fails, the subagent retries automatically before escalating - and configure each subagent to log its tool calls and outputs to a shared trace so the team can audit redundancy and contradictions after the fact.
D. Give each subagent read access to the other subagents' outputs so they can identify and resolve contradictions without routing back through the primary agent.

問題 #5
You are configuring Claude Code for a new project. The team needs to set permissions, default model selections, and environment-specific behavior at the project level so the configuration is consistent across all developers working on the repository.
The Claude Code mechanism you would use is...

A. Environment variables that each developer sets on their own machine when working with Claude Code on the project.
B. The settings.json file, scoped at the project level so the configuration applies consistently across developers and persists with the repository.
C. A shared spreadsheet that lists configuration values for team members to reference and update by hand as the project evolves.
D. A system prompt embedded in every Claude Code conversation by each developer at the start of every session in the project.

解説：