You maintain a Claude application that uses Claude Sonnet 4.5 across several production workflows.
Anthropic released Claude Sonnet 4.7, which your evaluation suite shows performing 8% better on your highest-volume task. However, this version produces different output formatting on two of your structured- extraction prompts that downstream consumers parse with regex-based code.
To roll out the upgrade, you would...

• A. Switch the model identifier in the application's configuration to the new model, then monitor production for parsing failures to be fixed as failures surface.
• B. Rewrite the downstream parsing code to accept a wider range of output formats to ensure the application is resilient to future model upgrades without requiring prompt changes.
• C. Adjust the prompts to constrain output format, re-run evaluations against the expected parsing-layer schema, then roll the model out with a feature flag and the option to revert per workflow.
• D. Keep the older model in production to preserve the fragile parsing layer that may risk breaking with any model change.

You are extending a Claude agent with a capability that needs to be reusable across multiple teams in the organization, with each team able to invoke and use it independently.
How would you build the capability?

• A. As a custom tool embedded in this team's agent only, with other teams able to copy the implementation into their own agents when they need the capability.
• B. As a Skill or MCP server because both are purpose-built for cross-team reuse independently by each consuming team.
• C. As a shared library that each team imports into its own Claude application code, with each team responsible for keeping the library up to date in its integration.
• D. As a wrapper around an existing built-in tool that adds the missing functionality, on the grounds that built-in tools cover the reuse pattern when extended carefully.

Your agent is processing tasks that take 30 to 60 minutes to complete. Each task has well-defined intermediate checkpoints, and the team wants the agent to be able to resume from the most recent checkpoint if a process is interrupted.
How would you implement this resumability?

• A. Apply a checkpointing pattern that persists the agent's intermediate state and reloads that state when resuming an interrupted task.
• B. Restart the task from the beginning whenever a process is interrupted.
• C. Run two copies of the agent in parallel for every task and use whichever one finishes first as the source of truth.
• D. Increase the agent's timeout to several hours so that interruptions become rare enough to ignore in practice.

Your Claude agent has too many tools, and many of them have overlapping functionality. The agent often picks an inappropriate tool when several could plausibly handle a request.
How would you address the tool selection problem?

• A. Add more tools to cover every variation of the requests the agent handles, on the grounds that more tools give the agent more accurate options to choose from.
• B. Remove all tools and rely on the agent's general capability instead, with the application losing the workflows that previously relied on tools.
• C. Restructure the tool set by consolidating overlapping tools, removing unused tools, and clarifying tool descriptions so each tool has a distinct purpose.
• D. Add detailed examples to each tool's description so the agent can match incoming requests to the right tool by example, treating the examples as the team's selection mechanism.

A Claude application is occasionally refusing to answer questions that should be in scope, including questions the application has answered correctly in the past. You want to investigate.
What is the first step of your investigation?

• A. Test whether removing or adjusting those instructions resolves the behavior.
• B. Assess whether the refusals cluster around specific input patterns, question types, or time periods by analyzing the distribution of refused requests in the logs.
• C. Examine traces of the refused requests to identify what triggers the refusal, whether input patterns, system prompt content, or other context.
• D. Identify which questions have been refused and compare them against previously answered questions to determine what changed between the successful and refused interactions.