Your Claude application has multi-step workflows where each step's output is needed only briefly before the agent moves on. The cumulative tool output is filling the context window with content that is no longer relevant.
How would you handle the accumulating tool output?

• A. Keep every tool output in the context indefinitely so the agent has the full record of every step it has executed during the workflow.
• B. Apply prompt caching to the accumulated tool outputs so the application does not re-pay for the older content on each subsequent step.
• C. Apply tool output pruning to remove tool outputs that are no longer needed by later steps in the workflow.
• D. Switch to a smaller Claude model that processes context more efficiently and treat any quality loss as a tradeoff for the cost reduction.

Your Claude application's content policy specifies categories of content it should not produce under any circumstance. The application currently has no mechanism to enforce this policy, and content matching these categories is appearing in the application's output.
How would you enforce the content policy?

• A. Move enforcement to users by asking them to report content policy violations after the violating content has already reached them in the application's responses.
• B. Remove the content policy entirely and let any output reach users during normal operation, accepting whatever content the application produces in response to incoming traffic.
• C. Add deterministic output filtering that checks responses against the content policy before they reach users.
• D. Enhance the system prompt to contain explicit instructions for the categories to avoid, complete with examples of each category. Treat the strengthened prompt as the primary enforcement mechanism for the application's content policy across all responses.

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. 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.
• B. Keep the older model in production to preserve the fragile parsing layer that may risk breaking with any model change.
• C. 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.
• D. 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.

The Claude application your team built has grown over six months, and the prompt-handling code has accumulated duplication and tangled control flow. The functionality is working, but new features are getting harder to add.
How would you address this?

• A. Refactor the prompt-handling code in small increments as part of each new feature ticket, treating the cleanup as a side effect of feature work.
• B. Continue adding features and plan a refactoring pass after the next two release cycles when the team has more bandwidth for internal work.
• C. Move all the prompt-handling code into a single large function to reduce the number of files developers have to navigate when reading the code.
• D. Plan a refactoring pass to consolidate duplicated logic, separate concerns, and simplify control flow before adding new features.

You are deciding between deploying a Claude-powered agent on Anthropic's hosted infrastructure or self- hosting under a "bring your own cloud" model in your own AWS account. The agent processes customer data subject to your enterprise's data residency policies, but the team wants to ship quickly and avoid managing infrastructure.
Which deployment model would you recommend?

• A. Self-hosting under BYOC to satisfy the data residency requirement, while working with the infrastructure team to reduce the operational overhead of managing the deployment.
• B. Deploying on Anthropic-hosted infrastructure to meet the team's shipping timeline, and flagging the data residency requirement for a follow-up compliance review after launch.
• C. Self-hosting under BYOC for an initial pilot, then evaluating whether to migrate to Anthropic-hosted infrastructure once the agent's data-handling patterns are better understood.
• D. Deploying on Anthropic-hosted infrastructure while the team drafts a request to update the enterprise data residency policy to accommodate hosted AI deployments.