When applying few-shot prompting to improve output consistency for a task with ambiguous boundaries, what is the optimal number of examples to provide?
You are designing an MCP tool interface for backend system integration. When a tool fails, which fields should be included in a well- designed structured error response? Choose 2 correct answers.
When using tool_use with a strict JSON schema for data extraction, what does the Claude API explicitly guarantee?
How can an architect split a massive, monolithic CLAUDE. md file to selectively include external standards files based on the current module?
When a tool lookup (e.g., get_customer) returns multiple matches, what is the most reliable way for the agent to proceed?