Anthropic workshop breaks down Claude prompting

This is less a prompt cheat code than a reminder that reliable agents need operating rules, not vibes.

• –Anthropic draws a hard line between chatbot prompting and agent prompting: tell the model when to use tools, not just what answer you want.
• –The strongest advice is to structure the environment first: tools, task scope, and success criteria, then let Claude reason and iterate.
• –The guidance maps directly to Claude Code and MCP-style workflows, where the prompt is only one piece of the harness.
• –Verification matters as much as generation because tool use makes false confidence and side effects more likely.
• –Net: the session is useful because it turns “prompt better” into “design the whole workflow better.”