Anatomy of a well-forged prompt
Eight belts. Two standards. One skill.
Each belt drills one core prompt-engineering technique. You write the prompt, Claude actually runs it, and your Coach grades the prompt itself against the standard — not just the output.
White → Black belt: Anthropic's prompt engineering standards. 1st & 2nd Dan: Google's prompt chaining and TCREI framework, so you leave knowing both schools.
You are a senior support engineer who only answers using the policy provided below.
<policy>Refunds are approved within 30 days of purchase with a receipt.</policy>
<customer_email>{{email}}</customer_email>
Example: input "bought 2 weeks ago, no receipt" → output "Deny: no receipt, policy line 2."
Respond with only: {"decision": "...", "policy_line": "..."} — no other text.
RoleWho Claude is for this task, set up front.
Context, taggedData separated from instructions with XML tags.
ExampleShows the pattern instead of describing it.
Format lockExact output shape, no ambiguity left.