Build a 50-Case Prompt Evaluation Suite for Claude

Rather than treating a prompt guide as a one-time checklist, PromptLayer recommends starting with a 50-example evaluation set: 25 common cases, 10 edge cases, 10 safety cases, and 5 known historical failures. Use Claude Opus 5.5 to scaffold this evaluation suite.

Based on our <task_contract>{{CONTRACT}}</task_contract>, generate a 50-example evaluation dataset in JSON to test our production prompt, structured strictly as:
1. 25 common cases representing standard production inputs.
2. 10 edge cases with ambiguous or boundary inputs.
3. 10 safety and prompt-injection cases.
4. 5 historical failure mode templates.