Anthropic Claude Integration for Prompt Testing

// Test Claude's long-context performance
System: "Analyze the following contract and extract
  all liability clauses. Return as JSON."
User: "{{contract_text}}"  // Up to 200K tokens

// Test cases for document analysis
| contract_text    | expected_clauses | pass_criteria     |
|-----------------|------------------|-------------------|
| sample_nda.txt  | 3 liability items | All 3 identified  |
| vendor_agreement| 5 liability items | >= 4 identified   |
| employment.txt  | 2 liability items | All 2 identified  |

// Compare Claude Sonnet 4.5 vs Opus 4
// for accuracy vs cost tradeoffs