04-subagent-verification.yaml 1.4 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859
  1. id: int-subagent-001
  2. name: Subagent Verification Integration Test
  3. description: |
  4. Test that verifies subagent delegation works correctly with the new
  5. SubagentEvaluator. Ensures task tool is called and subagent executes properly.
  6. category: business
  7. agent: openagent
  8. model: anthropic/claude-sonnet-4-5
  9. prompts:
  10. - text: |
  11. Create a simple utility module with a greet function that takes a name and returns "Hello, {name}!".
  12. Save it to evals/test_tmp/greet-utils.ts
  13. Delegate this to the coder agent to save time.
  14. expectContext: false
  15. - text: |
  16. Yes, delegate to the coder agent.
  17. delayMs: 2000
  18. # Expected behavior with delegation verification
  19. behavior:
  20. mustUseTools: [task]
  21. shouldDelegate: true
  22. minToolCalls: 1
  23. # NEW - Delegation/Subagent verification
  24. delegationExpectations:
  25. subagentType: "subagents/code/coder-agent"
  26. subagentMustUseTools: [write]
  27. subagentMinToolCalls: 1
  28. subagentMustComplete: true
  29. # NEW - Content validation for what the subagent writes
  30. contentExpectations:
  31. - filePath: "evals/test_tmp/greet-utils.ts"
  32. mustContain:
  33. - "greet"
  34. - "name"
  35. - "Hello"
  36. mustNotContain:
  37. - "console.log"
  38. minLength: 30
  39. # No expected violations - this is a positive test
  40. # Approval strategy
  41. approvalStrategy:
  42. type: auto-approve
  43. timeout: 120000
  44. tags:
  45. - integration
  46. - subagent-verification
  47. - delegation
  48. - v2-schema