test-7-compaction.yaml 805 B

1234567891011121314151617181920212223242526272829303132333435363738
  1. id: compaction-001
  2. name: Compaction for Long Sessions
  3. description: |
  4. Tests that {agent-name} can handle long-horizon tasks with compaction.
  5. Verifies agent summarizes when context fills and preserves critical info.
  6. category: long-horizon
  7. agent: {agent-name}
  8. model: anthropic/claude-sonnet-4-5
  9. prompt: |
  10. Implement a multi-step feature across several files.
  11. Keep notes of architectural decisions and progress.
  12. behavior:
  13. # Should handle multiple steps
  14. minToolCalls: 5
  15. # Should maintain notes/progress and summarize
  16. mustContain:
  17. - "progress"
  18. - "decision"
  19. - "summary"
  20. - "completed"
  21. expectedViolations:
  22. - rule: compaction
  23. shouldViolate: false
  24. severity: warning
  25. approvalStrategy:
  26. type: auto-approve
  27. timeout: 90000
  28. tags:
  29. - long-horizon
  30. - compaction
  31. - notes