mock_parity_scenarios.json 3.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109
  1. [
  2. {
  3. "name": "streaming_text",
  4. "category": "baseline",
  5. "description": "Validates streamed assistant text with no tool calls.",
  6. "parity_refs": [
  7. "Mock parity harness \u2014 milestone 1",
  8. "Streaming response support validated by the mock parity harness"
  9. ]
  10. },
  11. {
  12. "name": "read_file_roundtrip",
  13. "category": "file-tools",
  14. "description": "Exercises read_file tool execution and final assistant synthesis.",
  15. "parity_refs": [
  16. "Mock parity harness \u2014 milestone 1",
  17. "File tools \u2014 harness-validated flows"
  18. ]
  19. },
  20. {
  21. "name": "grep_chunk_assembly",
  22. "category": "file-tools",
  23. "description": "Validates grep_search partial JSON chunk assembly and follow-up synthesis.",
  24. "parity_refs": [
  25. "Mock parity harness \u2014 milestone 1",
  26. "File tools \u2014 harness-validated flows"
  27. ]
  28. },
  29. {
  30. "name": "write_file_allowed",
  31. "category": "file-tools",
  32. "description": "Confirms workspace-write write_file success and filesystem side effects.",
  33. "parity_refs": [
  34. "Mock parity harness \u2014 milestone 1",
  35. "File tools \u2014 harness-validated flows"
  36. ]
  37. },
  38. {
  39. "name": "write_file_denied",
  40. "category": "permissions",
  41. "description": "Confirms read-only mode blocks write_file with an error result.",
  42. "parity_refs": [
  43. "Mock parity harness \u2014 milestone 1",
  44. "Permission enforcement across tool paths"
  45. ]
  46. },
  47. {
  48. "name": "multi_tool_turn_roundtrip",
  49. "category": "multi-tool-turns",
  50. "description": "Executes read_file and grep_search in the same assistant turn before the final reply.",
  51. "parity_refs": [
  52. "Mock parity harness \u2014 milestone 2 (behavioral expansion)",
  53. "Multi-tool assistant turns"
  54. ]
  55. },
  56. {
  57. "name": "bash_stdout_roundtrip",
  58. "category": "bash",
  59. "description": "Validates bash execution and stdout roundtrip in danger-full-access mode.",
  60. "parity_refs": [
  61. "Mock parity harness \u2014 milestone 2 (behavioral expansion)",
  62. "Bash tool \u2014 upstream has 18 submodules, Rust has 1:"
  63. ]
  64. },
  65. {
  66. "name": "bash_permission_prompt_approved",
  67. "category": "permissions",
  68. "description": "Exercises workspace-write to bash escalation with a positive approval response.",
  69. "parity_refs": [
  70. "Mock parity harness \u2014 milestone 2 (behavioral expansion)",
  71. "Permission enforcement across tool paths"
  72. ]
  73. },
  74. {
  75. "name": "bash_permission_prompt_denied",
  76. "category": "permissions",
  77. "description": "Exercises workspace-write to bash escalation with a denied approval response.",
  78. "parity_refs": [
  79. "Mock parity harness \u2014 milestone 2 (behavioral expansion)",
  80. "Permission enforcement across tool paths"
  81. ]
  82. },
  83. {
  84. "name": "plugin_tool_roundtrip",
  85. "category": "plugin-paths",
  86. "description": "Loads an external plugin tool and executes it through the runtime tool registry.",
  87. "parity_refs": [
  88. "Mock parity harness \u2014 milestone 2 (behavioral expansion)",
  89. "Plugin tool execution path"
  90. ]
  91. },
  92. {
  93. "name": "auto_compact_triggered",
  94. "category": "session-compaction",
  95. "description": "Verifies auto-compact fires when cumulative input tokens exceed the configured threshold.",
  96. "parity_refs": [
  97. "Session compaction behavior matching",
  98. "auto_compaction threshold from env"
  99. ]
  100. },
  101. {
  102. "name": "token_cost_reporting",
  103. "category": "token-usage",
  104. "description": "Confirms usage token counts and estimated_cost appear in JSON output.",
  105. "parity_refs": [
  106. "Token counting / cost tracking accuracy"
  107. ]
  108. }
  109. ]