@openpond/evals 0.3.0 → 0.4.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (53) hide show
  1. package/CONTRACT.md +11 -3
  2. package/README.md +45 -9
  3. package/RELEASING.md +13 -46
  4. package/dist/benchmarks.js +218 -0
  5. package/dist/builtin-benchmarks/harness-refiner.js +1267 -0
  6. package/dist/compatibility.js +2 -2
  7. package/dist/conformance.js +1 -2
  8. package/dist/evidence/authoring.js +1 -1
  9. package/dist/evidence/conformance.js +1 -1
  10. package/dist/evidence/contracts.js +1 -2
  11. package/dist/evidence/eligibility.js +1 -1
  12. package/dist/graders.js +1 -1
  13. package/dist/harness.js +33 -101
  14. package/dist/index.js +3 -3
  15. package/dist/model-improvement-qualification.js +125 -0
  16. package/dist/review-conformance.js +209 -0
  17. package/dist/runs.js +1 -9
  18. package/dist/tasksets.js +2 -15
  19. package/dist/types/benchmarks.d.ts +293 -0
  20. package/dist/types/benchmarks.d.ts.map +1 -0
  21. package/dist/types/builtin-benchmarks/harness-refiner.d.ts +119 -0
  22. package/dist/types/builtin-benchmarks/harness-refiner.d.ts.map +1 -0
  23. package/dist/types/compatibility.d.ts +1 -1
  24. package/dist/types/compatibility.d.ts.map +1 -1
  25. package/dist/types/conformance.d.ts +2 -0
  26. package/dist/types/conformance.d.ts.map +1 -1
  27. package/dist/types/evidence/conformance.d.ts +24 -24
  28. package/dist/types/evidence/contracts.d.ts +26 -26
  29. package/dist/types/harness.d.ts +3 -409
  30. package/dist/types/harness.d.ts.map +1 -1
  31. package/dist/types/index.d.ts +3 -3
  32. package/dist/types/index.d.ts.map +1 -1
  33. package/dist/types/model-improvement-qualification.d.ts +222 -0
  34. package/dist/types/model-improvement-qualification.d.ts.map +1 -0
  35. package/dist/types/review-conformance.d.ts +607 -0
  36. package/dist/types/review-conformance.d.ts.map +1 -0
  37. package/dist/types/runs.d.ts +1 -9
  38. package/dist/types/runs.d.ts.map +1 -1
  39. package/dist/types/tasksets.d.ts +2 -25
  40. package/dist/types/tasksets.d.ts.map +1 -1
  41. package/package.json +18 -10
  42. package/dist/common.js +0 -67
  43. package/dist/harness-improvements.js +0 -329
  44. package/dist/harness-workspaces.js +0 -368
  45. package/dist/sha256.js +0 -91
  46. package/dist/types/common.d.ts +0 -51
  47. package/dist/types/common.d.ts.map +0 -1
  48. package/dist/types/harness-improvements.d.ts +0 -523
  49. package/dist/types/harness-improvements.d.ts.map +0 -1
  50. package/dist/types/harness-workspaces.d.ts +0 -802
  51. package/dist/types/harness-workspaces.d.ts.map +0 -1
  52. package/dist/types/sha256.d.ts +0 -2
  53. package/dist/types/sha256.d.ts.map +0 -1
@@ -13,9 +13,9 @@ export declare const EvidenceArtifactRefSchema: z.ZodObject<{
13
13
  id: z.ZodString;
14
14
  }, z.core.$strict>;
15
15
  export declare const WorkProcessStepKindSchema: z.ZodEnum<{
16
+ approval: "approval";
16
17
  validation: "validation";
17
18
  artifact: "artifact";
18
- approval: "approval";
19
19
  tool: "tool";
20
20
  state_transition: "state_transition";
21
21
  question: "question";
@@ -26,8 +26,8 @@ export declare const WorkProcessLayerSchema: z.ZodEnum<{
26
26
  environment: "environment";
27
27
  }>;
28
28
  export declare const WorkProcessActionSchema: z.ZodEnum<{
29
- turn_completed: "turn_completed";
30
29
  turn_started: "turn_started";
30
+ turn_completed: "turn_completed";
31
31
  turn_failed: "turn_failed";
32
32
  turn_cancelled: "turn_cancelled";
33
33
  turn_timed_out: "turn_timed_out";
@@ -57,20 +57,20 @@ export declare const WorkProcessStepStatusSchema: z.ZodEnum<{
57
57
  }>;
58
58
  export declare const WorkToolCategorySchema: z.ZodEnum<{
59
59
  agent: "agent";
60
- connected_app: "connected_app";
60
+ command: "command";
61
61
  filesystem: "filesystem";
62
62
  source_control: "source_control";
63
- command: "command";
64
63
  browser: "browser";
64
+ connected_app: "connected_app";
65
65
  sandbox: "sandbox";
66
66
  other: "other";
67
67
  }>;
68
68
  export declare const WorkValidationKindSchema: z.ZodEnum<{
69
69
  test: "test";
70
- other: "other";
71
70
  structural: "structural";
72
71
  visual: "visual";
73
72
  user_review: "user_review";
73
+ other: "other";
74
74
  }>;
75
75
  export declare const WorkTransitionStateSchema: z.ZodEnum<{
76
76
  timeout: "timeout";
@@ -88,10 +88,10 @@ export declare const WorkInterventionOutcomeSchema: z.ZodEnum<{
88
88
  }>;
89
89
  export declare const WorkProcessErrorClassSchema: z.ZodEnum<{
90
90
  unknown: "unknown";
91
- policy: "policy";
92
91
  timeout: "timeout";
93
92
  cancelled: "cancelled";
94
93
  validation: "validation";
94
+ policy: "policy";
95
95
  environment: "environment";
96
96
  infrastructure: "infrastructure";
97
97
  }>;
@@ -111,17 +111,17 @@ export declare const WorkProcessStepSchema: z.ZodObject<{
111
111
  environment: "environment";
112
112
  }>;
113
113
  kind: z.ZodEnum<{
114
+ approval: "approval";
114
115
  validation: "validation";
115
116
  artifact: "artifact";
116
- approval: "approval";
117
117
  tool: "tool";
118
118
  state_transition: "state_transition";
119
119
  question: "question";
120
120
  cleanup: "cleanup";
121
121
  }>;
122
122
  action: z.ZodEnum<{
123
- turn_completed: "turn_completed";
124
123
  turn_started: "turn_started";
124
+ turn_completed: "turn_completed";
125
125
  turn_failed: "turn_failed";
126
126
  turn_cancelled: "turn_cancelled";
127
127
  turn_timed_out: "turn_timed_out";
@@ -162,20 +162,20 @@ export declare const WorkProcessStepSchema: z.ZodObject<{
162
162
  attributes: z.ZodObject<{
163
163
  toolCategory: z.ZodNullable<z.ZodEnum<{
164
164
  agent: "agent";
165
- connected_app: "connected_app";
165
+ command: "command";
166
166
  filesystem: "filesystem";
167
167
  source_control: "source_control";
168
- command: "command";
169
168
  browser: "browser";
169
+ connected_app: "connected_app";
170
170
  sandbox: "sandbox";
171
171
  other: "other";
172
172
  }>>;
173
173
  validationKind: z.ZodNullable<z.ZodEnum<{
174
174
  test: "test";
175
- other: "other";
176
175
  structural: "structural";
177
176
  visual: "visual";
178
177
  user_review: "user_review";
178
+ other: "other";
179
179
  }>>;
180
180
  transitionState: z.ZodNullable<z.ZodEnum<{
181
181
  timeout: "timeout";
@@ -198,10 +198,10 @@ export declare const WorkProcessStepSchema: z.ZodObject<{
198
198
  memoryPeakBytes: z.ZodNullable<z.ZodNumber>;
199
199
  errorClass: z.ZodNullable<z.ZodEnum<{
200
200
  unknown: "unknown";
201
- policy: "policy";
202
201
  timeout: "timeout";
203
202
  cancelled: "cancelled";
204
203
  validation: "validation";
204
+ policy: "policy";
205
205
  environment: "environment";
206
206
  infrastructure: "infrastructure";
207
207
  }>>;
@@ -229,17 +229,17 @@ export declare const WorkProcessTraceContentSchema: z.ZodObject<{
229
229
  environment: "environment";
230
230
  }>;
231
231
  kind: z.ZodEnum<{
232
+ approval: "approval";
232
233
  validation: "validation";
233
234
  artifact: "artifact";
234
- approval: "approval";
235
235
  tool: "tool";
236
236
  state_transition: "state_transition";
237
237
  question: "question";
238
238
  cleanup: "cleanup";
239
239
  }>;
240
240
  action: z.ZodEnum<{
241
- turn_completed: "turn_completed";
242
241
  turn_started: "turn_started";
242
+ turn_completed: "turn_completed";
243
243
  turn_failed: "turn_failed";
244
244
  turn_cancelled: "turn_cancelled";
245
245
  turn_timed_out: "turn_timed_out";
@@ -280,20 +280,20 @@ export declare const WorkProcessTraceContentSchema: z.ZodObject<{
280
280
  attributes: z.ZodObject<{
281
281
  toolCategory: z.ZodNullable<z.ZodEnum<{
282
282
  agent: "agent";
283
- connected_app: "connected_app";
283
+ command: "command";
284
284
  filesystem: "filesystem";
285
285
  source_control: "source_control";
286
- command: "command";
287
286
  browser: "browser";
287
+ connected_app: "connected_app";
288
288
  sandbox: "sandbox";
289
289
  other: "other";
290
290
  }>>;
291
291
  validationKind: z.ZodNullable<z.ZodEnum<{
292
292
  test: "test";
293
- other: "other";
294
293
  structural: "structural";
295
294
  visual: "visual";
296
295
  user_review: "user_review";
296
+ other: "other";
297
297
  }>>;
298
298
  transitionState: z.ZodNullable<z.ZodEnum<{
299
299
  timeout: "timeout";
@@ -316,10 +316,10 @@ export declare const WorkProcessTraceContentSchema: z.ZodObject<{
316
316
  memoryPeakBytes: z.ZodNullable<z.ZodNumber>;
317
317
  errorClass: z.ZodNullable<z.ZodEnum<{
318
318
  unknown: "unknown";
319
- policy: "policy";
320
319
  timeout: "timeout";
321
320
  cancelled: "cancelled";
322
321
  validation: "validation";
322
+ policy: "policy";
323
323
  environment: "environment";
324
324
  infrastructure: "infrastructure";
325
325
  }>>;
@@ -348,17 +348,17 @@ export declare const WorkProcessTraceSchema: z.ZodObject<{
348
348
  environment: "environment";
349
349
  }>;
350
350
  kind: z.ZodEnum<{
351
+ approval: "approval";
351
352
  validation: "validation";
352
353
  artifact: "artifact";
353
- approval: "approval";
354
354
  tool: "tool";
355
355
  state_transition: "state_transition";
356
356
  question: "question";
357
357
  cleanup: "cleanup";
358
358
  }>;
359
359
  action: z.ZodEnum<{
360
- turn_completed: "turn_completed";
361
360
  turn_started: "turn_started";
361
+ turn_completed: "turn_completed";
362
362
  turn_failed: "turn_failed";
363
363
  turn_cancelled: "turn_cancelled";
364
364
  turn_timed_out: "turn_timed_out";
@@ -399,20 +399,20 @@ export declare const WorkProcessTraceSchema: z.ZodObject<{
399
399
  attributes: z.ZodObject<{
400
400
  toolCategory: z.ZodNullable<z.ZodEnum<{
401
401
  agent: "agent";
402
- connected_app: "connected_app";
402
+ command: "command";
403
403
  filesystem: "filesystem";
404
404
  source_control: "source_control";
405
- command: "command";
406
405
  browser: "browser";
406
+ connected_app: "connected_app";
407
407
  sandbox: "sandbox";
408
408
  other: "other";
409
409
  }>>;
410
410
  validationKind: z.ZodNullable<z.ZodEnum<{
411
411
  test: "test";
412
- other: "other";
413
412
  structural: "structural";
414
413
  visual: "visual";
415
414
  user_review: "user_review";
415
+ other: "other";
416
416
  }>>;
417
417
  transitionState: z.ZodNullable<z.ZodEnum<{
418
418
  timeout: "timeout";
@@ -435,10 +435,10 @@ export declare const WorkProcessTraceSchema: z.ZodObject<{
435
435
  memoryPeakBytes: z.ZodNullable<z.ZodNumber>;
436
436
  errorClass: z.ZodNullable<z.ZodEnum<{
437
437
  unknown: "unknown";
438
- policy: "policy";
439
438
  timeout: "timeout";
440
439
  cancelled: "cancelled";
441
440
  validation: "validation";
441
+ policy: "policy";
442
442
  environment: "environment";
443
443
  infrastructure: "infrastructure";
444
444
  }>>;
@@ -479,8 +479,8 @@ export declare const WorkEvidenceReceiptContentSchema: z.ZodObject<{
479
479
  id: z.ZodString;
480
480
  source: z.ZodObject<{
481
481
  surface: z.ZodEnum<{
482
- hosted: "hosted";
483
482
  desktop: "desktop";
483
+ hosted: "hosted";
484
484
  }>;
485
485
  experience: z.ZodEnum<{
486
486
  work: "work";
@@ -609,8 +609,8 @@ export declare const WorkEvidenceReceiptSchema: z.ZodObject<{
609
609
  id: z.ZodString;
610
610
  source: z.ZodObject<{
611
611
  surface: z.ZodEnum<{
612
- hosted: "hosted";
613
612
  desktop: "desktop";
613
+ hosted: "hosted";
614
614
  }>;
615
615
  experience: z.ZodEnum<{
616
616
  work: "work";
@@ -1,332 +1,7 @@
1
1
  import { z } from "zod";
2
- import { ImmutableArtifactRefSchema } from "./common.js";
2
+ import { HarnessTraceSchema, ImmutableArtifactRefSchema, type ModelAction, type ToolObservation } from "@openpond/harness";
3
3
  import { type AttemptReceipt, type RunManifest } from "./runs.js";
4
- export declare const PortabilityReportSchema: z.ZodObject<{
5
- portable: z.ZodBoolean;
6
- blockers: z.ZodArray<z.ZodString>;
7
- localOnlyAssetRefs: z.ZodArray<z.ZodString>;
8
- hostPrivateAssetRefs: z.ZodArray<z.ZodString>;
9
- }, z.core.$strict>;
10
- export declare const AgentSnapshotContentSchema: z.ZodObject<{
11
- schemaVersion: z.ZodLiteral<"openpond.agentSnapshot.v2">;
12
- id: z.ZodString;
13
- sourceRelease: z.ZodNullable<z.ZodObject<{
14
- id: z.ZodString;
15
- contentHash: z.ZodString;
16
- }, z.core.$strict>>;
17
- instructions: z.ZodArray<z.ZodObject<{
18
- id: z.ZodString;
19
- path: z.ZodString;
20
- contentHash: z.ZodString;
21
- sizeBytes: z.ZodNumber;
22
- mediaType: z.ZodString;
23
- visibility: z.ZodEnum<{
24
- policy: "policy";
25
- verifier: "verifier";
26
- host_private: "host_private";
27
- }>;
28
- }, z.core.$strict>>;
29
- skills: z.ZodArray<z.ZodObject<{
30
- id: z.ZodString;
31
- path: z.ZodString;
32
- contentHash: z.ZodString;
33
- sizeBytes: z.ZodNumber;
34
- mediaType: z.ZodString;
35
- visibility: z.ZodEnum<{
36
- policy: "policy";
37
- verifier: "verifier";
38
- host_private: "host_private";
39
- }>;
40
- }, z.core.$strict>>;
41
- agents: z.ZodArray<z.ZodObject<{
42
- id: z.ZodString;
43
- path: z.ZodString;
44
- contentHash: z.ZodString;
45
- sizeBytes: z.ZodNumber;
46
- mediaType: z.ZodString;
47
- visibility: z.ZodEnum<{
48
- policy: "policy";
49
- verifier: "verifier";
50
- host_private: "host_private";
51
- }>;
52
- }, z.core.$strict>>;
53
- toolDeclarations: z.ZodArray<z.ZodObject<{
54
- name: z.ZodString;
55
- description: z.ZodString;
56
- inputSchema: z.ZodRecord<z.ZodString, z.ZodUnknown>;
57
- inputSchemaHash: z.ZodString;
58
- sideEffect: z.ZodEnum<{
59
- read: "read";
60
- write: "write";
61
- }>;
62
- timeoutMs: z.ZodNumber;
63
- }, z.core.$strict>>;
64
- capabilityRequirements: z.ZodArray<z.ZodObject<{
65
- id: z.ZodString;
66
- required: z.ZodBoolean;
67
- scopes: z.ZodArray<z.ZodString>;
68
- }, z.core.$strict>>;
69
- dependencyLock: z.ZodObject<{
70
- id: z.ZodString;
71
- path: z.ZodString;
72
- contentHash: z.ZodString;
73
- sizeBytes: z.ZodNumber;
74
- mediaType: z.ZodString;
75
- visibility: z.ZodEnum<{
76
- policy: "policy";
77
- verifier: "verifier";
78
- host_private: "host_private";
79
- }>;
80
- }, z.core.$strict>;
81
- portability: z.ZodObject<{
82
- portable: z.ZodBoolean;
83
- blockers: z.ZodArray<z.ZodString>;
84
- localOnlyAssetRefs: z.ZodArray<z.ZodString>;
85
- hostPrivateAssetRefs: z.ZodArray<z.ZodString>;
86
- }, z.core.$strict>;
87
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
88
- }, z.core.$strict>;
89
- export declare const AgentSnapshotSchema: z.ZodObject<{
90
- schemaVersion: z.ZodLiteral<"openpond.agentSnapshot.v2">;
91
- id: z.ZodString;
92
- sourceRelease: z.ZodNullable<z.ZodObject<{
93
- id: z.ZodString;
94
- contentHash: z.ZodString;
95
- }, z.core.$strict>>;
96
- instructions: z.ZodArray<z.ZodObject<{
97
- id: z.ZodString;
98
- path: z.ZodString;
99
- contentHash: z.ZodString;
100
- sizeBytes: z.ZodNumber;
101
- mediaType: z.ZodString;
102
- visibility: z.ZodEnum<{
103
- policy: "policy";
104
- verifier: "verifier";
105
- host_private: "host_private";
106
- }>;
107
- }, z.core.$strict>>;
108
- skills: z.ZodArray<z.ZodObject<{
109
- id: z.ZodString;
110
- path: z.ZodString;
111
- contentHash: z.ZodString;
112
- sizeBytes: z.ZodNumber;
113
- mediaType: z.ZodString;
114
- visibility: z.ZodEnum<{
115
- policy: "policy";
116
- verifier: "verifier";
117
- host_private: "host_private";
118
- }>;
119
- }, z.core.$strict>>;
120
- agents: z.ZodArray<z.ZodObject<{
121
- id: z.ZodString;
122
- path: z.ZodString;
123
- contentHash: z.ZodString;
124
- sizeBytes: z.ZodNumber;
125
- mediaType: z.ZodString;
126
- visibility: z.ZodEnum<{
127
- policy: "policy";
128
- verifier: "verifier";
129
- host_private: "host_private";
130
- }>;
131
- }, z.core.$strict>>;
132
- toolDeclarations: z.ZodArray<z.ZodObject<{
133
- name: z.ZodString;
134
- description: z.ZodString;
135
- inputSchema: z.ZodRecord<z.ZodString, z.ZodUnknown>;
136
- inputSchemaHash: z.ZodString;
137
- sideEffect: z.ZodEnum<{
138
- read: "read";
139
- write: "write";
140
- }>;
141
- timeoutMs: z.ZodNumber;
142
- }, z.core.$strict>>;
143
- capabilityRequirements: z.ZodArray<z.ZodObject<{
144
- id: z.ZodString;
145
- required: z.ZodBoolean;
146
- scopes: z.ZodArray<z.ZodString>;
147
- }, z.core.$strict>>;
148
- dependencyLock: z.ZodObject<{
149
- id: z.ZodString;
150
- path: z.ZodString;
151
- contentHash: z.ZodString;
152
- sizeBytes: z.ZodNumber;
153
- mediaType: z.ZodString;
154
- visibility: z.ZodEnum<{
155
- policy: "policy";
156
- verifier: "verifier";
157
- host_private: "host_private";
158
- }>;
159
- }, z.core.$strict>;
160
- portability: z.ZodObject<{
161
- portable: z.ZodBoolean;
162
- blockers: z.ZodArray<z.ZodString>;
163
- localOnlyAssetRefs: z.ZodArray<z.ZodString>;
164
- hostPrivateAssetRefs: z.ZodArray<z.ZodString>;
165
- }, z.core.$strict>;
166
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
167
- contentHash: z.ZodString;
168
- }, z.core.$strict>;
169
- export declare const LifecycleContractSchema: z.ZodObject<{
170
- create: z.ZodLiteral<true>;
171
- reset: z.ZodLiteral<true>;
172
- step: z.ZodLiteral<true>;
173
- collect: z.ZodLiteral<true>;
174
- destroy: z.ZodLiteral<true>;
175
- resetScope: z.ZodEnum<{
176
- task: "task";
177
- attempt: "attempt";
178
- }>;
179
- }, z.core.$strict>;
180
- export declare const GraderInterfaceContractSchema: z.ZodObject<{
181
- visibleEvidence: z.ZodArray<z.ZodString>;
182
- privilegedEvidence: z.ZodArray<z.ZodString>;
183
- privateVerifierIsolation: z.ZodBoolean;
184
- }, z.core.$strict>;
185
- export declare const HarnessReleaseContentSchema: z.ZodObject<{
186
- schemaVersion: z.ZodLiteral<"openpond.harnessRelease.v2">;
187
- id: z.ZodString;
188
- agentSnapshot: z.ZodObject<{
189
- id: z.ZodString;
190
- contentHash: z.ZodString;
191
- }, z.core.$strict>;
192
- program: z.ZodObject<{
193
- id: z.ZodString;
194
- path: z.ZodString;
195
- contentHash: z.ZodString;
196
- sizeBytes: z.ZodNumber;
197
- mediaType: z.ZodString;
198
- visibility: z.ZodEnum<{
199
- policy: "policy";
200
- verifier: "verifier";
201
- host_private: "host_private";
202
- }>;
203
- }, z.core.$strict>;
204
- tools: z.ZodArray<z.ZodObject<{
205
- name: z.ZodString;
206
- description: z.ZodString;
207
- inputSchema: z.ZodRecord<z.ZodString, z.ZodUnknown>;
208
- inputSchemaHash: z.ZodString;
209
- sideEffect: z.ZodEnum<{
210
- read: "read";
211
- write: "write";
212
- }>;
213
- timeoutMs: z.ZodNumber;
214
- }, z.core.$strict>>;
215
- lifecycle: z.ZodObject<{
216
- create: z.ZodLiteral<true>;
217
- reset: z.ZodLiteral<true>;
218
- step: z.ZodLiteral<true>;
219
- collect: z.ZodLiteral<true>;
220
- destroy: z.ZodLiteral<true>;
221
- resetScope: z.ZodEnum<{
222
- task: "task";
223
- attempt: "attempt";
224
- }>;
225
- }, z.core.$strict>;
226
- graderInterface: z.ZodObject<{
227
- visibleEvidence: z.ZodArray<z.ZodString>;
228
- privilegedEvidence: z.ZodArray<z.ZodString>;
229
- privateVerifierIsolation: z.ZodBoolean;
230
- }, z.core.$strict>;
231
- files: z.ZodArray<z.ZodObject<{
232
- id: z.ZodString;
233
- path: z.ZodString;
234
- contentHash: z.ZodString;
235
- sizeBytes: z.ZodNumber;
236
- mediaType: z.ZodString;
237
- visibility: z.ZodEnum<{
238
- policy: "policy";
239
- verifier: "verifier";
240
- host_private: "host_private";
241
- }>;
242
- }, z.core.$strict>>;
243
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
244
- }, z.core.$strict>;
245
- export declare const HarnessReleaseSchema: z.ZodObject<{
246
- schemaVersion: z.ZodLiteral<"openpond.harnessRelease.v2">;
247
- id: z.ZodString;
248
- agentSnapshot: z.ZodObject<{
249
- id: z.ZodString;
250
- contentHash: z.ZodString;
251
- }, z.core.$strict>;
252
- program: z.ZodObject<{
253
- id: z.ZodString;
254
- path: z.ZodString;
255
- contentHash: z.ZodString;
256
- sizeBytes: z.ZodNumber;
257
- mediaType: z.ZodString;
258
- visibility: z.ZodEnum<{
259
- policy: "policy";
260
- verifier: "verifier";
261
- host_private: "host_private";
262
- }>;
263
- }, z.core.$strict>;
264
- tools: z.ZodArray<z.ZodObject<{
265
- name: z.ZodString;
266
- description: z.ZodString;
267
- inputSchema: z.ZodRecord<z.ZodString, z.ZodUnknown>;
268
- inputSchemaHash: z.ZodString;
269
- sideEffect: z.ZodEnum<{
270
- read: "read";
271
- write: "write";
272
- }>;
273
- timeoutMs: z.ZodNumber;
274
- }, z.core.$strict>>;
275
- lifecycle: z.ZodObject<{
276
- create: z.ZodLiteral<true>;
277
- reset: z.ZodLiteral<true>;
278
- step: z.ZodLiteral<true>;
279
- collect: z.ZodLiteral<true>;
280
- destroy: z.ZodLiteral<true>;
281
- resetScope: z.ZodEnum<{
282
- task: "task";
283
- attempt: "attempt";
284
- }>;
285
- }, z.core.$strict>;
286
- graderInterface: z.ZodObject<{
287
- visibleEvidence: z.ZodArray<z.ZodString>;
288
- privilegedEvidence: z.ZodArray<z.ZodString>;
289
- privateVerifierIsolation: z.ZodBoolean;
290
- }, z.core.$strict>;
291
- files: z.ZodArray<z.ZodObject<{
292
- id: z.ZodString;
293
- path: z.ZodString;
294
- contentHash: z.ZodString;
295
- sizeBytes: z.ZodNumber;
296
- mediaType: z.ZodString;
297
- visibility: z.ZodEnum<{
298
- policy: "policy";
299
- verifier: "verifier";
300
- host_private: "host_private";
301
- }>;
302
- }, z.core.$strict>>;
303
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
304
- contentHash: z.ZodString;
305
- }, z.core.$strict>;
306
- export declare const ModelActionSchema: z.ZodObject<{
307
- id: z.ZodString;
308
- turn: z.ZodNumber;
309
- kind: z.ZodEnum<{
310
- message: "message";
311
- terminal: "terminal";
312
- tool_call: "tool_call";
313
- }>;
314
- name: z.ZodNullable<z.ZodString>;
315
- arguments: z.ZodRecord<z.ZodString, z.ZodUnknown>;
316
- content: z.ZodNullable<z.ZodString>;
317
- }, z.core.$strict>;
318
- export declare const ToolObservationSchema: z.ZodObject<{
319
- actionId: z.ZodString;
320
- turn: z.ZodNumber;
321
- terminal: z.ZodBoolean;
322
- output: z.ZodRecord<z.ZodString, z.ZodUnknown>;
323
- artifactRefs: z.ZodArray<z.ZodObject<{
324
- id: z.ZodString;
325
- contentHash: z.ZodString;
326
- mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
327
- sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
328
- }, z.core.$strict>>;
329
- }, z.core.$strict>;
4
+ export * from "@openpond/harness/harness";
330
5
  export type HarnessLease = {
331
6
  id: string;
332
7
  metadata: Record<string, unknown>;
@@ -357,7 +32,7 @@ export type HarnessExecutionInput = {
357
32
  };
358
33
  export type HarnessExecutionResult = {
359
34
  receipt: AttemptReceipt;
360
- trace: unknown;
35
+ trace: z.infer<typeof HarnessTraceSchema>;
361
36
  output: Record<string, unknown>;
362
37
  };
363
38
  export interface HarnessExecutor {
@@ -381,85 +56,4 @@ export declare function executeRuntimeProtocol(input: {
381
56
  signal?: AbortSignal;
382
57
  now?: () => string;
383
58
  }): Promise<HarnessExecutionResult>;
384
- export declare function createAgentSnapshot(input: z.input<typeof AgentSnapshotContentSchema>): AgentSnapshot;
385
- export declare function createHarnessRelease(input: z.input<typeof HarnessReleaseContentSchema>): HarnessRelease;
386
- export declare const HarnessLifecycleEventSchema: z.ZodObject<{
387
- sequence: z.ZodNumber;
388
- type: z.ZodEnum<{
389
- reset: "reset";
390
- terminal: "terminal";
391
- created: "created";
392
- action: "action";
393
- observation: "observation";
394
- failure: "failure";
395
- collected: "collected";
396
- destroyed: "destroyed";
397
- }>;
398
- payloadHash: z.ZodString;
399
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
400
- }, z.core.$strict>;
401
- export declare const HarnessTraceSchema: z.ZodObject<{
402
- schemaVersion: z.ZodLiteral<"openpond.harnessTrace.v1">;
403
- manifest: z.ZodObject<{
404
- id: z.ZodString;
405
- contentHash: z.ZodString;
406
- }, z.core.$strict>;
407
- taskId: z.ZodString;
408
- seed: z.ZodString;
409
- events: z.ZodArray<z.ZodObject<{
410
- sequence: z.ZodNumber;
411
- type: z.ZodEnum<{
412
- reset: "reset";
413
- terminal: "terminal";
414
- created: "created";
415
- action: "action";
416
- observation: "observation";
417
- failure: "failure";
418
- collected: "collected";
419
- destroyed: "destroyed";
420
- }>;
421
- payloadHash: z.ZodString;
422
- metadata: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
423
- }, z.core.$strict>>;
424
- actions: z.ZodArray<z.ZodObject<{
425
- id: z.ZodString;
426
- turn: z.ZodNumber;
427
- kind: z.ZodEnum<{
428
- message: "message";
429
- terminal: "terminal";
430
- tool_call: "tool_call";
431
- }>;
432
- name: z.ZodNullable<z.ZodString>;
433
- arguments: z.ZodRecord<z.ZodString, z.ZodUnknown>;
434
- content: z.ZodNullable<z.ZodString>;
435
- }, z.core.$strict>>;
436
- observations: z.ZodArray<z.ZodObject<{
437
- actionId: z.ZodString;
438
- turn: z.ZodNumber;
439
- terminal: z.ZodBoolean;
440
- output: z.ZodRecord<z.ZodString, z.ZodUnknown>;
441
- artifactRefs: z.ZodArray<z.ZodObject<{
442
- id: z.ZodString;
443
- contentHash: z.ZodString;
444
- mediaType: z.ZodDefault<z.ZodNullable<z.ZodString>>;
445
- sizeBytes: z.ZodDefault<z.ZodNullable<z.ZodNumber>>;
446
- }, z.core.$strict>>;
447
- }, z.core.$strict>>;
448
- terminal: z.ZodBoolean;
449
- failureClass: z.ZodNullable<z.ZodEnum<{
450
- policy_failure: "policy_failure";
451
- grader_failure: "grader_failure";
452
- environment_failure: "environment_failure";
453
- infrastructure_failure: "infrastructure_failure";
454
- timeout: "timeout";
455
- cancelled: "cancelled";
456
- }>>;
457
- output: z.ZodRecord<z.ZodString, z.ZodUnknown>;
458
- contentHash: z.ZodString;
459
- }, z.core.$strict>;
460
- export type AgentSnapshot = z.infer<typeof AgentSnapshotSchema>;
461
- export type HarnessRelease = z.infer<typeof HarnessReleaseSchema>;
462
- export type ModelAction = z.infer<typeof ModelActionSchema>;
463
- export type ToolObservation = z.infer<typeof ToolObservationSchema>;
464
- export type HarnessTrace = z.infer<typeof HarnessTraceSchema>;
465
59
  //# sourceMappingURL=harness.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"harness.d.ts","sourceRoot":"","sources":["../../../../src/harness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAEL,0BAA0B,EAO3B,MAAM,aAAa,CAAC;AAErB,OAAO,EAAwB,KAAK,cAAc,EAAE,KAAK,WAAW,EAAE,MAAM,WAAW,CAAC;AAExF,eAAO,MAAM,uBAAuB;;;;;kBAKzB,CAAC;AAEZ,eAAO,MAAM,0BAA0B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAgB5B,CAAC;AACZ,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAiF,CAAC;AAElH,eAAO,MAAM,uBAAuB;;;;;;;;;;kBAOzB,CAAC;AACZ,eAAO,MAAM,6BAA6B;;;;kBAI/B,CAAC;AAEZ,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAU7B,CAAC;AACZ,eAAO,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAAkF,CAAC;AAEpH,eAAO,MAAM,iBAAiB;;;;;;;;;;;kBAOnB,CAAC;AACZ,eAAO,MAAM,qBAAqB;;;;;;;;;;;kBAMvB,CAAC;AAEZ,MAAM,MAAM,YAAY,GAAG;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,CAAC;AAC7E,MAAM,WAAW,cAAc;IAC7B,MAAM,CAAC,KAAK,EAAE;QAAE,QAAQ,EAAE,WAAW,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,WAAW,CAAA;KAAE,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;IACnH,KAAK,CAAC,KAAK,EAAE,YAAY,EAAE,KAAK,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,WAAW,CAAA;KAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACxF,IAAI,CAAC,KAAK,EAAE,YAAY,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC,eAAe,CAAC,CAAC;IAC9F,OAAO,CAAC,KAAK,EAAE,YAAY,EAAE,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC;QAAE,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC,EAAE,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAC,CAAC;IAC9J,OAAO,CAAC,KAAK,EAAE,YAAY,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7C;AAED,MAAM,MAAM,qBAAqB,GAAG;IAClC,QAAQ,EAAE,WAAW,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,CAAC;AACF,MAAM,MAAM,sBAAsB,GAAG;IACnC,OAAO,EAAE,cAAc,CAAC;IACxB,KAAK,EAAE,OAAO,CAAC;IACf,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC,CAAC;AACF,MAAM,WAAW,eAAe;IAC9B,OAAO,CAAC,KAAK,EAAE,qBAAqB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAAC;CACxE;AACD,MAAM,WAAW,gBAAgB;IAC/B,GAAG,CAAC,KAAK,EAAE;QACT,QAAQ,EAAE,WAAW,CAAC;QACtB,OAAO,EAAE,MAAM,EAAE,CAAC;QAClB,KAAK,EAAE,MAAM,EAAE,CAAC;QAChB,WAAW,EAAE,MAAM,CAAC;QACpB,MAAM,CAAC,EAAE,WAAW,CAAC;KACtB,GAAG,OAAO,CAAC,sBAAsB,EAAE,CAAC,CAAC;CACvC;AAED,wBAAsB,sBAAsB,CAAC,KAAK,EAAE;IAClD,QAAQ,EAAE,WAAW,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,WAAW,EAAE,CAAC;IACvB,OAAO,EAAE,cAAc,CAAC;IACxB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;CACpB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAuGlC;AAED,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC,GAAG,aAAa,CAGpG;AACD,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,2BAA2B,CAAC,GAAG,cAAc,CAGvG;AAED,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;kBAK7B,CAAC;AACZ,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAYpB,CAAC;AAEZ,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC;AAChE,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC;AAClE,MAAM,MAAM,WAAW,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iBAAiB,CAAC,CAAC;AAC5D,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AACpE,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC"}
1
+ {"version":3,"file":"harness.d.ts","sourceRoot":"","sources":["../../../../src/harness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAGL,kBAAkB,EAClB,0BAA0B,EAI1B,KAAK,WAAW,EAChB,KAAK,eAAe,EACrB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,EAEL,KAAK,cAAc,EACnB,KAAK,WAAW,EACjB,MAAM,WAAW,CAAC;AAEnB,cAAc,2BAA2B,CAAC;AAE1C,MAAM,MAAM,YAAY,GAAG;IAAE,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,CAAC;AAE7E,MAAM,WAAW,cAAc;IAC7B,MAAM,CAAC,KAAK,EAAE;QACZ,QAAQ,EAAE,WAAW,CAAC;QACtB,MAAM,EAAE,MAAM,CAAC;QACf,IAAI,EAAE,MAAM,CAAC;QACb,MAAM,EAAE,WAAW,CAAC;KACrB,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;IAC1B,KAAK,CACH,KAAK,EAAE,YAAY,EACnB,KAAK,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,WAAW,CAAA;KAAE,GAC3C,OAAO,CAAC,IAAI,CAAC,CAAC;IACjB,IAAI,CACF,KAAK,EAAE,YAAY,EACnB,MAAM,EAAE,WAAW,EACnB,MAAM,EAAE,WAAW,GAClB,OAAO,CAAC,eAAe,CAAC,CAAC;IAC5B,OAAO,CACL,KAAK,EAAE,YAAY,EACnB,MAAM,EAAE,WAAW,GAClB,OAAO,CAAC;QACT,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,0BAA0B,CAAC,EAAE,CAAC;QAC3D,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KACnC,CAAC,CAAC;IACH,OAAO,CAAC,KAAK,EAAE,YAAY,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CAC7C;AAED,MAAM,MAAM,qBAAqB,GAAG;IAClC,QAAQ,EAAE,WAAW,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,WAAW,CAAC;CACtB,CAAC;AAEF,MAAM,MAAM,sBAAsB,GAAG;IACnC,OAAO,EAAE,cAAc,CAAC;IACxB,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;IAC1C,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC,CAAC;AAEF,MAAM,WAAW,eAAe;IAC9B,OAAO,CAAC,KAAK,EAAE,qBAAqB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAAC;CACxE;AAED,MAAM,WAAW,gBAAgB;IAC/B,GAAG,CAAC,KAAK,EAAE;QACT,QAAQ,EAAE,WAAW,CAAC;QACtB,OAAO,EAAE,MAAM,EAAE,CAAC;QAClB,KAAK,EAAE,MAAM,EAAE,CAAC;QAChB,WAAW,EAAE,MAAM,CAAC;QACpB,MAAM,CAAC,EAAE,WAAW,CAAC;KACtB,GAAG,OAAO,CAAC,sBAAsB,EAAE,CAAC,CAAC;CACvC;AAED,wBAAsB,sBAAsB,CAAC,KAAK,EAAE;IAClD,QAAQ,EAAE,WAAW,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,WAAW,EAAE,CAAC;IACvB,OAAO,EAAE,cAAc,CAAC;IACxB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;CACpB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CA8JlC"}