@ora-ai/cli 0.1.0 → 0.2.1
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.
- package/README.md +75 -0
- package/dist/core/http.js +11 -2
- package/dist/core/http.js.map +1 -1
- package/dist/core/output.js +22 -1
- package/dist/core/output.js.map +1 -1
- package/dist/generated/operations.d.ts +2 -1
- package/dist/generated/operations.d.ts.map +1 -1
- package/dist/generated/operations.js +498 -2
- package/dist/generated/operations.js.map +1 -1
- package/dist/generated/types.d.ts +603 -0
- package/dist/generated/types.d.ts.map +1 -1
- package/dist/generated/types.js +1 -1
- package/dist/watch.d.ts.map +1 -1
- package/dist/watch.js +14 -0
- package/dist/watch.js.map +1 -1
- package/package.json +1 -1
|
@@ -16,6 +16,584 @@ export type AddonsListResponse = {
|
|
|
16
16
|
"harnesses": Array<string>;
|
|
17
17
|
}>;
|
|
18
18
|
};
|
|
19
|
+
export type BenchmarksListResponse = {
|
|
20
|
+
"benchmarks": Array<{
|
|
21
|
+
"id": string;
|
|
22
|
+
"plannedRunCount": number;
|
|
23
|
+
"organizationId": string;
|
|
24
|
+
"createdBy": string | null;
|
|
25
|
+
"name": string;
|
|
26
|
+
"slug": string;
|
|
27
|
+
"description": string | null;
|
|
28
|
+
"skillRef": string | null;
|
|
29
|
+
"skillSlug": string | null;
|
|
30
|
+
"skillName": string | null;
|
|
31
|
+
"skillPromise": string;
|
|
32
|
+
"subjectKind": "skill" | "tools" | "addon" | "custom";
|
|
33
|
+
"subjectName": string | null;
|
|
34
|
+
"subjectLabel": string;
|
|
35
|
+
"armConfig": {
|
|
36
|
+
"baseline"?: {
|
|
37
|
+
"tools"?: "default" | "none" | {
|
|
38
|
+
"tools": Array<string>;
|
|
39
|
+
};
|
|
40
|
+
"addons"?: Record<string, unknown>;
|
|
41
|
+
};
|
|
42
|
+
"treatment"?: {
|
|
43
|
+
"tools"?: "default" | "none" | {
|
|
44
|
+
"tools": Array<string>;
|
|
45
|
+
};
|
|
46
|
+
"addons"?: Record<string, unknown>;
|
|
47
|
+
};
|
|
48
|
+
};
|
|
49
|
+
"comparisonMode": "ab_vs_none" | "ab_custom" | "single";
|
|
50
|
+
"baselineLabel": string | null;
|
|
51
|
+
"treatmentLabel": string | null;
|
|
52
|
+
"promiseMetrics": Array<{
|
|
53
|
+
"metric": "total_tokens" | "output_tokens" | "cost_usd" | "duration_ms" | "num_turns" | "success_rate";
|
|
54
|
+
"direction": "down" | "up";
|
|
55
|
+
"label": string;
|
|
56
|
+
}>;
|
|
57
|
+
"matrix": Array<{
|
|
58
|
+
"harness": string;
|
|
59
|
+
"model": string;
|
|
60
|
+
"provider"?: string;
|
|
61
|
+
}>;
|
|
62
|
+
"taskCount": number;
|
|
63
|
+
"status": string;
|
|
64
|
+
"error": string | null;
|
|
65
|
+
"visibility": "private" | "public";
|
|
66
|
+
"publishedAt": string | null;
|
|
67
|
+
"verdict": unknown;
|
|
68
|
+
"createdAt": string;
|
|
69
|
+
"updatedAt": string;
|
|
70
|
+
}>;
|
|
71
|
+
};
|
|
72
|
+
export type BenchmarksCreateRequest = {
|
|
73
|
+
"name": string;
|
|
74
|
+
"description"?: string;
|
|
75
|
+
"subjectKind"?: "skill" | "tools" | "addon" | "custom";
|
|
76
|
+
"skillRef"?: string;
|
|
77
|
+
"subjectName"?: string;
|
|
78
|
+
"armConfig"?: {
|
|
79
|
+
"baseline"?: {
|
|
80
|
+
"tools"?: "default" | "none" | {
|
|
81
|
+
"tools": Array<string>;
|
|
82
|
+
};
|
|
83
|
+
"addons"?: Record<string, unknown>;
|
|
84
|
+
};
|
|
85
|
+
"treatment"?: {
|
|
86
|
+
"tools"?: "default" | "none" | {
|
|
87
|
+
"tools": Array<string>;
|
|
88
|
+
};
|
|
89
|
+
"addons"?: Record<string, unknown>;
|
|
90
|
+
};
|
|
91
|
+
};
|
|
92
|
+
"comparisonMode"?: "ab_vs_none" | "ab_custom" | "single";
|
|
93
|
+
"baselineLabel"?: string;
|
|
94
|
+
"treatmentLabel"?: string;
|
|
95
|
+
"taskCount"?: number;
|
|
96
|
+
"matrix": Array<{
|
|
97
|
+
"harness": string;
|
|
98
|
+
"model": string;
|
|
99
|
+
"provider"?: string;
|
|
100
|
+
}>;
|
|
101
|
+
"promise"?: string;
|
|
102
|
+
};
|
|
103
|
+
export type BenchmarksCreateResponse = {
|
|
104
|
+
"benchmark": {
|
|
105
|
+
"id": string;
|
|
106
|
+
"plannedRunCount": number;
|
|
107
|
+
"organizationId": string;
|
|
108
|
+
"createdBy": string | null;
|
|
109
|
+
"name": string;
|
|
110
|
+
"slug": string;
|
|
111
|
+
"description": string | null;
|
|
112
|
+
"skillRef": string | null;
|
|
113
|
+
"skillSlug": string | null;
|
|
114
|
+
"skillName": string | null;
|
|
115
|
+
"skillPromise": string;
|
|
116
|
+
"subjectKind": "skill" | "tools" | "addon" | "custom";
|
|
117
|
+
"subjectName": string | null;
|
|
118
|
+
"subjectLabel": string;
|
|
119
|
+
"armConfig": {
|
|
120
|
+
"baseline"?: {
|
|
121
|
+
"tools"?: "default" | "none" | {
|
|
122
|
+
"tools": Array<string>;
|
|
123
|
+
};
|
|
124
|
+
"addons"?: Record<string, unknown>;
|
|
125
|
+
};
|
|
126
|
+
"treatment"?: {
|
|
127
|
+
"tools"?: "default" | "none" | {
|
|
128
|
+
"tools": Array<string>;
|
|
129
|
+
};
|
|
130
|
+
"addons"?: Record<string, unknown>;
|
|
131
|
+
};
|
|
132
|
+
};
|
|
133
|
+
"comparisonMode": "ab_vs_none" | "ab_custom" | "single";
|
|
134
|
+
"baselineLabel": string | null;
|
|
135
|
+
"treatmentLabel": string | null;
|
|
136
|
+
"promiseMetrics": Array<{
|
|
137
|
+
"metric": "total_tokens" | "output_tokens" | "cost_usd" | "duration_ms" | "num_turns" | "success_rate";
|
|
138
|
+
"direction": "down" | "up";
|
|
139
|
+
"label": string;
|
|
140
|
+
}>;
|
|
141
|
+
"matrix": Array<{
|
|
142
|
+
"harness": string;
|
|
143
|
+
"model": string;
|
|
144
|
+
"provider"?: string;
|
|
145
|
+
}>;
|
|
146
|
+
"taskCount": number;
|
|
147
|
+
"status": string;
|
|
148
|
+
"error": string | null;
|
|
149
|
+
"visibility": "private" | "public";
|
|
150
|
+
"publishedAt": string | null;
|
|
151
|
+
"verdict": unknown;
|
|
152
|
+
"createdAt": string;
|
|
153
|
+
"updatedAt": string;
|
|
154
|
+
};
|
|
155
|
+
"streamUrl": string;
|
|
156
|
+
};
|
|
157
|
+
export type BenchmarksGetResponse = {
|
|
158
|
+
"benchmark": {
|
|
159
|
+
"id": string;
|
|
160
|
+
"plannedRunCount": number;
|
|
161
|
+
"organizationId": string;
|
|
162
|
+
"createdBy": string | null;
|
|
163
|
+
"name": string;
|
|
164
|
+
"slug": string;
|
|
165
|
+
"description": string | null;
|
|
166
|
+
"skillRef": string | null;
|
|
167
|
+
"skillSlug": string | null;
|
|
168
|
+
"skillName": string | null;
|
|
169
|
+
"skillPromise": string;
|
|
170
|
+
"subjectKind": "skill" | "tools" | "addon" | "custom";
|
|
171
|
+
"subjectName": string | null;
|
|
172
|
+
"subjectLabel": string;
|
|
173
|
+
"armConfig": {
|
|
174
|
+
"baseline"?: {
|
|
175
|
+
"tools"?: "default" | "none" | {
|
|
176
|
+
"tools": Array<string>;
|
|
177
|
+
};
|
|
178
|
+
"addons"?: Record<string, unknown>;
|
|
179
|
+
};
|
|
180
|
+
"treatment"?: {
|
|
181
|
+
"tools"?: "default" | "none" | {
|
|
182
|
+
"tools": Array<string>;
|
|
183
|
+
};
|
|
184
|
+
"addons"?: Record<string, unknown>;
|
|
185
|
+
};
|
|
186
|
+
};
|
|
187
|
+
"comparisonMode": "ab_vs_none" | "ab_custom" | "single";
|
|
188
|
+
"baselineLabel": string | null;
|
|
189
|
+
"treatmentLabel": string | null;
|
|
190
|
+
"promiseMetrics": Array<{
|
|
191
|
+
"metric": "total_tokens" | "output_tokens" | "cost_usd" | "duration_ms" | "num_turns" | "success_rate";
|
|
192
|
+
"direction": "down" | "up";
|
|
193
|
+
"label": string;
|
|
194
|
+
}>;
|
|
195
|
+
"matrix": Array<{
|
|
196
|
+
"harness": string;
|
|
197
|
+
"model": string;
|
|
198
|
+
"provider"?: string;
|
|
199
|
+
}>;
|
|
200
|
+
"taskCount": number;
|
|
201
|
+
"status": string;
|
|
202
|
+
"error": string | null;
|
|
203
|
+
"visibility": "private" | "public";
|
|
204
|
+
"publishedAt": string | null;
|
|
205
|
+
"verdict": unknown;
|
|
206
|
+
"createdAt": string;
|
|
207
|
+
"updatedAt": string;
|
|
208
|
+
};
|
|
209
|
+
"tasks": Array<{
|
|
210
|
+
"id": string;
|
|
211
|
+
"benchmarkId": string;
|
|
212
|
+
"position": number;
|
|
213
|
+
"prompt": string;
|
|
214
|
+
"rationale": string | null;
|
|
215
|
+
"category": string | null;
|
|
216
|
+
"requiresWeb": boolean;
|
|
217
|
+
"rubric": Array<string>;
|
|
218
|
+
}>;
|
|
219
|
+
"runs": Array<{
|
|
220
|
+
"runId": string;
|
|
221
|
+
"taskId": string;
|
|
222
|
+
"arm": "baseline" | "skill";
|
|
223
|
+
"harness": string;
|
|
224
|
+
"model": string;
|
|
225
|
+
"provider": string | null;
|
|
226
|
+
"status": string;
|
|
227
|
+
"outcome": string | null;
|
|
228
|
+
"outcomeReason": string | null;
|
|
229
|
+
"totalTokens": number | null;
|
|
230
|
+
"inputTokens": number | null;
|
|
231
|
+
"outputTokens": number | null;
|
|
232
|
+
"cacheReadTokens": number | null;
|
|
233
|
+
"cacheCreationTokens": number | null;
|
|
234
|
+
"costUsd": number | null;
|
|
235
|
+
"numTurns": number | null;
|
|
236
|
+
"durationMs": number | null;
|
|
237
|
+
"intentSatisfied": boolean | null;
|
|
238
|
+
"frictionCount": number | null;
|
|
239
|
+
"insightSummary": string | null;
|
|
240
|
+
"qualityScore": number | null;
|
|
241
|
+
"rubricResults": Array<{
|
|
242
|
+
"criterion": string;
|
|
243
|
+
"passed": boolean;
|
|
244
|
+
"note"?: string;
|
|
245
|
+
"votes"?: {
|
|
246
|
+
"passed": number;
|
|
247
|
+
"judges": number;
|
|
248
|
+
};
|
|
249
|
+
}> | null;
|
|
250
|
+
"gradedAt": string | null;
|
|
251
|
+
"judgeModel": string | null;
|
|
252
|
+
"pairVerdict": "skill" | "baseline" | "tie" | null;
|
|
253
|
+
"pairGradedAt": string | null;
|
|
254
|
+
"runCreatedAt": string | null;
|
|
255
|
+
"startedAt": string | null;
|
|
256
|
+
"finishedAt": string | null;
|
|
257
|
+
}>;
|
|
258
|
+
"aggregate": {
|
|
259
|
+
"summary": {
|
|
260
|
+
"verdict": "delivered" | "partial" | "not_delivered" | "inconclusive" | "measured";
|
|
261
|
+
"headline": Array<unknown>;
|
|
262
|
+
"successRate": {
|
|
263
|
+
"baseline": number | null;
|
|
264
|
+
"skill": number | null;
|
|
265
|
+
};
|
|
266
|
+
"pairwise": {
|
|
267
|
+
"pairs": number;
|
|
268
|
+
"skillWins": number;
|
|
269
|
+
"baselineWins": number;
|
|
270
|
+
"ties": number;
|
|
271
|
+
"winRate": number | null;
|
|
272
|
+
};
|
|
273
|
+
"leaderboard"?: Array<unknown>;
|
|
274
|
+
"totals": {
|
|
275
|
+
"cells": number;
|
|
276
|
+
"runs": number;
|
|
277
|
+
"completed": number;
|
|
278
|
+
"failed": number;
|
|
279
|
+
"costUsd": number;
|
|
280
|
+
"tokens": number;
|
|
281
|
+
};
|
|
282
|
+
};
|
|
283
|
+
"perCell": Array<unknown>;
|
|
284
|
+
"perHarness": Array<unknown>;
|
|
285
|
+
"perModel": Array<unknown>;
|
|
286
|
+
"perTask": Array<unknown>;
|
|
287
|
+
"webSplit": unknown;
|
|
288
|
+
};
|
|
289
|
+
};
|
|
290
|
+
export type BenchmarksDeleteResponse = unknown;
|
|
291
|
+
export type BenchmarksGenerateResponse = {
|
|
292
|
+
"benchmark": {
|
|
293
|
+
"id": string;
|
|
294
|
+
"plannedRunCount": number;
|
|
295
|
+
"organizationId": string;
|
|
296
|
+
"createdBy": string | null;
|
|
297
|
+
"name": string;
|
|
298
|
+
"slug": string;
|
|
299
|
+
"description": string | null;
|
|
300
|
+
"skillRef": string | null;
|
|
301
|
+
"skillSlug": string | null;
|
|
302
|
+
"skillName": string | null;
|
|
303
|
+
"skillPromise": string;
|
|
304
|
+
"subjectKind": "skill" | "tools" | "addon" | "custom";
|
|
305
|
+
"subjectName": string | null;
|
|
306
|
+
"subjectLabel": string;
|
|
307
|
+
"armConfig": {
|
|
308
|
+
"baseline"?: {
|
|
309
|
+
"tools"?: "default" | "none" | {
|
|
310
|
+
"tools": Array<string>;
|
|
311
|
+
};
|
|
312
|
+
"addons"?: Record<string, unknown>;
|
|
313
|
+
};
|
|
314
|
+
"treatment"?: {
|
|
315
|
+
"tools"?: "default" | "none" | {
|
|
316
|
+
"tools": Array<string>;
|
|
317
|
+
};
|
|
318
|
+
"addons"?: Record<string, unknown>;
|
|
319
|
+
};
|
|
320
|
+
};
|
|
321
|
+
"comparisonMode": "ab_vs_none" | "ab_custom" | "single";
|
|
322
|
+
"baselineLabel": string | null;
|
|
323
|
+
"treatmentLabel": string | null;
|
|
324
|
+
"promiseMetrics": Array<{
|
|
325
|
+
"metric": "total_tokens" | "output_tokens" | "cost_usd" | "duration_ms" | "num_turns" | "success_rate";
|
|
326
|
+
"direction": "down" | "up";
|
|
327
|
+
"label": string;
|
|
328
|
+
}>;
|
|
329
|
+
"matrix": Array<{
|
|
330
|
+
"harness": string;
|
|
331
|
+
"model": string;
|
|
332
|
+
"provider"?: string;
|
|
333
|
+
}>;
|
|
334
|
+
"taskCount": number;
|
|
335
|
+
"status": string;
|
|
336
|
+
"error": string | null;
|
|
337
|
+
"visibility": "private" | "public";
|
|
338
|
+
"publishedAt": string | null;
|
|
339
|
+
"verdict": unknown;
|
|
340
|
+
"createdAt": string;
|
|
341
|
+
"updatedAt": string;
|
|
342
|
+
};
|
|
343
|
+
"streamUrl": string;
|
|
344
|
+
};
|
|
345
|
+
export type BenchmarksLaunchResponse = {
|
|
346
|
+
"benchmark": {
|
|
347
|
+
"id": string;
|
|
348
|
+
"plannedRunCount": number;
|
|
349
|
+
"organizationId": string;
|
|
350
|
+
"createdBy": string | null;
|
|
351
|
+
"name": string;
|
|
352
|
+
"slug": string;
|
|
353
|
+
"description": string | null;
|
|
354
|
+
"skillRef": string | null;
|
|
355
|
+
"skillSlug": string | null;
|
|
356
|
+
"skillName": string | null;
|
|
357
|
+
"skillPromise": string;
|
|
358
|
+
"subjectKind": "skill" | "tools" | "addon" | "custom";
|
|
359
|
+
"subjectName": string | null;
|
|
360
|
+
"subjectLabel": string;
|
|
361
|
+
"armConfig": {
|
|
362
|
+
"baseline"?: {
|
|
363
|
+
"tools"?: "default" | "none" | {
|
|
364
|
+
"tools": Array<string>;
|
|
365
|
+
};
|
|
366
|
+
"addons"?: Record<string, unknown>;
|
|
367
|
+
};
|
|
368
|
+
"treatment"?: {
|
|
369
|
+
"tools"?: "default" | "none" | {
|
|
370
|
+
"tools": Array<string>;
|
|
371
|
+
};
|
|
372
|
+
"addons"?: Record<string, unknown>;
|
|
373
|
+
};
|
|
374
|
+
};
|
|
375
|
+
"comparisonMode": "ab_vs_none" | "ab_custom" | "single";
|
|
376
|
+
"baselineLabel": string | null;
|
|
377
|
+
"treatmentLabel": string | null;
|
|
378
|
+
"promiseMetrics": Array<{
|
|
379
|
+
"metric": "total_tokens" | "output_tokens" | "cost_usd" | "duration_ms" | "num_turns" | "success_rate";
|
|
380
|
+
"direction": "down" | "up";
|
|
381
|
+
"label": string;
|
|
382
|
+
}>;
|
|
383
|
+
"matrix": Array<{
|
|
384
|
+
"harness": string;
|
|
385
|
+
"model": string;
|
|
386
|
+
"provider"?: string;
|
|
387
|
+
}>;
|
|
388
|
+
"taskCount": number;
|
|
389
|
+
"status": string;
|
|
390
|
+
"error": string | null;
|
|
391
|
+
"visibility": "private" | "public";
|
|
392
|
+
"publishedAt": string | null;
|
|
393
|
+
"verdict": unknown;
|
|
394
|
+
"createdAt": string;
|
|
395
|
+
"updatedAt": string;
|
|
396
|
+
};
|
|
397
|
+
"streamUrl": string;
|
|
398
|
+
};
|
|
399
|
+
export type BenchmarksPublishRequest = {
|
|
400
|
+
"slug"?: string;
|
|
401
|
+
};
|
|
402
|
+
export type BenchmarksPublishResponse = {
|
|
403
|
+
"benchmark": {
|
|
404
|
+
"id": string;
|
|
405
|
+
"plannedRunCount": number;
|
|
406
|
+
"organizationId": string;
|
|
407
|
+
"createdBy": string | null;
|
|
408
|
+
"name": string;
|
|
409
|
+
"slug": string;
|
|
410
|
+
"description": string | null;
|
|
411
|
+
"skillRef": string | null;
|
|
412
|
+
"skillSlug": string | null;
|
|
413
|
+
"skillName": string | null;
|
|
414
|
+
"skillPromise": string;
|
|
415
|
+
"subjectKind": "skill" | "tools" | "addon" | "custom";
|
|
416
|
+
"subjectName": string | null;
|
|
417
|
+
"subjectLabel": string;
|
|
418
|
+
"armConfig": {
|
|
419
|
+
"baseline"?: {
|
|
420
|
+
"tools"?: "default" | "none" | {
|
|
421
|
+
"tools": Array<string>;
|
|
422
|
+
};
|
|
423
|
+
"addons"?: Record<string, unknown>;
|
|
424
|
+
};
|
|
425
|
+
"treatment"?: {
|
|
426
|
+
"tools"?: "default" | "none" | {
|
|
427
|
+
"tools": Array<string>;
|
|
428
|
+
};
|
|
429
|
+
"addons"?: Record<string, unknown>;
|
|
430
|
+
};
|
|
431
|
+
};
|
|
432
|
+
"comparisonMode": "ab_vs_none" | "ab_custom" | "single";
|
|
433
|
+
"baselineLabel": string | null;
|
|
434
|
+
"treatmentLabel": string | null;
|
|
435
|
+
"promiseMetrics": Array<{
|
|
436
|
+
"metric": "total_tokens" | "output_tokens" | "cost_usd" | "duration_ms" | "num_turns" | "success_rate";
|
|
437
|
+
"direction": "down" | "up";
|
|
438
|
+
"label": string;
|
|
439
|
+
}>;
|
|
440
|
+
"matrix": Array<{
|
|
441
|
+
"harness": string;
|
|
442
|
+
"model": string;
|
|
443
|
+
"provider"?: string;
|
|
444
|
+
}>;
|
|
445
|
+
"taskCount": number;
|
|
446
|
+
"status": string;
|
|
447
|
+
"error": string | null;
|
|
448
|
+
"visibility": "private" | "public";
|
|
449
|
+
"publishedAt": string | null;
|
|
450
|
+
"verdict": unknown;
|
|
451
|
+
"createdAt": string;
|
|
452
|
+
"updatedAt": string;
|
|
453
|
+
};
|
|
454
|
+
};
|
|
455
|
+
export type BenchmarksRegradeResponse = {
|
|
456
|
+
"benchmark": {
|
|
457
|
+
"id": string;
|
|
458
|
+
"plannedRunCount": number;
|
|
459
|
+
"organizationId": string;
|
|
460
|
+
"createdBy": string | null;
|
|
461
|
+
"name": string;
|
|
462
|
+
"slug": string;
|
|
463
|
+
"description": string | null;
|
|
464
|
+
"skillRef": string | null;
|
|
465
|
+
"skillSlug": string | null;
|
|
466
|
+
"skillName": string | null;
|
|
467
|
+
"skillPromise": string;
|
|
468
|
+
"subjectKind": "skill" | "tools" | "addon" | "custom";
|
|
469
|
+
"subjectName": string | null;
|
|
470
|
+
"subjectLabel": string;
|
|
471
|
+
"armConfig": {
|
|
472
|
+
"baseline"?: {
|
|
473
|
+
"tools"?: "default" | "none" | {
|
|
474
|
+
"tools": Array<string>;
|
|
475
|
+
};
|
|
476
|
+
"addons"?: Record<string, unknown>;
|
|
477
|
+
};
|
|
478
|
+
"treatment"?: {
|
|
479
|
+
"tools"?: "default" | "none" | {
|
|
480
|
+
"tools": Array<string>;
|
|
481
|
+
};
|
|
482
|
+
"addons"?: Record<string, unknown>;
|
|
483
|
+
};
|
|
484
|
+
};
|
|
485
|
+
"comparisonMode": "ab_vs_none" | "ab_custom" | "single";
|
|
486
|
+
"baselineLabel": string | null;
|
|
487
|
+
"treatmentLabel": string | null;
|
|
488
|
+
"promiseMetrics": Array<{
|
|
489
|
+
"metric": "total_tokens" | "output_tokens" | "cost_usd" | "duration_ms" | "num_turns" | "success_rate";
|
|
490
|
+
"direction": "down" | "up";
|
|
491
|
+
"label": string;
|
|
492
|
+
}>;
|
|
493
|
+
"matrix": Array<{
|
|
494
|
+
"harness": string;
|
|
495
|
+
"model": string;
|
|
496
|
+
"provider"?: string;
|
|
497
|
+
}>;
|
|
498
|
+
"taskCount": number;
|
|
499
|
+
"status": string;
|
|
500
|
+
"error": string | null;
|
|
501
|
+
"visibility": "private" | "public";
|
|
502
|
+
"publishedAt": string | null;
|
|
503
|
+
"verdict": unknown;
|
|
504
|
+
"createdAt": string;
|
|
505
|
+
"updatedAt": string;
|
|
506
|
+
};
|
|
507
|
+
"streamUrl": string;
|
|
508
|
+
};
|
|
509
|
+
export type BenchmarksStreamResponse = unknown;
|
|
510
|
+
export type BenchmarksUpdateTaskRequest = {
|
|
511
|
+
"prompt": string;
|
|
512
|
+
};
|
|
513
|
+
export type BenchmarksUpdateTaskResponse = {
|
|
514
|
+
"task": {
|
|
515
|
+
"id": string;
|
|
516
|
+
"benchmarkId": string;
|
|
517
|
+
"position": number;
|
|
518
|
+
"prompt": string;
|
|
519
|
+
"rationale": string | null;
|
|
520
|
+
"category": string | null;
|
|
521
|
+
"requiresWeb": boolean;
|
|
522
|
+
"rubric": Array<string>;
|
|
523
|
+
};
|
|
524
|
+
};
|
|
525
|
+
export type BenchmarksUnpublishResponse = {
|
|
526
|
+
"benchmark": {
|
|
527
|
+
"id": string;
|
|
528
|
+
"plannedRunCount": number;
|
|
529
|
+
"organizationId": string;
|
|
530
|
+
"createdBy": string | null;
|
|
531
|
+
"name": string;
|
|
532
|
+
"slug": string;
|
|
533
|
+
"description": string | null;
|
|
534
|
+
"skillRef": string | null;
|
|
535
|
+
"skillSlug": string | null;
|
|
536
|
+
"skillName": string | null;
|
|
537
|
+
"skillPromise": string;
|
|
538
|
+
"subjectKind": "skill" | "tools" | "addon" | "custom";
|
|
539
|
+
"subjectName": string | null;
|
|
540
|
+
"subjectLabel": string;
|
|
541
|
+
"armConfig": {
|
|
542
|
+
"baseline"?: {
|
|
543
|
+
"tools"?: "default" | "none" | {
|
|
544
|
+
"tools": Array<string>;
|
|
545
|
+
};
|
|
546
|
+
"addons"?: Record<string, unknown>;
|
|
547
|
+
};
|
|
548
|
+
"treatment"?: {
|
|
549
|
+
"tools"?: "default" | "none" | {
|
|
550
|
+
"tools": Array<string>;
|
|
551
|
+
};
|
|
552
|
+
"addons"?: Record<string, unknown>;
|
|
553
|
+
};
|
|
554
|
+
};
|
|
555
|
+
"comparisonMode": "ab_vs_none" | "ab_custom" | "single";
|
|
556
|
+
"baselineLabel": string | null;
|
|
557
|
+
"treatmentLabel": string | null;
|
|
558
|
+
"promiseMetrics": Array<{
|
|
559
|
+
"metric": "total_tokens" | "output_tokens" | "cost_usd" | "duration_ms" | "num_turns" | "success_rate";
|
|
560
|
+
"direction": "down" | "up";
|
|
561
|
+
"label": string;
|
|
562
|
+
}>;
|
|
563
|
+
"matrix": Array<{
|
|
564
|
+
"harness": string;
|
|
565
|
+
"model": string;
|
|
566
|
+
"provider"?: string;
|
|
567
|
+
}>;
|
|
568
|
+
"taskCount": number;
|
|
569
|
+
"status": string;
|
|
570
|
+
"error": string | null;
|
|
571
|
+
"visibility": "private" | "public";
|
|
572
|
+
"publishedAt": string | null;
|
|
573
|
+
"verdict": unknown;
|
|
574
|
+
"createdAt": string;
|
|
575
|
+
"updatedAt": string;
|
|
576
|
+
};
|
|
577
|
+
};
|
|
578
|
+
export type ExperimentCatalogResponse = {
|
|
579
|
+
"harnesses": Array<{
|
|
580
|
+
"id": string;
|
|
581
|
+
"name": string;
|
|
582
|
+
"maker": string;
|
|
583
|
+
"offerable": boolean;
|
|
584
|
+
"reason": string | null;
|
|
585
|
+
"modes": Array<"task" | "session">;
|
|
586
|
+
"defaultProvider": string | null;
|
|
587
|
+
"providers": Array<{
|
|
588
|
+
"id": string;
|
|
589
|
+
"models": Array<{
|
|
590
|
+
"id": string;
|
|
591
|
+
"name": string;
|
|
592
|
+
}>;
|
|
593
|
+
"defaultModel": string | null;
|
|
594
|
+
}>;
|
|
595
|
+
}>;
|
|
596
|
+
};
|
|
19
597
|
export type ExperimentRunsListResponse = {
|
|
20
598
|
"results": Array<{
|
|
21
599
|
"id": string;
|
|
@@ -101,6 +679,31 @@ export type ExperimentRunsInsightResponse = unknown;
|
|
|
101
679
|
export type ExperimentRunsRawTraceResponse = unknown;
|
|
102
680
|
export type ExperimentRunsStreamResponse = unknown;
|
|
103
681
|
export type ExperimentRunsTrajectoryResponse = unknown;
|
|
682
|
+
export type ExperimentSecretsListResponse = {
|
|
683
|
+
"secrets": Array<{
|
|
684
|
+
"id": string;
|
|
685
|
+
"organizationId": string;
|
|
686
|
+
"name": string;
|
|
687
|
+
"createdBy": string | null;
|
|
688
|
+
"createdAt": string;
|
|
689
|
+
"updatedAt": string;
|
|
690
|
+
}>;
|
|
691
|
+
};
|
|
692
|
+
export type ExperimentSecretsCreateRequest = {
|
|
693
|
+
"name": string;
|
|
694
|
+
"value": string;
|
|
695
|
+
};
|
|
696
|
+
export type ExperimentSecretsCreateResponse = {
|
|
697
|
+
"secret": {
|
|
698
|
+
"id": string;
|
|
699
|
+
"organizationId": string;
|
|
700
|
+
"name": string;
|
|
701
|
+
"createdBy": string | null;
|
|
702
|
+
"createdAt": string;
|
|
703
|
+
"updatedAt": string;
|
|
704
|
+
};
|
|
705
|
+
};
|
|
706
|
+
export type ExperimentSecretsDeleteResponse = unknown;
|
|
104
707
|
export type SessionsListResponse = {
|
|
105
708
|
"sessions": Array<{
|
|
106
709
|
"id": string;
|