@kortyx/agent 0.31.1 → 0.33.0
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/CHANGELOG.md +23 -0
- package/dist/evals.d.ts +199 -16
- package/dist/evals.js +417 -270
- package/dist/evals.js.map +1 -1
- package/dist/index.d.ts +2 -2
- package/dist/index.js +914 -672
- package/dist/index.js.map +1 -1
- package/dist/{studio-judge-CG86eozt.d.ts → studio-judge-DpRLvf8j.d.ts} +700 -57
- package/package.json +4 -4
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,28 @@
|
|
|
1
1
|
# Changelog
|
|
2
2
|
|
|
3
|
+
## [0.33.0](https://github.com/kortyx-io/kortyx/compare/agent-v0.32.0...agent-v0.33.0) (2026-10-08)
|
|
4
|
+
|
|
5
|
+
|
|
6
|
+
### Features
|
|
7
|
+
|
|
8
|
+
* **hooks:** support mixed tools outputs and interrupts ([#297](https://github.com/kortyx-io/kortyx/issues/297)) ([0aa4aa6](https://github.com/kortyx-io/kortyx/commit/0aa4aa6b08cd2ef44f603b8dbe8e35b30c54cbdc))
|
|
9
|
+
|
|
10
|
+
|
|
11
|
+
### Dependencies
|
|
12
|
+
|
|
13
|
+
* The following workspace dependencies were updated
|
|
14
|
+
* dependencies
|
|
15
|
+
* @kortyx/runtime bumped to 0.22.0
|
|
16
|
+
* devDependencies
|
|
17
|
+
* @kortyx/hooks bumped to 0.32.0
|
|
18
|
+
|
|
19
|
+
## [0.32.0](https://github.com/kortyx-io/kortyx/compare/agent-v0.31.1...agent-v0.32.0) (2026-10-08)
|
|
20
|
+
|
|
21
|
+
|
|
22
|
+
### Features
|
|
23
|
+
|
|
24
|
+
* **evals:** compact judge evidence and add suite selection ([#293](https://github.com/kortyx-io/kortyx/issues/293)) ([9754f7c](https://github.com/kortyx-io/kortyx/commit/9754f7cacc61b5945a254ccaa66e17b1b5c8d13f))
|
|
25
|
+
|
|
3
26
|
## [0.31.1](https://github.com/kortyx-io/kortyx/compare/agent-v0.31.0...agent-v0.31.1) (2026-10-08)
|
|
4
27
|
|
|
5
28
|
|
package/dist/evals.d.ts
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
export { C as CreateEvalsOptions, E as EvalCase, a as EvalCaseResult, b as EvalCaseResultSchema, c as EvalCommand, d as EvalConfigurationError, e as EvalContext, f as EvalCriterion, g as EvalCriterionResult, h as EvalDefaults, i as
|
|
1
|
+
export { C as CreateEvalsOptions, E as EvalCase, a as EvalCaseResult, b as EvalCaseResultSchema, c as EvalCommand, d as EvalConfigurationError, e as EvalContext, f as EvalCriterion, g as EvalCriterionResult, h as EvalDefaults, i as EvalEvidenceEvent, j as EvalEvidenceEventType, k as EvalEvidenceFilters, l as EvalEvidencePolicy, m as EvalEvidencePolicySchema, n as EvalExecution, o as EvalExecutionStatus, p as EvalExpectation, q as EvalGradeInput, r as EvalHandlerRef, s as EvalInterrupt, t as EvalIssue, u as EvalJson, v as EvalJudge, w as EvalJudgeEvidence, x as EvalJudgeIdentity, y as EvalJudgeIdentitySchema, z as EvalJudgeOptions, A as EvalJudgeUsage, B as EvalJudgeUsageSchema, D as EvalManifest, F as EvalManifestSchema, G as EvalObservation, H as EvalObservationSchema, I as EvalOutputExpectation, J as EvalOutputSelector, K as EvalPhase, L as EvalProgress, M as EvalProgressSchema, N as EvalRemoteRunRequest, O as EvalRemoteRunRequestSchema, P as EvalResponder, Q as EvalResumeResponseSchema, R as EvalRunOptions, S as EvalRunResult, T as EvalRunResultSchema, U as EvalSetupContext, V as EvalStatus, W as EvalStep, X as EvalStepResult, Y as EvalStepResultSchema, Z as EvalSuite, _ as EvalSuiteSchema, $ as EvalVerdict, a0 as EvalVerdictSchema, a1 as EvalWireEvent, a2 as EvalWireEventSchema, a3 as EvaluationMetadataSchema, a4 as StudioEvalJudgeOptions, a5 as StudioEvalJudgeRequestSchema, a6 as StudioEvalJudgeResponseSchema, a7 as StudioEvalStartRequestSchema, a8 as StudioEvaluationDetail, a9 as StudioEvaluationDetailSchema, aa as StudioEvaluationHistorySchema, ab as StudioEvaluationResultsSchema, ac as StudioEvaluationStartRequest, ad as StudioEvaluationStartRequestSchema, ae as StudioEvaluationSummary, af as StudioEvaluationSummarySchema, ag as createStudioEvalJudge, ah as defineSuite, ai as getEvalGradeEvidence, aj as parseEvalSuite } from './studio-judge-DpRLvf8j.js';
|
|
2
2
|
import { z } from 'zod';
|
|
3
3
|
import '@kortyx/providers';
|
|
4
4
|
import '@kortyx/core/errors';
|
|
@@ -39,6 +39,32 @@ declare const StudioEvalTargetsResponseSchema: z.ZodObject<{
|
|
|
39
39
|
schemaVersion: z.ZodLiteral<1>;
|
|
40
40
|
studioJudging: z.ZodOptional<z.ZodLiteral<true>>;
|
|
41
41
|
suites: z.ZodArray<z.ZodObject<{
|
|
42
|
+
evidence: z.ZodOptional<z.ZodObject<{
|
|
43
|
+
history: z.ZodOptional<z.ZodBoolean>;
|
|
44
|
+
events: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<false>, z.ZodArray<z.ZodEnum<{
|
|
45
|
+
"tool-call-start": "tool-call-start";
|
|
46
|
+
"tool-call-result": "tool-call-result";
|
|
47
|
+
"tool-call-error": "tool-call-error";
|
|
48
|
+
"tool-result": "tool-result";
|
|
49
|
+
"structured-data": "structured-data";
|
|
50
|
+
"structured-data-invalidated": "structured-data-invalidated";
|
|
51
|
+
interrupt: "interrupt";
|
|
52
|
+
error: "error";
|
|
53
|
+
cancelled: "cancelled";
|
|
54
|
+
"limit-reached": "limit-reached";
|
|
55
|
+
}>>, z.ZodObject<{
|
|
56
|
+
using: z.ZodString;
|
|
57
|
+
params: z.ZodOptional<z.ZodJSONSchema>;
|
|
58
|
+
}, z.core.$strict>]>>;
|
|
59
|
+
outputs: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<false>, z.ZodArray<z.ZodObject<{
|
|
60
|
+
dataType: z.ZodOptional<z.ZodString>;
|
|
61
|
+
schemaId: z.ZodOptional<z.ZodString>;
|
|
62
|
+
schemaVersion: z.ZodOptional<z.ZodString>;
|
|
63
|
+
}, z.core.$strict>>, z.ZodObject<{
|
|
64
|
+
using: z.ZodString;
|
|
65
|
+
params: z.ZodOptional<z.ZodJSONSchema>;
|
|
66
|
+
}, z.core.$strict>]>>;
|
|
67
|
+
}, z.core.$strict>>;
|
|
42
68
|
id: z.ZodString;
|
|
43
69
|
name: z.ZodOptional<z.ZodString>;
|
|
44
70
|
cases: z.ZodArray<z.ZodObject<{
|
|
@@ -181,9 +207,9 @@ declare const StudioEvalRunSummarySchema: z.ZodObject<{
|
|
|
181
207
|
suiteId: z.ZodString;
|
|
182
208
|
suiteRevision: z.ZodString;
|
|
183
209
|
status: z.ZodEnum<{
|
|
210
|
+
error: "error";
|
|
184
211
|
cancelled: "cancelled";
|
|
185
212
|
failed: "failed";
|
|
186
|
-
error: "error";
|
|
187
213
|
passed: "passed";
|
|
188
214
|
queued: "queued";
|
|
189
215
|
running: "running";
|
|
@@ -252,9 +278,9 @@ declare const StudioEvalHistorySchema: z.ZodObject<{
|
|
|
252
278
|
suiteId: z.ZodString;
|
|
253
279
|
suiteRevision: z.ZodString;
|
|
254
280
|
status: z.ZodEnum<{
|
|
281
|
+
error: "error";
|
|
255
282
|
cancelled: "cancelled";
|
|
256
283
|
failed: "failed";
|
|
257
|
-
error: "error";
|
|
258
284
|
passed: "passed";
|
|
259
285
|
queued: "queued";
|
|
260
286
|
running: "running";
|
|
@@ -324,9 +350,9 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
324
350
|
suiteId: z.ZodString;
|
|
325
351
|
suiteRevision: z.ZodString;
|
|
326
352
|
status: z.ZodEnum<{
|
|
353
|
+
error: "error";
|
|
327
354
|
cancelled: "cancelled";
|
|
328
355
|
failed: "failed";
|
|
329
|
-
error: "error";
|
|
330
356
|
passed: "passed";
|
|
331
357
|
queued: "queued";
|
|
332
358
|
running: "running";
|
|
@@ -440,6 +466,32 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
440
466
|
concurrency: z.ZodOptional<z.ZodNumber>;
|
|
441
467
|
}, z.core.$strip>>;
|
|
442
468
|
suite: z.ZodObject<{
|
|
469
|
+
evidence: z.ZodOptional<z.ZodObject<{
|
|
470
|
+
history: z.ZodOptional<z.ZodBoolean>;
|
|
471
|
+
events: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<false>, z.ZodArray<z.ZodEnum<{
|
|
472
|
+
"tool-call-start": "tool-call-start";
|
|
473
|
+
"tool-call-result": "tool-call-result";
|
|
474
|
+
"tool-call-error": "tool-call-error";
|
|
475
|
+
"tool-result": "tool-result";
|
|
476
|
+
"structured-data": "structured-data";
|
|
477
|
+
"structured-data-invalidated": "structured-data-invalidated";
|
|
478
|
+
interrupt: "interrupt";
|
|
479
|
+
error: "error";
|
|
480
|
+
cancelled: "cancelled";
|
|
481
|
+
"limit-reached": "limit-reached";
|
|
482
|
+
}>>, z.ZodObject<{
|
|
483
|
+
using: z.ZodString;
|
|
484
|
+
params: z.ZodOptional<z.ZodJSONSchema>;
|
|
485
|
+
}, z.core.$strict>]>>;
|
|
486
|
+
outputs: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<false>, z.ZodArray<z.ZodObject<{
|
|
487
|
+
dataType: z.ZodOptional<z.ZodString>;
|
|
488
|
+
schemaId: z.ZodOptional<z.ZodString>;
|
|
489
|
+
schemaVersion: z.ZodOptional<z.ZodString>;
|
|
490
|
+
}, z.core.$strict>>, z.ZodObject<{
|
|
491
|
+
using: z.ZodString;
|
|
492
|
+
params: z.ZodOptional<z.ZodJSONSchema>;
|
|
493
|
+
}, z.core.$strict>]>>;
|
|
494
|
+
}, z.core.$strict>>;
|
|
443
495
|
id: z.ZodString;
|
|
444
496
|
name: z.ZodOptional<z.ZodString>;
|
|
445
497
|
cases: z.ZodArray<z.ZodObject<{
|
|
@@ -507,6 +559,32 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
507
559
|
suiteId: z.ZodString;
|
|
508
560
|
suiteRevision: z.ZodString;
|
|
509
561
|
suite: z.ZodObject<{
|
|
562
|
+
evidence: z.ZodOptional<z.ZodObject<{
|
|
563
|
+
history: z.ZodOptional<z.ZodBoolean>;
|
|
564
|
+
events: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<false>, z.ZodArray<z.ZodEnum<{
|
|
565
|
+
"tool-call-start": "tool-call-start";
|
|
566
|
+
"tool-call-result": "tool-call-result";
|
|
567
|
+
"tool-call-error": "tool-call-error";
|
|
568
|
+
"tool-result": "tool-result";
|
|
569
|
+
"structured-data": "structured-data";
|
|
570
|
+
"structured-data-invalidated": "structured-data-invalidated";
|
|
571
|
+
interrupt: "interrupt";
|
|
572
|
+
error: "error";
|
|
573
|
+
cancelled: "cancelled";
|
|
574
|
+
"limit-reached": "limit-reached";
|
|
575
|
+
}>>, z.ZodObject<{
|
|
576
|
+
using: z.ZodString;
|
|
577
|
+
params: z.ZodOptional<z.ZodJSONSchema>;
|
|
578
|
+
}, z.core.$strict>]>>;
|
|
579
|
+
outputs: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<false>, z.ZodArray<z.ZodObject<{
|
|
580
|
+
dataType: z.ZodOptional<z.ZodString>;
|
|
581
|
+
schemaId: z.ZodOptional<z.ZodString>;
|
|
582
|
+
schemaVersion: z.ZodOptional<z.ZodString>;
|
|
583
|
+
}, z.core.$strict>>, z.ZodObject<{
|
|
584
|
+
using: z.ZodString;
|
|
585
|
+
params: z.ZodOptional<z.ZodJSONSchema>;
|
|
586
|
+
}, z.core.$strict>]>>;
|
|
587
|
+
}, z.core.$strict>>;
|
|
510
588
|
id: z.ZodString;
|
|
511
589
|
name: z.ZodOptional<z.ZodString>;
|
|
512
590
|
cases: z.ZodArray<z.ZodObject<{
|
|
@@ -580,9 +658,9 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
580
658
|
startedAt: z.ZodISODateTime;
|
|
581
659
|
durationMs: z.ZodNumber;
|
|
582
660
|
status: z.ZodEnum<{
|
|
661
|
+
error: "error";
|
|
583
662
|
cancelled: "cancelled";
|
|
584
663
|
failed: "failed";
|
|
585
|
-
error: "error";
|
|
586
664
|
passed: "passed";
|
|
587
665
|
ungraded: "ungraded";
|
|
588
666
|
}>;
|
|
@@ -598,14 +676,49 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
598
676
|
repetition: z.ZodNumber;
|
|
599
677
|
sessionId: z.ZodString;
|
|
600
678
|
status: z.ZodEnum<{
|
|
679
|
+
error: "error";
|
|
601
680
|
cancelled: "cancelled";
|
|
602
681
|
failed: "failed";
|
|
603
|
-
error: "error";
|
|
604
682
|
passed: "passed";
|
|
605
683
|
ungraded: "ungraded";
|
|
606
684
|
}>;
|
|
607
685
|
durationMs: z.ZodNumber;
|
|
608
686
|
steps: z.ZodArray<z.ZodObject<{
|
|
687
|
+
evidence: z.ZodOptional<z.ZodObject<{
|
|
688
|
+
version: z.ZodLiteral<"compact-v1">;
|
|
689
|
+
history: z.ZodBoolean;
|
|
690
|
+
observation: z.ZodObject<{
|
|
691
|
+
type: z.ZodEnum<{
|
|
692
|
+
interrupt: "interrupt";
|
|
693
|
+
error: "error";
|
|
694
|
+
cancelled: "cancelled";
|
|
695
|
+
answer: "answer";
|
|
696
|
+
}>;
|
|
697
|
+
text: z.ZodString;
|
|
698
|
+
structured: z.ZodArray<z.ZodJSONSchema>;
|
|
699
|
+
events: z.ZodOptional<z.ZodArray<z.ZodJSONSchema>>;
|
|
700
|
+
runId: z.ZodOptional<z.ZodString>;
|
|
701
|
+
checkpointId: z.ZodOptional<z.ZodString>;
|
|
702
|
+
interrupt: z.ZodOptional<z.ZodObject<{
|
|
703
|
+
requestId: z.ZodString;
|
|
704
|
+
kind: z.ZodEnum<{
|
|
705
|
+
custom: "custom";
|
|
706
|
+
text: "text";
|
|
707
|
+
"multi-choice": "multi-choice";
|
|
708
|
+
choice: "choice";
|
|
709
|
+
}>;
|
|
710
|
+
question: z.ZodOptional<z.ZodString>;
|
|
711
|
+
schemaId: z.ZodOptional<z.ZodString>;
|
|
712
|
+
schemaVersion: z.ZodOptional<z.ZodString>;
|
|
713
|
+
options: z.ZodArray<z.ZodObject<{
|
|
714
|
+
id: z.ZodString;
|
|
715
|
+
label: z.ZodString;
|
|
716
|
+
description: z.ZodOptional<z.ZodString>;
|
|
717
|
+
}, z.core.$strict>>;
|
|
718
|
+
request: z.ZodOptional<z.ZodJSONSchema>;
|
|
719
|
+
}, z.core.$strict>>;
|
|
720
|
+
}, z.core.$strict>;
|
|
721
|
+
}, z.core.$strict>>;
|
|
609
722
|
index: z.ZodNumber;
|
|
610
723
|
input: z.ZodUnion<readonly [z.ZodObject<{
|
|
611
724
|
message: z.ZodString;
|
|
@@ -642,9 +755,9 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
642
755
|
}, z.core.$strict>;
|
|
643
756
|
observation: z.ZodObject<{
|
|
644
757
|
type: z.ZodEnum<{
|
|
645
|
-
cancelled: "cancelled";
|
|
646
758
|
interrupt: "interrupt";
|
|
647
759
|
error: "error";
|
|
760
|
+
cancelled: "cancelled";
|
|
648
761
|
answer: "answer";
|
|
649
762
|
}>;
|
|
650
763
|
text: z.ZodString;
|
|
@@ -655,8 +768,8 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
655
768
|
interrupt: z.ZodOptional<z.ZodObject<{
|
|
656
769
|
requestId: z.ZodString;
|
|
657
770
|
kind: z.ZodEnum<{
|
|
658
|
-
text: "text";
|
|
659
771
|
custom: "custom";
|
|
772
|
+
text: "text";
|
|
660
773
|
"multi-choice": "multi-choice";
|
|
661
774
|
choice: "choice";
|
|
662
775
|
}>;
|
|
@@ -673,9 +786,9 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
673
786
|
}, z.core.$strict>;
|
|
674
787
|
reference: z.ZodOptional<z.ZodJSONSchema>;
|
|
675
788
|
status: z.ZodEnum<{
|
|
789
|
+
error: "error";
|
|
676
790
|
cancelled: "cancelled";
|
|
677
791
|
failed: "failed";
|
|
678
|
-
error: "error";
|
|
679
792
|
passed: "passed";
|
|
680
793
|
ungraded: "ungraded";
|
|
681
794
|
}>;
|
|
@@ -778,6 +891,41 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
778
891
|
caseId: z.ZodString;
|
|
779
892
|
repetition: z.ZodNumber;
|
|
780
893
|
step: z.ZodObject<{
|
|
894
|
+
evidence: z.ZodOptional<z.ZodObject<{
|
|
895
|
+
version: z.ZodLiteral<"compact-v1">;
|
|
896
|
+
history: z.ZodBoolean;
|
|
897
|
+
observation: z.ZodObject<{
|
|
898
|
+
type: z.ZodEnum<{
|
|
899
|
+
interrupt: "interrupt";
|
|
900
|
+
error: "error";
|
|
901
|
+
cancelled: "cancelled";
|
|
902
|
+
answer: "answer";
|
|
903
|
+
}>;
|
|
904
|
+
text: z.ZodString;
|
|
905
|
+
structured: z.ZodArray<z.ZodJSONSchema>;
|
|
906
|
+
events: z.ZodOptional<z.ZodArray<z.ZodJSONSchema>>;
|
|
907
|
+
runId: z.ZodOptional<z.ZodString>;
|
|
908
|
+
checkpointId: z.ZodOptional<z.ZodString>;
|
|
909
|
+
interrupt: z.ZodOptional<z.ZodObject<{
|
|
910
|
+
requestId: z.ZodString;
|
|
911
|
+
kind: z.ZodEnum<{
|
|
912
|
+
custom: "custom";
|
|
913
|
+
text: "text";
|
|
914
|
+
"multi-choice": "multi-choice";
|
|
915
|
+
choice: "choice";
|
|
916
|
+
}>;
|
|
917
|
+
question: z.ZodOptional<z.ZodString>;
|
|
918
|
+
schemaId: z.ZodOptional<z.ZodString>;
|
|
919
|
+
schemaVersion: z.ZodOptional<z.ZodString>;
|
|
920
|
+
options: z.ZodArray<z.ZodObject<{
|
|
921
|
+
id: z.ZodString;
|
|
922
|
+
label: z.ZodString;
|
|
923
|
+
description: z.ZodOptional<z.ZodString>;
|
|
924
|
+
}, z.core.$strict>>;
|
|
925
|
+
request: z.ZodOptional<z.ZodJSONSchema>;
|
|
926
|
+
}, z.core.$strict>>;
|
|
927
|
+
}, z.core.$strict>;
|
|
928
|
+
}, z.core.$strict>>;
|
|
781
929
|
index: z.ZodNumber;
|
|
782
930
|
input: z.ZodUnion<readonly [z.ZodObject<{
|
|
783
931
|
message: z.ZodString;
|
|
@@ -814,9 +962,9 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
814
962
|
}, z.core.$strict>;
|
|
815
963
|
observation: z.ZodObject<{
|
|
816
964
|
type: z.ZodEnum<{
|
|
817
|
-
cancelled: "cancelled";
|
|
818
965
|
interrupt: "interrupt";
|
|
819
966
|
error: "error";
|
|
967
|
+
cancelled: "cancelled";
|
|
820
968
|
answer: "answer";
|
|
821
969
|
}>;
|
|
822
970
|
text: z.ZodString;
|
|
@@ -827,8 +975,8 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
827
975
|
interrupt: z.ZodOptional<z.ZodObject<{
|
|
828
976
|
requestId: z.ZodString;
|
|
829
977
|
kind: z.ZodEnum<{
|
|
830
|
-
text: "text";
|
|
831
978
|
custom: "custom";
|
|
979
|
+
text: "text";
|
|
832
980
|
"multi-choice": "multi-choice";
|
|
833
981
|
choice: "choice";
|
|
834
982
|
}>;
|
|
@@ -845,9 +993,9 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
845
993
|
}, z.core.$strict>;
|
|
846
994
|
reference: z.ZodOptional<z.ZodJSONSchema>;
|
|
847
995
|
status: z.ZodEnum<{
|
|
996
|
+
error: "error";
|
|
848
997
|
cancelled: "cancelled";
|
|
849
998
|
failed: "failed";
|
|
850
|
-
error: "error";
|
|
851
999
|
passed: "passed";
|
|
852
1000
|
ungraded: "ungraded";
|
|
853
1001
|
}>;
|
|
@@ -895,14 +1043,49 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
895
1043
|
repetition: z.ZodNumber;
|
|
896
1044
|
sessionId: z.ZodString;
|
|
897
1045
|
status: z.ZodEnum<{
|
|
1046
|
+
error: "error";
|
|
898
1047
|
cancelled: "cancelled";
|
|
899
1048
|
failed: "failed";
|
|
900
|
-
error: "error";
|
|
901
1049
|
passed: "passed";
|
|
902
1050
|
ungraded: "ungraded";
|
|
903
1051
|
}>;
|
|
904
1052
|
durationMs: z.ZodNumber;
|
|
905
1053
|
steps: z.ZodArray<z.ZodObject<{
|
|
1054
|
+
evidence: z.ZodOptional<z.ZodObject<{
|
|
1055
|
+
version: z.ZodLiteral<"compact-v1">;
|
|
1056
|
+
history: z.ZodBoolean;
|
|
1057
|
+
observation: z.ZodObject<{
|
|
1058
|
+
type: z.ZodEnum<{
|
|
1059
|
+
interrupt: "interrupt";
|
|
1060
|
+
error: "error";
|
|
1061
|
+
cancelled: "cancelled";
|
|
1062
|
+
answer: "answer";
|
|
1063
|
+
}>;
|
|
1064
|
+
text: z.ZodString;
|
|
1065
|
+
structured: z.ZodArray<z.ZodJSONSchema>;
|
|
1066
|
+
events: z.ZodOptional<z.ZodArray<z.ZodJSONSchema>>;
|
|
1067
|
+
runId: z.ZodOptional<z.ZodString>;
|
|
1068
|
+
checkpointId: z.ZodOptional<z.ZodString>;
|
|
1069
|
+
interrupt: z.ZodOptional<z.ZodObject<{
|
|
1070
|
+
requestId: z.ZodString;
|
|
1071
|
+
kind: z.ZodEnum<{
|
|
1072
|
+
custom: "custom";
|
|
1073
|
+
text: "text";
|
|
1074
|
+
"multi-choice": "multi-choice";
|
|
1075
|
+
choice: "choice";
|
|
1076
|
+
}>;
|
|
1077
|
+
question: z.ZodOptional<z.ZodString>;
|
|
1078
|
+
schemaId: z.ZodOptional<z.ZodString>;
|
|
1079
|
+
schemaVersion: z.ZodOptional<z.ZodString>;
|
|
1080
|
+
options: z.ZodArray<z.ZodObject<{
|
|
1081
|
+
id: z.ZodString;
|
|
1082
|
+
label: z.ZodString;
|
|
1083
|
+
description: z.ZodOptional<z.ZodString>;
|
|
1084
|
+
}, z.core.$strict>>;
|
|
1085
|
+
request: z.ZodOptional<z.ZodJSONSchema>;
|
|
1086
|
+
}, z.core.$strict>>;
|
|
1087
|
+
}, z.core.$strict>;
|
|
1088
|
+
}, z.core.$strict>>;
|
|
906
1089
|
index: z.ZodNumber;
|
|
907
1090
|
input: z.ZodUnion<readonly [z.ZodObject<{
|
|
908
1091
|
message: z.ZodString;
|
|
@@ -939,9 +1122,9 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
939
1122
|
}, z.core.$strict>;
|
|
940
1123
|
observation: z.ZodObject<{
|
|
941
1124
|
type: z.ZodEnum<{
|
|
942
|
-
cancelled: "cancelled";
|
|
943
1125
|
interrupt: "interrupt";
|
|
944
1126
|
error: "error";
|
|
1127
|
+
cancelled: "cancelled";
|
|
945
1128
|
answer: "answer";
|
|
946
1129
|
}>;
|
|
947
1130
|
text: z.ZodString;
|
|
@@ -952,8 +1135,8 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
952
1135
|
interrupt: z.ZodOptional<z.ZodObject<{
|
|
953
1136
|
requestId: z.ZodString;
|
|
954
1137
|
kind: z.ZodEnum<{
|
|
955
|
-
text: "text";
|
|
956
1138
|
custom: "custom";
|
|
1139
|
+
text: "text";
|
|
957
1140
|
"multi-choice": "multi-choice";
|
|
958
1141
|
choice: "choice";
|
|
959
1142
|
}>;
|
|
@@ -970,9 +1153,9 @@ declare const StudioEvalDetailSchema: z.ZodObject<{
|
|
|
970
1153
|
}, z.core.$strict>;
|
|
971
1154
|
reference: z.ZodOptional<z.ZodJSONSchema>;
|
|
972
1155
|
status: z.ZodEnum<{
|
|
1156
|
+
error: "error";
|
|
973
1157
|
cancelled: "cancelled";
|
|
974
1158
|
failed: "failed";
|
|
975
|
-
error: "error";
|
|
976
1159
|
passed: "passed";
|
|
977
1160
|
ungraded: "ungraded";
|
|
978
1161
|
}>;
|