@assemble-dev/evals 0.1.0 → 0.2.0

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/dist/index.d.cts CHANGED
@@ -412,6 +412,7 @@ declare const EvalRunFileSchema: z.ZodObject<{
412
412
  agentName: z.ZodString;
413
413
  output: z.ZodType<_assemble_dev_shared_types.JsonValue, unknown, z.core.$ZodTypeInternals<_assemble_dev_shared_types.JsonValue, unknown>>;
414
414
  latencyMs: z.ZodNumber;
415
+ modelName: z.ZodOptional<z.ZodString>;
415
416
  }, z.core.$strip>, z.ZodObject<{
416
417
  id: z.ZodString;
417
418
  runId: z.ZodString;
@@ -701,6 +702,79 @@ declare const EvalRunFileSchema: z.ZodObject<{
701
702
  type: z.ZodLiteral<"replay.completed">;
702
703
  sourceRunId: z.ZodString;
703
704
  latencyMs: z.ZodNumber;
705
+ }, z.core.$strip>, z.ZodObject<{
706
+ id: z.ZodString;
707
+ runId: z.ZodString;
708
+ workflowName: z.ZodString;
709
+ timestamp: z.ZodISODateTime;
710
+ parentEventId: z.ZodNullable<z.ZodString>;
711
+ redacted: z.ZodBoolean;
712
+ metadata: z.ZodOptional<z.ZodObject<{
713
+ sessionId: z.ZodOptional<z.ZodString>;
714
+ userId: z.ZodOptional<z.ZodString>;
715
+ tags: z.ZodOptional<z.ZodArray<z.ZodString>>;
716
+ parentTraceId: z.ZodOptional<z.ZodString>;
717
+ parentSpanId: z.ZodOptional<z.ZodString>;
718
+ }, z.core.$strip>>;
719
+ type: z.ZodLiteral<"batch.submitted">;
720
+ batchId: z.ZodString;
721
+ requestCount: z.ZodNumber;
722
+ models: z.ZodArray<z.ZodString>;
723
+ }, z.core.$strip>, z.ZodObject<{
724
+ id: z.ZodString;
725
+ runId: z.ZodString;
726
+ workflowName: z.ZodString;
727
+ timestamp: z.ZodISODateTime;
728
+ parentEventId: z.ZodNullable<z.ZodString>;
729
+ redacted: z.ZodBoolean;
730
+ metadata: z.ZodOptional<z.ZodObject<{
731
+ sessionId: z.ZodOptional<z.ZodString>;
732
+ userId: z.ZodOptional<z.ZodString>;
733
+ tags: z.ZodOptional<z.ZodArray<z.ZodString>>;
734
+ parentTraceId: z.ZodOptional<z.ZodString>;
735
+ parentSpanId: z.ZodOptional<z.ZodString>;
736
+ }, z.core.$strip>>;
737
+ type: z.ZodLiteral<"batch.completed">;
738
+ batchId: z.ZodString;
739
+ requestCounts: z.ZodObject<{
740
+ processing: z.ZodNumber;
741
+ succeeded: z.ZodNumber;
742
+ errored: z.ZodNumber;
743
+ canceled: z.ZodNumber;
744
+ expired: z.ZodNumber;
745
+ }, z.core.$strip>;
746
+ latencyMs: z.ZodNumber;
747
+ }, z.core.$strip>, z.ZodObject<{
748
+ id: z.ZodString;
749
+ runId: z.ZodString;
750
+ workflowName: z.ZodString;
751
+ timestamp: z.ZodISODateTime;
752
+ parentEventId: z.ZodNullable<z.ZodString>;
753
+ redacted: z.ZodBoolean;
754
+ metadata: z.ZodOptional<z.ZodObject<{
755
+ sessionId: z.ZodOptional<z.ZodString>;
756
+ userId: z.ZodOptional<z.ZodString>;
757
+ tags: z.ZodOptional<z.ZodArray<z.ZodString>>;
758
+ parentTraceId: z.ZodOptional<z.ZodString>;
759
+ parentSpanId: z.ZodOptional<z.ZodString>;
760
+ }, z.core.$strip>>;
761
+ type: z.ZodLiteral<"batch.result">;
762
+ batchId: z.ZodString;
763
+ customId: z.ZodString;
764
+ outcomeType: z.ZodEnum<{
765
+ canceled: "canceled";
766
+ succeeded: "succeeded";
767
+ errored: "errored";
768
+ expired: "expired";
769
+ }>;
770
+ tokenUsage: z.ZodOptional<z.ZodObject<{
771
+ inputTokens: z.ZodNumber;
772
+ outputTokens: z.ZodNumber;
773
+ totalTokens: z.ZodNumber;
774
+ cacheCreationInputTokens: z.ZodOptional<z.ZodNumber>;
775
+ cacheReadInputTokens: z.ZodOptional<z.ZodNumber>;
776
+ }, z.core.$strip>>;
777
+ costUsd: z.ZodNullable<z.ZodNumber>;
704
778
  }, z.core.$strip>], "type">>;
705
779
  }, z.core.$strip>;
706
780
  type EvalCaseRunResult = z.infer<typeof EvalCaseResultSchema>;
package/dist/index.d.ts CHANGED
@@ -412,6 +412,7 @@ declare const EvalRunFileSchema: z.ZodObject<{
412
412
  agentName: z.ZodString;
413
413
  output: z.ZodType<_assemble_dev_shared_types.JsonValue, unknown, z.core.$ZodTypeInternals<_assemble_dev_shared_types.JsonValue, unknown>>;
414
414
  latencyMs: z.ZodNumber;
415
+ modelName: z.ZodOptional<z.ZodString>;
415
416
  }, z.core.$strip>, z.ZodObject<{
416
417
  id: z.ZodString;
417
418
  runId: z.ZodString;
@@ -701,6 +702,79 @@ declare const EvalRunFileSchema: z.ZodObject<{
701
702
  type: z.ZodLiteral<"replay.completed">;
702
703
  sourceRunId: z.ZodString;
703
704
  latencyMs: z.ZodNumber;
705
+ }, z.core.$strip>, z.ZodObject<{
706
+ id: z.ZodString;
707
+ runId: z.ZodString;
708
+ workflowName: z.ZodString;
709
+ timestamp: z.ZodISODateTime;
710
+ parentEventId: z.ZodNullable<z.ZodString>;
711
+ redacted: z.ZodBoolean;
712
+ metadata: z.ZodOptional<z.ZodObject<{
713
+ sessionId: z.ZodOptional<z.ZodString>;
714
+ userId: z.ZodOptional<z.ZodString>;
715
+ tags: z.ZodOptional<z.ZodArray<z.ZodString>>;
716
+ parentTraceId: z.ZodOptional<z.ZodString>;
717
+ parentSpanId: z.ZodOptional<z.ZodString>;
718
+ }, z.core.$strip>>;
719
+ type: z.ZodLiteral<"batch.submitted">;
720
+ batchId: z.ZodString;
721
+ requestCount: z.ZodNumber;
722
+ models: z.ZodArray<z.ZodString>;
723
+ }, z.core.$strip>, z.ZodObject<{
724
+ id: z.ZodString;
725
+ runId: z.ZodString;
726
+ workflowName: z.ZodString;
727
+ timestamp: z.ZodISODateTime;
728
+ parentEventId: z.ZodNullable<z.ZodString>;
729
+ redacted: z.ZodBoolean;
730
+ metadata: z.ZodOptional<z.ZodObject<{
731
+ sessionId: z.ZodOptional<z.ZodString>;
732
+ userId: z.ZodOptional<z.ZodString>;
733
+ tags: z.ZodOptional<z.ZodArray<z.ZodString>>;
734
+ parentTraceId: z.ZodOptional<z.ZodString>;
735
+ parentSpanId: z.ZodOptional<z.ZodString>;
736
+ }, z.core.$strip>>;
737
+ type: z.ZodLiteral<"batch.completed">;
738
+ batchId: z.ZodString;
739
+ requestCounts: z.ZodObject<{
740
+ processing: z.ZodNumber;
741
+ succeeded: z.ZodNumber;
742
+ errored: z.ZodNumber;
743
+ canceled: z.ZodNumber;
744
+ expired: z.ZodNumber;
745
+ }, z.core.$strip>;
746
+ latencyMs: z.ZodNumber;
747
+ }, z.core.$strip>, z.ZodObject<{
748
+ id: z.ZodString;
749
+ runId: z.ZodString;
750
+ workflowName: z.ZodString;
751
+ timestamp: z.ZodISODateTime;
752
+ parentEventId: z.ZodNullable<z.ZodString>;
753
+ redacted: z.ZodBoolean;
754
+ metadata: z.ZodOptional<z.ZodObject<{
755
+ sessionId: z.ZodOptional<z.ZodString>;
756
+ userId: z.ZodOptional<z.ZodString>;
757
+ tags: z.ZodOptional<z.ZodArray<z.ZodString>>;
758
+ parentTraceId: z.ZodOptional<z.ZodString>;
759
+ parentSpanId: z.ZodOptional<z.ZodString>;
760
+ }, z.core.$strip>>;
761
+ type: z.ZodLiteral<"batch.result">;
762
+ batchId: z.ZodString;
763
+ customId: z.ZodString;
764
+ outcomeType: z.ZodEnum<{
765
+ canceled: "canceled";
766
+ succeeded: "succeeded";
767
+ errored: "errored";
768
+ expired: "expired";
769
+ }>;
770
+ tokenUsage: z.ZodOptional<z.ZodObject<{
771
+ inputTokens: z.ZodNumber;
772
+ outputTokens: z.ZodNumber;
773
+ totalTokens: z.ZodNumber;
774
+ cacheCreationInputTokens: z.ZodOptional<z.ZodNumber>;
775
+ cacheReadInputTokens: z.ZodOptional<z.ZodNumber>;
776
+ }, z.core.$strip>>;
777
+ costUsd: z.ZodNullable<z.ZodNumber>;
704
778
  }, z.core.$strip>], "type">>;
705
779
  }, z.core.$strip>;
706
780
  type EvalCaseRunResult = z.infer<typeof EvalCaseResultSchema>;
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@assemble-dev/evals",
3
- "version": "0.1.0",
3
+ "version": "0.2.0",
4
4
  "license": "Elastic-2.0",
5
5
  "description": "Evaluation harness for Assemble agents and workflows",
6
6
  "repository": {
@@ -27,10 +27,10 @@
27
27
  ],
28
28
  "dependencies": {
29
29
  "zod": "^4.1.13",
30
- "@assemble-dev/providers": "0.1.0",
31
- "@assemble-dev/sdk": "0.1.0",
32
- "@assemble-dev/shared-types": "0.1.0",
33
- "@assemble-dev/shared-utils": "0.1.0"
30
+ "@assemble-dev/providers": "0.2.0",
31
+ "@assemble-dev/sdk": "0.2.0",
32
+ "@assemble-dev/shared-types": "0.2.0",
33
+ "@assemble-dev/shared-utils": "0.2.0"
34
34
  },
35
35
  "devDependencies": {
36
36
  "@types/node": "^20.19.0",