@wix/evalforge-types 0.60.0 → 0.61.0
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/build/index.js +34 -10
- package/build/index.js.map +2 -2
- package/build/index.mjs +34 -10
- package/build/index.mjs.map +2 -2
- package/build/types/assertion/assertion.d.ts +21 -6
- package/build/types/scenario/test-scenario.d.ts +21 -3
- package/package.json +2 -2
package/build/index.js
CHANGED
|
@@ -831,8 +831,8 @@ var CostConfigSchema = import_zod22.z.strictObject({
|
|
|
831
831
|
var ToolCalledWithParamConfigSchema = import_zod22.z.strictObject({
|
|
832
832
|
/** Name of the tool that must have been called */
|
|
833
833
|
toolName: import_zod22.z.string().min(1),
|
|
834
|
-
/** JSON string of key-value pairs for expected parameters (substring match) */
|
|
835
|
-
expectedParams: import_zod22.z.string().min(1),
|
|
834
|
+
/** JSON string of key-value pairs for expected parameters (substring match). Optional — when omitted, only checks tool presence. */
|
|
835
|
+
expectedParams: import_zod22.z.string().min(1).optional(),
|
|
836
836
|
/** If true, the matching tool call must also have succeeded (step.success === true) */
|
|
837
837
|
requireSuccess: import_zod22.z.boolean().optional()
|
|
838
838
|
});
|
|
@@ -869,23 +869,33 @@ var LlmJudgeConfigSchema = import_zod22.z.object({
|
|
|
869
869
|
/** User-defined parameters for this assertion */
|
|
870
870
|
parameters: import_zod22.z.array(AssertionParameterSchema).optional()
|
|
871
871
|
});
|
|
872
|
+
var AssertionBaseFields = {
|
|
873
|
+
/** When true, the assertion's pass/fail logic is inverted (NOT operator). */
|
|
874
|
+
negate: import_zod22.z.boolean().optional()
|
|
875
|
+
};
|
|
872
876
|
var SkillWasCalledAssertionSchema = SkillWasCalledConfigSchema.extend({
|
|
873
|
-
type: import_zod22.z.literal("skill_was_called")
|
|
877
|
+
type: import_zod22.z.literal("skill_was_called"),
|
|
878
|
+
...AssertionBaseFields
|
|
874
879
|
});
|
|
875
880
|
var ToolCalledWithParamAssertionSchema = ToolCalledWithParamConfigSchema.extend({
|
|
876
|
-
type: import_zod22.z.literal("tool_called_with_param")
|
|
881
|
+
type: import_zod22.z.literal("tool_called_with_param"),
|
|
882
|
+
...AssertionBaseFields
|
|
877
883
|
});
|
|
878
884
|
var BuildPassedAssertionSchema = BuildPassedConfigSchema.extend({
|
|
879
|
-
type: import_zod22.z.literal("build_passed")
|
|
885
|
+
type: import_zod22.z.literal("build_passed"),
|
|
886
|
+
...AssertionBaseFields
|
|
880
887
|
});
|
|
881
888
|
var CostAssertionSchema = CostConfigSchema.extend({
|
|
882
|
-
type: import_zod22.z.literal("cost")
|
|
889
|
+
type: import_zod22.z.literal("cost"),
|
|
890
|
+
...AssertionBaseFields
|
|
883
891
|
});
|
|
884
892
|
var LlmJudgeAssertionSchema = LlmJudgeConfigSchema.extend({
|
|
885
|
-
type: import_zod22.z.literal("llm_judge")
|
|
893
|
+
type: import_zod22.z.literal("llm_judge"),
|
|
894
|
+
...AssertionBaseFields
|
|
886
895
|
});
|
|
887
896
|
var TimeAssertionSchema = TimeConfigSchema.extend({
|
|
888
|
-
type: import_zod22.z.literal("time_limit")
|
|
897
|
+
type: import_zod22.z.literal("time_limit"),
|
|
898
|
+
...AssertionBaseFields
|
|
889
899
|
});
|
|
890
900
|
var AssertionSchema = import_zod22.z.union([
|
|
891
901
|
SkillWasCalledAssertionSchema,
|
|
@@ -901,7 +911,7 @@ var AssertionConfigSchema = import_zod22.z.union([
|
|
|
901
911
|
SkillWasCalledConfigSchema,
|
|
902
912
|
// requires skillNames
|
|
903
913
|
ToolCalledWithParamConfigSchema,
|
|
904
|
-
// requires toolName
|
|
914
|
+
// requires toolName, uses strictObject
|
|
905
915
|
TimeConfigSchema,
|
|
906
916
|
// requires maxDurationMs, uses strictObject
|
|
907
917
|
CostConfigSchema,
|
|
@@ -1613,6 +1623,13 @@ var SYSTEM_ASSERTIONS = {
|
|
|
1613
1623
|
label: "Skills",
|
|
1614
1624
|
type: "string",
|
|
1615
1625
|
required: true
|
|
1626
|
+
},
|
|
1627
|
+
{
|
|
1628
|
+
name: "negate",
|
|
1629
|
+
label: "Negate (NOT operator)",
|
|
1630
|
+
type: "boolean",
|
|
1631
|
+
required: false,
|
|
1632
|
+
defaultValue: false
|
|
1616
1633
|
}
|
|
1617
1634
|
]
|
|
1618
1635
|
},
|
|
@@ -1632,7 +1649,7 @@ var SYSTEM_ASSERTIONS = {
|
|
|
1632
1649
|
name: "expectedParams",
|
|
1633
1650
|
label: "Expected Parameters (JSON, substring match)",
|
|
1634
1651
|
type: "string",
|
|
1635
|
-
required:
|
|
1652
|
+
required: false
|
|
1636
1653
|
},
|
|
1637
1654
|
{
|
|
1638
1655
|
name: "requireSuccess",
|
|
@@ -1641,6 +1658,13 @@ var SYSTEM_ASSERTIONS = {
|
|
|
1641
1658
|
required: false,
|
|
1642
1659
|
defaultValue: false,
|
|
1643
1660
|
advanced: true
|
|
1661
|
+
},
|
|
1662
|
+
{
|
|
1663
|
+
name: "negate",
|
|
1664
|
+
label: "Negate (NOT operator)",
|
|
1665
|
+
type: "boolean",
|
|
1666
|
+
required: false,
|
|
1667
|
+
defaultValue: false
|
|
1644
1668
|
}
|
|
1645
1669
|
]
|
|
1646
1670
|
},
|