@wix/evalforge-types 0.58.0 → 0.60.0
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 +53 -63
- package/build/index.js.map +4 -4
- package/build/index.mjs +49 -57
- package/build/index.mjs.map +4 -4
- package/build/types/assertion/assertion.d.ts +1 -187
- package/build/types/assertion/index.d.ts +1 -1
- package/build/types/assertion/system-assertions.d.ts +1 -1
- package/build/types/evaluation/eval-run-folder.d.ts +43 -0
- package/build/types/evaluation/index.d.ts +1 -0
- package/package.json +2 -2
package/build/index.js
CHANGED
|
@@ -59,7 +59,7 @@ __export(index_exports, {
|
|
|
59
59
|
CostAssertionSchema: () => CostAssertionSchema,
|
|
60
60
|
CostConfigSchema: () => CostConfigSchema,
|
|
61
61
|
CreateAgentInputSchema: () => CreateAgentInputSchema,
|
|
62
|
-
|
|
62
|
+
CreateEvalRunFolderInputSchema: () => CreateEvalRunFolderInputSchema,
|
|
63
63
|
CreateEvalRunInputSchema: () => CreateEvalRunInputSchema,
|
|
64
64
|
CreateEvalScheduleInputSchema: () => CreateEvalScheduleInputSchema,
|
|
65
65
|
CreateMcpInputSchema: () => CreateMcpInputSchema,
|
|
@@ -72,7 +72,6 @@ __export(index_exports, {
|
|
|
72
72
|
CreateTemplateInputSchema: () => CreateTemplateInputSchema,
|
|
73
73
|
CreateTestScenarioInputSchema: () => CreateTestScenarioInputSchema,
|
|
74
74
|
CreateTestSuiteInputSchema: () => CreateTestSuiteInputSchema,
|
|
75
|
-
CustomAssertionSchema: () => CustomAssertionSchema,
|
|
76
75
|
DEFAULT_EVALUATOR_SYSTEM_PROMPT: () => DEFAULT_EVALUATOR_SYSTEM_PROMPT,
|
|
77
76
|
DEFAULT_JUDGE_MODEL: () => DEFAULT_JUDGE_MODEL,
|
|
78
77
|
DiffContentSchema: () => DiffContentSchema,
|
|
@@ -80,6 +79,8 @@ __export(index_exports, {
|
|
|
80
79
|
DiffLineTypeSchema: () => DiffLineTypeSchema,
|
|
81
80
|
EnvironmentSchema: () => EnvironmentSchema,
|
|
82
81
|
EvalMetricsSchema: () => EvalMetricsSchema,
|
|
82
|
+
EvalRunFolderMembershipSchema: () => EvalRunFolderMembershipSchema,
|
|
83
|
+
EvalRunFolderSchema: () => EvalRunFolderSchema,
|
|
83
84
|
EvalRunResultSchema: () => EvalRunResultSchema,
|
|
84
85
|
EvalRunSchema: () => EvalRunSchema,
|
|
85
86
|
EvalScheduleSchema: () => EvalScheduleSchema,
|
|
@@ -172,7 +173,7 @@ __export(index_exports, {
|
|
|
172
173
|
TriggerSchema: () => TriggerSchema,
|
|
173
174
|
TriggerType: () => TriggerType,
|
|
174
175
|
UpdateAgentInputSchema: () => UpdateAgentInputSchema,
|
|
175
|
-
|
|
176
|
+
UpdateEvalRunFolderInputSchema: () => UpdateEvalRunFolderInputSchema,
|
|
176
177
|
UpdateEvalScheduleInputSchema: () => UpdateEvalScheduleInputSchema,
|
|
177
178
|
UpdateMcpInputSchema: () => UpdateMcpInputSchema,
|
|
178
179
|
UpdatePresetInputSchema: () => UpdatePresetInputSchema,
|
|
@@ -185,9 +186,6 @@ __export(index_exports, {
|
|
|
185
186
|
UpdateTestSuiteInputSchema: () => UpdateTestSuiteInputSchema,
|
|
186
187
|
VitestTestSchema: () => VitestTestSchema,
|
|
187
188
|
formatTraceEventLine: () => formatTraceEventLine,
|
|
188
|
-
getBuildPassedConfig: () => getBuildPassedConfig,
|
|
189
|
-
getLlmJudgeConfig: () => getLlmJudgeConfig,
|
|
190
|
-
getSkillWasCalledConfig: () => getSkillWasCalledConfig,
|
|
191
189
|
getSystemAssertion: () => getSystemAssertion,
|
|
192
190
|
getSystemAssertions: () => getSystemAssertions,
|
|
193
191
|
isSystemAssertionId: () => isSystemAssertionId,
|
|
@@ -814,7 +812,7 @@ var AssertionParameterSchema = import_zod22.z.object({
|
|
|
814
812
|
advanced: import_zod22.z.boolean().optional()
|
|
815
813
|
});
|
|
816
814
|
var ScenarioAssertionLinkSchema = import_zod22.z.object({
|
|
817
|
-
/** ID of the
|
|
815
|
+
/** ID of the system assertion (e.g., 'system:skill_was_called') */
|
|
818
816
|
assertionId: import_zod22.z.string(),
|
|
819
817
|
/** Parameter values for this assertion in this scenario */
|
|
820
818
|
params: import_zod22.z.record(
|
|
@@ -913,19 +911,6 @@ var AssertionConfigSchema = import_zod22.z.union([
|
|
|
913
911
|
import_zod22.z.object({})
|
|
914
912
|
// fallback empty config
|
|
915
913
|
]);
|
|
916
|
-
var CustomAssertionSchema = TenantEntitySchema.extend({
|
|
917
|
-
/** The assertion type */
|
|
918
|
-
type: AssertionTypeSchema,
|
|
919
|
-
/** Type-specific configuration */
|
|
920
|
-
config: AssertionConfigSchema
|
|
921
|
-
});
|
|
922
|
-
var CreateCustomAssertionInputSchema = CustomAssertionSchema.omit({
|
|
923
|
-
id: true,
|
|
924
|
-
createdAt: true,
|
|
925
|
-
updatedAt: true,
|
|
926
|
-
deleted: true
|
|
927
|
-
});
|
|
928
|
-
var UpdateCustomAssertionInputSchema = CreateCustomAssertionInputSchema.partial();
|
|
929
914
|
function validateAssertionConfig(type, config) {
|
|
930
915
|
switch (type) {
|
|
931
916
|
case "skill_was_called":
|
|
@@ -944,21 +929,6 @@ function validateAssertionConfig(type, config) {
|
|
|
944
929
|
return false;
|
|
945
930
|
}
|
|
946
931
|
}
|
|
947
|
-
function getSkillWasCalledConfig(assertion) {
|
|
948
|
-
if (assertion.type !== "skill_was_called") return null;
|
|
949
|
-
const result = SkillWasCalledConfigSchema.safeParse(assertion.config);
|
|
950
|
-
return result.success ? result.data : null;
|
|
951
|
-
}
|
|
952
|
-
function getBuildPassedConfig(assertion) {
|
|
953
|
-
if (assertion.type !== "build_passed") return null;
|
|
954
|
-
const result = BuildPassedConfigSchema.safeParse(assertion.config);
|
|
955
|
-
return result.success ? result.data : null;
|
|
956
|
-
}
|
|
957
|
-
function getLlmJudgeConfig(assertion) {
|
|
958
|
-
if (assertion.type !== "llm_judge") return null;
|
|
959
|
-
const result = LlmJudgeConfigSchema.safeParse(assertion.config);
|
|
960
|
-
return result.success ? result.data : null;
|
|
961
|
-
}
|
|
962
932
|
|
|
963
933
|
// src/scenario/test-scenario.ts
|
|
964
934
|
var ExpectedFileSchema = import_zod23.z.object({
|
|
@@ -1482,14 +1452,36 @@ var LeanEvaluationResultSchema = import_zod29.z.object({
|
|
|
1482
1452
|
microcentsSpent: import_zod29.z.number().optional()
|
|
1483
1453
|
});
|
|
1484
1454
|
|
|
1485
|
-
// src/
|
|
1455
|
+
// src/evaluation/eval-run-folder.ts
|
|
1486
1456
|
var import_zod30 = require("zod");
|
|
1457
|
+
var EvalRunFolderSchema = TenantEntitySchema.extend({});
|
|
1458
|
+
var CreateEvalRunFolderInputSchema = EvalRunFolderSchema.omit({
|
|
1459
|
+
id: true,
|
|
1460
|
+
createdAt: true,
|
|
1461
|
+
updatedAt: true,
|
|
1462
|
+
deleted: true
|
|
1463
|
+
});
|
|
1464
|
+
var UpdateEvalRunFolderInputSchema = EvalRunFolderSchema.omit({
|
|
1465
|
+
id: true,
|
|
1466
|
+
createdAt: true,
|
|
1467
|
+
updatedAt: true,
|
|
1468
|
+
deleted: true
|
|
1469
|
+
}).partial();
|
|
1470
|
+
var EvalRunFolderMembershipSchema = import_zod30.z.object({
|
|
1471
|
+
folderId: import_zod30.z.string(),
|
|
1472
|
+
evalRunId: import_zod30.z.string(),
|
|
1473
|
+
projectId: import_zod30.z.string(),
|
|
1474
|
+
createdAt: import_zod30.z.string()
|
|
1475
|
+
});
|
|
1476
|
+
|
|
1477
|
+
// src/project/project.ts
|
|
1478
|
+
var import_zod31 = require("zod");
|
|
1487
1479
|
var ProjectSchema = BaseEntitySchema.extend({
|
|
1488
|
-
appId:
|
|
1489
|
-
appSecret:
|
|
1490
|
-
useWixAuth:
|
|
1491
|
-
useBase44Auth:
|
|
1492
|
-
scenarioTags:
|
|
1480
|
+
appId: import_zod31.z.string().optional().describe("The ID of the app in Dev Center"),
|
|
1481
|
+
appSecret: import_zod31.z.string().optional().describe("The secret of the app in Dev Center"),
|
|
1482
|
+
useWixAuth: import_zod31.z.boolean().optional().describe("Enable Wix CLI/MCP auth for evaluations"),
|
|
1483
|
+
useBase44Auth: import_zod31.z.boolean().optional().describe("Enable Base44 auth for evaluations"),
|
|
1484
|
+
scenarioTags: import_zod31.z.array(import_zod31.z.string()).optional().describe("Project-level tag vocabulary for scenarios")
|
|
1493
1485
|
});
|
|
1494
1486
|
var CreateProjectInputSchema = ProjectSchema.omit({
|
|
1495
1487
|
id: true,
|
|
@@ -1513,7 +1505,7 @@ var CreateTemplateInputSchema = TemplateSchema.omit({
|
|
|
1513
1505
|
var UpdateTemplateInputSchema = CreateTemplateInputSchema.partial();
|
|
1514
1506
|
|
|
1515
1507
|
// src/schedule/eval-schedule.ts
|
|
1516
|
-
var
|
|
1508
|
+
var import_zod32 = require("zod");
|
|
1517
1509
|
var FrequencyType = /* @__PURE__ */ ((FrequencyType2) => {
|
|
1518
1510
|
FrequencyType2["DAILY"] = "daily";
|
|
1519
1511
|
FrequencyType2["WEEKDAY"] = "weekday";
|
|
@@ -1523,29 +1515,29 @@ var FrequencyType = /* @__PURE__ */ ((FrequencyType2) => {
|
|
|
1523
1515
|
})(FrequencyType || {});
|
|
1524
1516
|
var EvalScheduleSchema = TenantEntitySchema.extend({
|
|
1525
1517
|
/** Whether the schedule is active */
|
|
1526
|
-
enabled:
|
|
1518
|
+
enabled: import_zod32.z.boolean(),
|
|
1527
1519
|
/** Test suite to run */
|
|
1528
|
-
suiteId:
|
|
1520
|
+
suiteId: import_zod32.z.string(),
|
|
1529
1521
|
/** Preset that provides agent + entities for this schedule */
|
|
1530
|
-
presetId:
|
|
1522
|
+
presetId: import_zod32.z.string(),
|
|
1531
1523
|
/** How often to run */
|
|
1532
|
-
frequencyType:
|
|
1524
|
+
frequencyType: import_zod32.z.nativeEnum(FrequencyType),
|
|
1533
1525
|
/** Time of day in 24h format (HH:MM), hours 00-23, minutes 00-59 */
|
|
1534
|
-
timeOfDay:
|
|
1526
|
+
timeOfDay: import_zod32.z.string().regex(/^([01]\d|2[0-3]):[0-5]\d$/),
|
|
1535
1527
|
/** Day of week (0=Sun, 6=Sat) for weekly schedules */
|
|
1536
|
-
dayOfWeek:
|
|
1528
|
+
dayOfWeek: import_zod32.z.number().min(0).max(6).optional(),
|
|
1537
1529
|
/** Day of month (1-31) for monthly schedules */
|
|
1538
|
-
dayOfMonth:
|
|
1530
|
+
dayOfMonth: import_zod32.z.number().min(1).max(31).optional(),
|
|
1539
1531
|
/** IANA timezone (e.g., 'America/New_York') */
|
|
1540
|
-
timezone:
|
|
1532
|
+
timezone: import_zod32.z.string(),
|
|
1541
1533
|
/** ID of the last eval run created by this schedule */
|
|
1542
|
-
lastRunId:
|
|
1534
|
+
lastRunId: import_zod32.z.string().optional(),
|
|
1543
1535
|
/** Denormalized status of the last run */
|
|
1544
|
-
lastRunStatus:
|
|
1536
|
+
lastRunStatus: import_zod32.z.string().optional(),
|
|
1545
1537
|
/** ISO timestamp of the last run */
|
|
1546
|
-
lastRunAt:
|
|
1538
|
+
lastRunAt: import_zod32.z.string().optional(),
|
|
1547
1539
|
/** Next scheduled run time in UTC (pre-computed for efficient querying, set by backend) */
|
|
1548
|
-
nextRunAt:
|
|
1540
|
+
nextRunAt: import_zod32.z.string().optional()
|
|
1549
1541
|
});
|
|
1550
1542
|
function isValidTimezone(tz) {
|
|
1551
1543
|
try {
|
|
@@ -1558,14 +1550,14 @@ function isValidTimezone(tz) {
|
|
|
1558
1550
|
function validateScheduleFields(data, ctx, options) {
|
|
1559
1551
|
if (data.frequencyType === "weekly" /* WEEKLY */ && data.dayOfWeek == null) {
|
|
1560
1552
|
ctx.addIssue({
|
|
1561
|
-
code:
|
|
1553
|
+
code: import_zod32.z.ZodIssueCode.custom,
|
|
1562
1554
|
message: "dayOfWeek is required for weekly schedules",
|
|
1563
1555
|
path: ["dayOfWeek"]
|
|
1564
1556
|
});
|
|
1565
1557
|
}
|
|
1566
1558
|
if (data.frequencyType === "monthly" /* MONTHLY */ && data.dayOfMonth == null) {
|
|
1567
1559
|
ctx.addIssue({
|
|
1568
|
-
code:
|
|
1560
|
+
code: import_zod32.z.ZodIssueCode.custom,
|
|
1569
1561
|
message: "dayOfMonth is required for monthly schedules",
|
|
1570
1562
|
path: ["dayOfMonth"]
|
|
1571
1563
|
});
|
|
@@ -1573,7 +1565,7 @@ function validateScheduleFields(data, ctx, options) {
|
|
|
1573
1565
|
const shouldValidateTz = options.partial ? data.timezone !== void 0 : true;
|
|
1574
1566
|
if (shouldValidateTz && !isValidTimezone(data.timezone)) {
|
|
1575
1567
|
ctx.addIssue({
|
|
1576
|
-
code:
|
|
1568
|
+
code: import_zod32.z.ZodIssueCode.custom,
|
|
1577
1569
|
message: "Invalid IANA timezone",
|
|
1578
1570
|
path: ["timezone"]
|
|
1579
1571
|
});
|
|
@@ -1794,7 +1786,7 @@ function getSystemAssertion(id) {
|
|
|
1794
1786
|
CostAssertionSchema,
|
|
1795
1787
|
CostConfigSchema,
|
|
1796
1788
|
CreateAgentInputSchema,
|
|
1797
|
-
|
|
1789
|
+
CreateEvalRunFolderInputSchema,
|
|
1798
1790
|
CreateEvalRunInputSchema,
|
|
1799
1791
|
CreateEvalScheduleInputSchema,
|
|
1800
1792
|
CreateMcpInputSchema,
|
|
@@ -1807,7 +1799,6 @@ function getSystemAssertion(id) {
|
|
|
1807
1799
|
CreateTemplateInputSchema,
|
|
1808
1800
|
CreateTestScenarioInputSchema,
|
|
1809
1801
|
CreateTestSuiteInputSchema,
|
|
1810
|
-
CustomAssertionSchema,
|
|
1811
1802
|
DEFAULT_EVALUATOR_SYSTEM_PROMPT,
|
|
1812
1803
|
DEFAULT_JUDGE_MODEL,
|
|
1813
1804
|
DiffContentSchema,
|
|
@@ -1815,6 +1806,8 @@ function getSystemAssertion(id) {
|
|
|
1815
1806
|
DiffLineTypeSchema,
|
|
1816
1807
|
EnvironmentSchema,
|
|
1817
1808
|
EvalMetricsSchema,
|
|
1809
|
+
EvalRunFolderMembershipSchema,
|
|
1810
|
+
EvalRunFolderSchema,
|
|
1818
1811
|
EvalRunResultSchema,
|
|
1819
1812
|
EvalRunSchema,
|
|
1820
1813
|
EvalScheduleSchema,
|
|
@@ -1907,7 +1900,7 @@ function getSystemAssertion(id) {
|
|
|
1907
1900
|
TriggerSchema,
|
|
1908
1901
|
TriggerType,
|
|
1909
1902
|
UpdateAgentInputSchema,
|
|
1910
|
-
|
|
1903
|
+
UpdateEvalRunFolderInputSchema,
|
|
1911
1904
|
UpdateEvalScheduleInputSchema,
|
|
1912
1905
|
UpdateMcpInputSchema,
|
|
1913
1906
|
UpdatePresetInputSchema,
|
|
@@ -1920,9 +1913,6 @@ function getSystemAssertion(id) {
|
|
|
1920
1913
|
UpdateTestSuiteInputSchema,
|
|
1921
1914
|
VitestTestSchema,
|
|
1922
1915
|
formatTraceEventLine,
|
|
1923
|
-
getBuildPassedConfig,
|
|
1924
|
-
getLlmJudgeConfig,
|
|
1925
|
-
getSkillWasCalledConfig,
|
|
1926
1916
|
getSystemAssertion,
|
|
1927
1917
|
getSystemAssertions,
|
|
1928
1918
|
isSystemAssertionId,
|