@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.mjs
CHANGED
|
@@ -614,7 +614,7 @@ var AssertionParameterSchema = z22.object({
|
|
|
614
614
|
advanced: z22.boolean().optional()
|
|
615
615
|
});
|
|
616
616
|
var ScenarioAssertionLinkSchema = z22.object({
|
|
617
|
-
/** ID of the
|
|
617
|
+
/** ID of the system assertion (e.g., 'system:skill_was_called') */
|
|
618
618
|
assertionId: z22.string(),
|
|
619
619
|
/** Parameter values for this assertion in this scenario */
|
|
620
620
|
params: z22.record(
|
|
@@ -713,19 +713,6 @@ var AssertionConfigSchema = z22.union([
|
|
|
713
713
|
z22.object({})
|
|
714
714
|
// fallback empty config
|
|
715
715
|
]);
|
|
716
|
-
var CustomAssertionSchema = TenantEntitySchema.extend({
|
|
717
|
-
/** The assertion type */
|
|
718
|
-
type: AssertionTypeSchema,
|
|
719
|
-
/** Type-specific configuration */
|
|
720
|
-
config: AssertionConfigSchema
|
|
721
|
-
});
|
|
722
|
-
var CreateCustomAssertionInputSchema = CustomAssertionSchema.omit({
|
|
723
|
-
id: true,
|
|
724
|
-
createdAt: true,
|
|
725
|
-
updatedAt: true,
|
|
726
|
-
deleted: true
|
|
727
|
-
});
|
|
728
|
-
var UpdateCustomAssertionInputSchema = CreateCustomAssertionInputSchema.partial();
|
|
729
716
|
function validateAssertionConfig(type, config) {
|
|
730
717
|
switch (type) {
|
|
731
718
|
case "skill_was_called":
|
|
@@ -744,21 +731,6 @@ function validateAssertionConfig(type, config) {
|
|
|
744
731
|
return false;
|
|
745
732
|
}
|
|
746
733
|
}
|
|
747
|
-
function getSkillWasCalledConfig(assertion) {
|
|
748
|
-
if (assertion.type !== "skill_was_called") return null;
|
|
749
|
-
const result = SkillWasCalledConfigSchema.safeParse(assertion.config);
|
|
750
|
-
return result.success ? result.data : null;
|
|
751
|
-
}
|
|
752
|
-
function getBuildPassedConfig(assertion) {
|
|
753
|
-
if (assertion.type !== "build_passed") return null;
|
|
754
|
-
const result = BuildPassedConfigSchema.safeParse(assertion.config);
|
|
755
|
-
return result.success ? result.data : null;
|
|
756
|
-
}
|
|
757
|
-
function getLlmJudgeConfig(assertion) {
|
|
758
|
-
if (assertion.type !== "llm_judge") return null;
|
|
759
|
-
const result = LlmJudgeConfigSchema.safeParse(assertion.config);
|
|
760
|
-
return result.success ? result.data : null;
|
|
761
|
-
}
|
|
762
734
|
|
|
763
735
|
// src/scenario/test-scenario.ts
|
|
764
736
|
var ExpectedFileSchema = z23.object({
|
|
@@ -1282,14 +1254,36 @@ var LeanEvaluationResultSchema = z29.object({
|
|
|
1282
1254
|
microcentsSpent: z29.number().optional()
|
|
1283
1255
|
});
|
|
1284
1256
|
|
|
1285
|
-
// src/
|
|
1257
|
+
// src/evaluation/eval-run-folder.ts
|
|
1286
1258
|
import { z as z30 } from "zod";
|
|
1259
|
+
var EvalRunFolderSchema = TenantEntitySchema.extend({});
|
|
1260
|
+
var CreateEvalRunFolderInputSchema = EvalRunFolderSchema.omit({
|
|
1261
|
+
id: true,
|
|
1262
|
+
createdAt: true,
|
|
1263
|
+
updatedAt: true,
|
|
1264
|
+
deleted: true
|
|
1265
|
+
});
|
|
1266
|
+
var UpdateEvalRunFolderInputSchema = EvalRunFolderSchema.omit({
|
|
1267
|
+
id: true,
|
|
1268
|
+
createdAt: true,
|
|
1269
|
+
updatedAt: true,
|
|
1270
|
+
deleted: true
|
|
1271
|
+
}).partial();
|
|
1272
|
+
var EvalRunFolderMembershipSchema = z30.object({
|
|
1273
|
+
folderId: z30.string(),
|
|
1274
|
+
evalRunId: z30.string(),
|
|
1275
|
+
projectId: z30.string(),
|
|
1276
|
+
createdAt: z30.string()
|
|
1277
|
+
});
|
|
1278
|
+
|
|
1279
|
+
// src/project/project.ts
|
|
1280
|
+
import { z as z31 } from "zod";
|
|
1287
1281
|
var ProjectSchema = BaseEntitySchema.extend({
|
|
1288
|
-
appId:
|
|
1289
|
-
appSecret:
|
|
1290
|
-
useWixAuth:
|
|
1291
|
-
useBase44Auth:
|
|
1292
|
-
scenarioTags:
|
|
1282
|
+
appId: z31.string().optional().describe("The ID of the app in Dev Center"),
|
|
1283
|
+
appSecret: z31.string().optional().describe("The secret of the app in Dev Center"),
|
|
1284
|
+
useWixAuth: z31.boolean().optional().describe("Enable Wix CLI/MCP auth for evaluations"),
|
|
1285
|
+
useBase44Auth: z31.boolean().optional().describe("Enable Base44 auth for evaluations"),
|
|
1286
|
+
scenarioTags: z31.array(z31.string()).optional().describe("Project-level tag vocabulary for scenarios")
|
|
1293
1287
|
});
|
|
1294
1288
|
var CreateProjectInputSchema = ProjectSchema.omit({
|
|
1295
1289
|
id: true,
|
|
@@ -1313,7 +1307,7 @@ var CreateTemplateInputSchema = TemplateSchema.omit({
|
|
|
1313
1307
|
var UpdateTemplateInputSchema = CreateTemplateInputSchema.partial();
|
|
1314
1308
|
|
|
1315
1309
|
// src/schedule/eval-schedule.ts
|
|
1316
|
-
import { z as
|
|
1310
|
+
import { z as z32 } from "zod";
|
|
1317
1311
|
var FrequencyType = /* @__PURE__ */ ((FrequencyType2) => {
|
|
1318
1312
|
FrequencyType2["DAILY"] = "daily";
|
|
1319
1313
|
FrequencyType2["WEEKDAY"] = "weekday";
|
|
@@ -1323,29 +1317,29 @@ var FrequencyType = /* @__PURE__ */ ((FrequencyType2) => {
|
|
|
1323
1317
|
})(FrequencyType || {});
|
|
1324
1318
|
var EvalScheduleSchema = TenantEntitySchema.extend({
|
|
1325
1319
|
/** Whether the schedule is active */
|
|
1326
|
-
enabled:
|
|
1320
|
+
enabled: z32.boolean(),
|
|
1327
1321
|
/** Test suite to run */
|
|
1328
|
-
suiteId:
|
|
1322
|
+
suiteId: z32.string(),
|
|
1329
1323
|
/** Preset that provides agent + entities for this schedule */
|
|
1330
|
-
presetId:
|
|
1324
|
+
presetId: z32.string(),
|
|
1331
1325
|
/** How often to run */
|
|
1332
|
-
frequencyType:
|
|
1326
|
+
frequencyType: z32.nativeEnum(FrequencyType),
|
|
1333
1327
|
/** Time of day in 24h format (HH:MM), hours 00-23, minutes 00-59 */
|
|
1334
|
-
timeOfDay:
|
|
1328
|
+
timeOfDay: z32.string().regex(/^([01]\d|2[0-3]):[0-5]\d$/),
|
|
1335
1329
|
/** Day of week (0=Sun, 6=Sat) for weekly schedules */
|
|
1336
|
-
dayOfWeek:
|
|
1330
|
+
dayOfWeek: z32.number().min(0).max(6).optional(),
|
|
1337
1331
|
/** Day of month (1-31) for monthly schedules */
|
|
1338
|
-
dayOfMonth:
|
|
1332
|
+
dayOfMonth: z32.number().min(1).max(31).optional(),
|
|
1339
1333
|
/** IANA timezone (e.g., 'America/New_York') */
|
|
1340
|
-
timezone:
|
|
1334
|
+
timezone: z32.string(),
|
|
1341
1335
|
/** ID of the last eval run created by this schedule */
|
|
1342
|
-
lastRunId:
|
|
1336
|
+
lastRunId: z32.string().optional(),
|
|
1343
1337
|
/** Denormalized status of the last run */
|
|
1344
|
-
lastRunStatus:
|
|
1338
|
+
lastRunStatus: z32.string().optional(),
|
|
1345
1339
|
/** ISO timestamp of the last run */
|
|
1346
|
-
lastRunAt:
|
|
1340
|
+
lastRunAt: z32.string().optional(),
|
|
1347
1341
|
/** Next scheduled run time in UTC (pre-computed for efficient querying, set by backend) */
|
|
1348
|
-
nextRunAt:
|
|
1342
|
+
nextRunAt: z32.string().optional()
|
|
1349
1343
|
});
|
|
1350
1344
|
function isValidTimezone(tz) {
|
|
1351
1345
|
try {
|
|
@@ -1358,14 +1352,14 @@ function isValidTimezone(tz) {
|
|
|
1358
1352
|
function validateScheduleFields(data, ctx, options) {
|
|
1359
1353
|
if (data.frequencyType === "weekly" /* WEEKLY */ && data.dayOfWeek == null) {
|
|
1360
1354
|
ctx.addIssue({
|
|
1361
|
-
code:
|
|
1355
|
+
code: z32.ZodIssueCode.custom,
|
|
1362
1356
|
message: "dayOfWeek is required for weekly schedules",
|
|
1363
1357
|
path: ["dayOfWeek"]
|
|
1364
1358
|
});
|
|
1365
1359
|
}
|
|
1366
1360
|
if (data.frequencyType === "monthly" /* MONTHLY */ && data.dayOfMonth == null) {
|
|
1367
1361
|
ctx.addIssue({
|
|
1368
|
-
code:
|
|
1362
|
+
code: z32.ZodIssueCode.custom,
|
|
1369
1363
|
message: "dayOfMonth is required for monthly schedules",
|
|
1370
1364
|
path: ["dayOfMonth"]
|
|
1371
1365
|
});
|
|
@@ -1373,7 +1367,7 @@ function validateScheduleFields(data, ctx, options) {
|
|
|
1373
1367
|
const shouldValidateTz = options.partial ? data.timezone !== void 0 : true;
|
|
1374
1368
|
if (shouldValidateTz && !isValidTimezone(data.timezone)) {
|
|
1375
1369
|
ctx.addIssue({
|
|
1376
|
-
code:
|
|
1370
|
+
code: z32.ZodIssueCode.custom,
|
|
1377
1371
|
message: "Invalid IANA timezone",
|
|
1378
1372
|
path: ["timezone"]
|
|
1379
1373
|
});
|
|
@@ -1593,7 +1587,7 @@ export {
|
|
|
1593
1587
|
CostAssertionSchema,
|
|
1594
1588
|
CostConfigSchema,
|
|
1595
1589
|
CreateAgentInputSchema,
|
|
1596
|
-
|
|
1590
|
+
CreateEvalRunFolderInputSchema,
|
|
1597
1591
|
CreateEvalRunInputSchema,
|
|
1598
1592
|
CreateEvalScheduleInputSchema,
|
|
1599
1593
|
CreateMcpInputSchema,
|
|
@@ -1606,7 +1600,6 @@ export {
|
|
|
1606
1600
|
CreateTemplateInputSchema,
|
|
1607
1601
|
CreateTestScenarioInputSchema,
|
|
1608
1602
|
CreateTestSuiteInputSchema,
|
|
1609
|
-
CustomAssertionSchema,
|
|
1610
1603
|
DEFAULT_EVALUATOR_SYSTEM_PROMPT,
|
|
1611
1604
|
DEFAULT_JUDGE_MODEL,
|
|
1612
1605
|
DiffContentSchema,
|
|
@@ -1614,6 +1607,8 @@ export {
|
|
|
1614
1607
|
DiffLineTypeSchema,
|
|
1615
1608
|
EnvironmentSchema,
|
|
1616
1609
|
EvalMetricsSchema,
|
|
1610
|
+
EvalRunFolderMembershipSchema,
|
|
1611
|
+
EvalRunFolderSchema,
|
|
1617
1612
|
EvalRunResultSchema,
|
|
1618
1613
|
EvalRunSchema,
|
|
1619
1614
|
EvalScheduleSchema,
|
|
@@ -1706,7 +1701,7 @@ export {
|
|
|
1706
1701
|
TriggerSchema,
|
|
1707
1702
|
TriggerType,
|
|
1708
1703
|
UpdateAgentInputSchema,
|
|
1709
|
-
|
|
1704
|
+
UpdateEvalRunFolderInputSchema,
|
|
1710
1705
|
UpdateEvalScheduleInputSchema,
|
|
1711
1706
|
UpdateMcpInputSchema,
|
|
1712
1707
|
UpdatePresetInputSchema,
|
|
@@ -1719,9 +1714,6 @@ export {
|
|
|
1719
1714
|
UpdateTestSuiteInputSchema,
|
|
1720
1715
|
VitestTestSchema,
|
|
1721
1716
|
formatTraceEventLine,
|
|
1722
|
-
getBuildPassedConfig,
|
|
1723
|
-
getLlmJudgeConfig,
|
|
1724
|
-
getSkillWasCalledConfig,
|
|
1725
1717
|
getSystemAssertion,
|
|
1726
1718
|
getSystemAssertions,
|
|
1727
1719
|
isSystemAssertionId,
|