@wix/evalforge-evaluator 0.144.0 → 0.145.0
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 +1 -71
- package/build/index.js.map +2 -2
- package/build/index.mjs +1 -71
- package/build/index.mjs.map +2 -2
- package/build/types/api-client.d.ts +1 -7
- package/package.json +5 -5
package/build/index.mjs
CHANGED
|
@@ -132,34 +132,12 @@ function createApiClient(serverUrl, options = "") {
|
|
|
132
132
|
getScenario(projectId2, id) {
|
|
133
133
|
return fetchJson(`/projects/${projectId2}/test-scenarios/${id}`);
|
|
134
134
|
},
|
|
135
|
-
getSkill(projectId2, id) {
|
|
136
|
-
return fetchJson(`/projects/${projectId2}/skills/${id}`);
|
|
137
|
-
},
|
|
138
|
-
getSkillVersion(projectId2, skillId, versionId) {
|
|
139
|
-
return fetchJson(
|
|
140
|
-
`/projects/${projectId2}/skills/${skillId}/versions/${versionId}`
|
|
141
|
-
);
|
|
142
|
-
},
|
|
143
|
-
getLatestSkillVersion(projectId2, skillId) {
|
|
144
|
-
return fetchJson(
|
|
145
|
-
`/projects/${projectId2}/skills/${skillId}/versions/latest`
|
|
146
|
-
);
|
|
147
|
-
},
|
|
148
135
|
getAgent(projectId2, id) {
|
|
149
136
|
return fetchJson(`/projects/${projectId2}/agents/${id}`);
|
|
150
137
|
},
|
|
151
138
|
getTemplate(projectId2, id) {
|
|
152
139
|
return fetchJson(`/projects/${projectId2}/templates/${id}`);
|
|
153
140
|
},
|
|
154
|
-
getMcp(projectId2, id) {
|
|
155
|
-
return fetchJson(`/projects/${projectId2}/mcps/${id}`);
|
|
156
|
-
},
|
|
157
|
-
getSubAgent(projectId2, id) {
|
|
158
|
-
return fetchJson(`/projects/${projectId2}/sub-agents/${id}`);
|
|
159
|
-
},
|
|
160
|
-
getRule(projectId2, id) {
|
|
161
|
-
return fetchJson(`/projects/${projectId2}/rules/${id}`);
|
|
162
|
-
},
|
|
163
141
|
getPreset(projectId2, id) {
|
|
164
142
|
return fetchJson(`/projects/${projectId2}/presets/${id}`);
|
|
165
143
|
},
|
|
@@ -382,54 +360,6 @@ async function fetchEvaluationData(api, projectId2, evalRunId2) {
|
|
|
382
360
|
break;
|
|
383
361
|
}
|
|
384
362
|
}
|
|
385
|
-
} else {
|
|
386
|
-
const resolvedSkillIds = evalRun.skillIds ?? [];
|
|
387
|
-
if (resolvedSkillIds.length > 0) {
|
|
388
|
-
const fetchResults = await Promise.allSettled(
|
|
389
|
-
resolvedSkillIds.map((id) => api.getSkill(projectId2, id))
|
|
390
|
-
);
|
|
391
|
-
skills = fetchResults.filter(
|
|
392
|
-
(r) => r.status === "fulfilled"
|
|
393
|
-
).map((r) => r.value).filter((s) => !s.deleted);
|
|
394
|
-
if (evalRun.skillVersions && Object.keys(evalRun.skillVersions).length > 0) {
|
|
395
|
-
skills = await Promise.all(
|
|
396
|
-
skills.map(async (skill) => {
|
|
397
|
-
const versionId = evalRun.skillVersions?.[skill.id];
|
|
398
|
-
if (versionId) {
|
|
399
|
-
const version = await api.getSkillVersion(
|
|
400
|
-
projectId2,
|
|
401
|
-
skill.id,
|
|
402
|
-
versionId
|
|
403
|
-
);
|
|
404
|
-
return { ...skill, latestVersion: version };
|
|
405
|
-
}
|
|
406
|
-
return skill;
|
|
407
|
-
})
|
|
408
|
-
);
|
|
409
|
-
}
|
|
410
|
-
skills = skills.map((skill) => {
|
|
411
|
-
const hasPinnedVersion = evalRun.skillVersions?.[skill.id];
|
|
412
|
-
if (!hasPinnedVersion && skill.source) {
|
|
413
|
-
return { ...skill, latestVersion: void 0 };
|
|
414
|
-
}
|
|
415
|
-
return skill;
|
|
416
|
-
});
|
|
417
|
-
}
|
|
418
|
-
if (evalRun.mcpIds && evalRun.mcpIds.length > 0) {
|
|
419
|
-
mcps = await Promise.all(
|
|
420
|
-
evalRun.mcpIds.map((id) => api.getMcp(projectId2, id))
|
|
421
|
-
);
|
|
422
|
-
}
|
|
423
|
-
if (evalRun.subAgentIds && evalRun.subAgentIds.length > 0) {
|
|
424
|
-
subAgents = await Promise.all(
|
|
425
|
-
evalRun.subAgentIds.map((id) => api.getSubAgent(projectId2, id))
|
|
426
|
-
);
|
|
427
|
-
}
|
|
428
|
-
if (evalRun.ruleIds && evalRun.ruleIds.length > 0) {
|
|
429
|
-
rules = await Promise.all(
|
|
430
|
-
evalRun.ruleIds.map((id) => api.getRule(projectId2, id))
|
|
431
|
-
);
|
|
432
|
-
}
|
|
433
363
|
}
|
|
434
364
|
const templateIds = [
|
|
435
365
|
...new Set(
|
|
@@ -5042,7 +4972,7 @@ async function runEvaluation(projectId2, evalRunId2) {
|
|
|
5042
4972
|
hasAgent: !!agent,
|
|
5043
4973
|
agentId: evalData.evalRun.agentId,
|
|
5044
4974
|
presetId: evalData.evalRun.presetId,
|
|
5045
|
-
|
|
4975
|
+
capabilityIds: evalData.evalRun.capabilityIds
|
|
5046
4976
|
};
|
|
5047
4977
|
if (scenarioItems.length > 0 && !agent) {
|
|
5048
4978
|
throw new Error(
|