expert_eval_cli 2.0.0 → 2.0.2
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/client.d.ts +8 -1
- package/dist/client.d.ts.map +1 -1
- package/dist/client.js +23 -0
- package/dist/client.js.map +1 -1
- package/dist/commands/eval-cases.d.ts +3 -0
- package/dist/commands/eval-cases.d.ts.map +1 -0
- package/dist/commands/eval-cases.js +273 -0
- package/dist/commands/eval-cases.js.map +1 -0
- package/dist/commands/eval-tasks.d.ts.map +1 -1
- package/dist/commands/eval-tasks.js +36 -0
- package/dist/commands/eval-tasks.js.map +1 -1
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +3 -1
- package/dist/index.js.map +1 -1
- package/dist/types.d.ts +38 -1
- package/dist/types.d.ts.map +1 -1
- package/package.json +1 -1
- package/skill/SKILL.md +96 -5
- package/skill/references/api-reference.md +44 -0
package/dist/client.d.ts
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import type { AgentAccount, BatchUpdateEvalRecordsPayload, CreateEvalTaskPayload, CreateExpertPayload, CreateVersionPayload, EvalRecord, EvalTask, Expert, ListAgentAccountsParams, ListEvalRecordsParams, ListEvalTasksParams, ListExpertsParams, ListVersionsParams, PaginatedResponse, UpdateEvalRecordPayload, UpdateExpertPayload, Version } from './types.js';
|
|
1
|
+
import type { AgentAccount, BatchUpdateEvalRecordsPayload, CreateEvalCasePayload, CreateEvalTaskPayload, CreateExpertPayload, CreateVersionPayload, EvalCase, EvalRecord, EvalTask, Expert, ListAgentAccountsParams, ListEvalCasesParams, ListEvalRecordsParams, ListEvalTasksParams, ListExpertsParams, ListVersionsParams, PaginatedResponse, UpdateEvalCasePayload, UpdateEvalTaskPayload, UpdateEvalRecordPayload, UpdateExpertPayload, Version } from './types.js';
|
|
2
2
|
export declare class ApiError extends Error {
|
|
3
3
|
readonly statusCode: number;
|
|
4
4
|
readonly apiCode?: string;
|
|
@@ -28,11 +28,18 @@ export declare function deleteVersion(id: string): Promise<{
|
|
|
28
28
|
export declare function listEvalTasks(params?: ListEvalTasksParams): Promise<PaginatedResponse<EvalTask>>;
|
|
29
29
|
export declare function getEvalTask(id: string): Promise<EvalTask>;
|
|
30
30
|
export declare function createEvalTask(payload: CreateEvalTaskPayload): Promise<EvalTask>;
|
|
31
|
+
export declare function updateEvalTask(id: string, payload: UpdateEvalTaskPayload): Promise<EvalTask>;
|
|
31
32
|
export declare function deleteEvalTask(id: string): Promise<{
|
|
32
33
|
id: string;
|
|
33
34
|
}>;
|
|
34
35
|
export declare function listEvalTaskRecords(id: string): Promise<EvalRecord[]>;
|
|
35
36
|
export declare function getEvalTaskAgentCard(id: string): Promise<Record<string, unknown>>;
|
|
37
|
+
export declare function listEvalCases(params?: ListEvalCasesParams): Promise<EvalCase[]>;
|
|
38
|
+
export declare function createEvalCase(payload: CreateEvalCasePayload): Promise<EvalCase>;
|
|
39
|
+
export declare function updateEvalCase(id: string, payload: UpdateEvalCasePayload): Promise<EvalCase>;
|
|
40
|
+
export declare function deleteEvalCase(id: string): Promise<{
|
|
41
|
+
id: string;
|
|
42
|
+
}>;
|
|
36
43
|
export declare function listEvalRecords(params: ListEvalRecordsParams): Promise<EvalRecord[]>;
|
|
37
44
|
export declare function updateEvalRecord(id: string, payload: UpdateEvalRecordPayload): Promise<EvalRecord>;
|
|
38
45
|
export declare function batchUpdateEvalRecords(payload: BatchUpdateEvalRecordsPayload): Promise<{
|
package/dist/client.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../src/client.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,YAAY,EAEZ,6BAA6B,EAC7B,qBAAqB,EACrB,mBAAmB,EACnB,oBAAoB,EACpB,UAAU,EACV,QAAQ,EACR,MAAM,EACN,uBAAuB,EACvB,qBAAqB,EACrB,mBAAmB,EACnB,iBAAiB,EACjB,kBAAkB,EAClB,iBAAiB,EACjB,uBAAuB,EACvB,mBAAmB,EACnB,OAAO,EACR,MAAM,YAAY,CAAC;AAGpB,qBAAa,QAAS,SAAQ,KAAK;IACjC,SAAgB,UAAU,EAAE,MAAM,CAAC;IACnC,SAAgB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjC,SAAgB,OAAO,CAAC,EAAE,OAAO,CAAC;gBAEtB,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,OAAO;CAOrF;AAED,qBAAa,SAAU,SAAQ,KAAK;gBACtB,OAAO,SAAgG;CAIpH;AAiGD,wBAAgB,WAAW,CAAC,MAAM,CAAC,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC,CAE1F;AAED,wBAAgB,SAAS,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAE3D;AAED,wBAAgB,YAAY,CAAC,OAAO,EAAE,mBAAmB,GAAG,OAAO,CAAC,MAAM,CAAC,CAE1E;AAED,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,mBAAmB,GAAG,OAAO,CAAC,MAAM,CAAC,CAE5F;AAED,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC;IAAE,EAAE,EAAE,MAAM,CAAA;CAAE,CAAC,CAEtE;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAErF;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAEpE;AAED,0CAA0C;AAC1C,eAAO,MAAM,YAAY,2BAAqB,CAAC;AAM/C,wBAAgB,YAAY,CAAC,MAAM,CAAC,EAAE,kBAAkB,GAAG,OAAO,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC,CAE7F;AAED,wBAAgB,UAAU,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAEvD;AAED,wBAAgB,aAAa,CAAC,OAAO,EAAE,oBAAoB,GAAG,OAAO,CAAC,OAAO,CAAC,CAE7E;AAED,wBAAgB,aAAa,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC;IAAE,EAAE,EAAE,MAAM,CAAA;CAAE,CAAC,CAEjE;AAMD,wBAAgB,aAAa,CAAC,MAAM,CAAC,EAAE,mBAAmB,GAAG,OAAO,CAAC,iBAAiB,CAAC,QAAQ,CAAC,CAAC,CAEhG;AAED,wBAAgB,WAAW,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,QAAQ,CAAC,CAEzD;AAED,wBAAgB,cAAc,CAAC,OAAO,EAAE,qBAAqB,GAAG,OAAO,CAAC,QAAQ,CAAC,CAEhF;AAED,wBAAgB,cAAc,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC;IAAE,EAAE,EAAE,MAAM,CAAA;CAAE,CAAC,CAElE;AAED,wBAAgB,mBAAmB,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC,CAErE;AAED,wBAAgB,oBAAoB,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAEjF;AAMD,wBAAgB,eAAe,CAAC,MAAM,EAAE,qBAAqB,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC,CAEpF;AAED,wBAAgB,gBAAgB,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,uBAAuB,GAAG,OAAO,CAAC,UAAU,CAAC,CAElG;AAED,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,6BAA6B,GAAG,OAAO,CAAC;IAAE,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,CAE3G;AAMD,wBAAgB,iBAAiB,CAAC,MAAM,CAAC,EAAE,uBAAuB,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC,CAE3F;AAED,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,CAAC,CAExE;AAED,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,CAAC,CAEvE"}
|
|
1
|
+
{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../src/client.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,YAAY,EAEZ,6BAA6B,EAC7B,qBAAqB,EACrB,qBAAqB,EACrB,mBAAmB,EACnB,oBAAoB,EACpB,QAAQ,EACR,UAAU,EACV,QAAQ,EACR,MAAM,EACN,uBAAuB,EACvB,mBAAmB,EACnB,qBAAqB,EACrB,mBAAmB,EACnB,iBAAiB,EACjB,kBAAkB,EAClB,iBAAiB,EACjB,qBAAqB,EACrB,qBAAqB,EACrB,uBAAuB,EACvB,mBAAmB,EACnB,OAAO,EACR,MAAM,YAAY,CAAC;AAGpB,qBAAa,QAAS,SAAQ,KAAK;IACjC,SAAgB,UAAU,EAAE,MAAM,CAAC;IACnC,SAAgB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjC,SAAgB,OAAO,CAAC,EAAE,OAAO,CAAC;gBAEtB,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,OAAO;CAOrF;AAED,qBAAa,SAAU,SAAQ,KAAK;gBACtB,OAAO,SAAgG;CAIpH;AAiGD,wBAAgB,WAAW,CAAC,MAAM,CAAC,EAAE,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC,CAE1F;AAED,wBAAgB,SAAS,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAE3D;AAED,wBAAgB,YAAY,CAAC,OAAO,EAAE,mBAAmB,GAAG,OAAO,CAAC,MAAM,CAAC,CAE1E;AAED,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,mBAAmB,GAAG,OAAO,CAAC,MAAM,CAAC,CAE5F;AAED,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC;IAAE,EAAE,EAAE,MAAM,CAAA;CAAE,CAAC,CAEtE;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAErF;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAEpE;AAED,0CAA0C;AAC1C,eAAO,MAAM,YAAY,2BAAqB,CAAC;AAM/C,wBAAgB,YAAY,CAAC,MAAM,CAAC,EAAE,kBAAkB,GAAG,OAAO,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC,CAE7F;AAED,wBAAgB,UAAU,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAEvD;AAED,wBAAgB,aAAa,CAAC,OAAO,EAAE,oBAAoB,GAAG,OAAO,CAAC,OAAO,CAAC,CAE7E;AAED,wBAAgB,aAAa,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC;IAAE,EAAE,EAAE,MAAM,CAAA;CAAE,CAAC,CAEjE;AAMD,wBAAgB,aAAa,CAAC,MAAM,CAAC,EAAE,mBAAmB,GAAG,OAAO,CAAC,iBAAiB,CAAC,QAAQ,CAAC,CAAC,CAEhG;AAED,wBAAgB,WAAW,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,QAAQ,CAAC,CAEzD;AAED,wBAAgB,cAAc,CAAC,OAAO,EAAE,qBAAqB,GAAG,OAAO,CAAC,QAAQ,CAAC,CAEhF;AAED,wBAAgB,cAAc,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,qBAAqB,GAAG,OAAO,CAAC,QAAQ,CAAC,CAE5F;AAED,wBAAgB,cAAc,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC;IAAE,EAAE,EAAE,MAAM,CAAA;CAAE,CAAC,CAElE;AAED,wBAAgB,mBAAmB,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC,CAErE;AAED,wBAAgB,oBAAoB,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAEjF;AAMD,wBAAgB,aAAa,CAAC,MAAM,CAAC,EAAE,mBAAmB,GAAG,OAAO,CAAC,QAAQ,EAAE,CAAC,CAE/E;AAED,wBAAgB,cAAc,CAAC,OAAO,EAAE,qBAAqB,GAAG,OAAO,CAAC,QAAQ,CAAC,CAEhF;AAED,wBAAgB,cAAc,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,qBAAqB,GAAG,OAAO,CAAC,QAAQ,CAAC,CAE5F;AAED,wBAAgB,cAAc,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC;IAAE,EAAE,EAAE,MAAM,CAAA;CAAE,CAAC,CAElE;AAMD,wBAAgB,eAAe,CAAC,MAAM,EAAE,qBAAqB,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC,CAEpF;AAED,wBAAgB,gBAAgB,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,uBAAuB,GAAG,OAAO,CAAC,UAAU,CAAC,CAElG;AAED,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,6BAA6B,GAAG,OAAO,CAAC;IAAE,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,CAE3G;AAMD,wBAAgB,iBAAiB,CAAC,MAAM,CAAC,EAAE,uBAAuB,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC,CAE3F;AAED,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,CAAC,CAExE;AAED,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,CAAC,CAEvE"}
|
package/dist/client.js
CHANGED
|
@@ -15,9 +15,14 @@ exports.deleteVersion = deleteVersion;
|
|
|
15
15
|
exports.listEvalTasks = listEvalTasks;
|
|
16
16
|
exports.getEvalTask = getEvalTask;
|
|
17
17
|
exports.createEvalTask = createEvalTask;
|
|
18
|
+
exports.updateEvalTask = updateEvalTask;
|
|
18
19
|
exports.deleteEvalTask = deleteEvalTask;
|
|
19
20
|
exports.listEvalTaskRecords = listEvalTaskRecords;
|
|
20
21
|
exports.getEvalTaskAgentCard = getEvalTaskAgentCard;
|
|
22
|
+
exports.listEvalCases = listEvalCases;
|
|
23
|
+
exports.createEvalCase = createEvalCase;
|
|
24
|
+
exports.updateEvalCase = updateEvalCase;
|
|
25
|
+
exports.deleteEvalCase = deleteEvalCase;
|
|
21
26
|
exports.listEvalRecords = listEvalRecords;
|
|
22
27
|
exports.updateEvalRecord = updateEvalRecord;
|
|
23
28
|
exports.batchUpdateEvalRecords = batchUpdateEvalRecords;
|
|
@@ -162,6 +167,9 @@ function getEvalTask(id) {
|
|
|
162
167
|
function createEvalTask(payload) {
|
|
163
168
|
return request('POST', '/openapi/eval-tasks', { body: payload });
|
|
164
169
|
}
|
|
170
|
+
function updateEvalTask(id, payload) {
|
|
171
|
+
return request('PATCH', `/openapi/eval-tasks/${encodeURIComponent(id)}`, { body: payload });
|
|
172
|
+
}
|
|
165
173
|
function deleteEvalTask(id) {
|
|
166
174
|
return request('DELETE', `/openapi/eval-tasks/${encodeURIComponent(id)}`);
|
|
167
175
|
}
|
|
@@ -172,6 +180,21 @@ function getEvalTaskAgentCard(id) {
|
|
|
172
180
|
return request('GET', `/openapi/eval-tasks/${encodeURIComponent(id)}/agent-card`);
|
|
173
181
|
}
|
|
174
182
|
// ============================
|
|
183
|
+
// Eval Cases
|
|
184
|
+
// ============================
|
|
185
|
+
function listEvalCases(params) {
|
|
186
|
+
return request('GET', '/openapi/eval-cases', { query: params });
|
|
187
|
+
}
|
|
188
|
+
function createEvalCase(payload) {
|
|
189
|
+
return request('POST', '/openapi/eval-cases', { body: payload });
|
|
190
|
+
}
|
|
191
|
+
function updateEvalCase(id, payload) {
|
|
192
|
+
return request('PATCH', `/openapi/eval-cases/${encodeURIComponent(id)}`, { body: payload });
|
|
193
|
+
}
|
|
194
|
+
function deleteEvalCase(id) {
|
|
195
|
+
return request('DELETE', `/openapi/eval-cases/${encodeURIComponent(id)}`);
|
|
196
|
+
}
|
|
197
|
+
// ============================
|
|
175
198
|
// Eval Records
|
|
176
199
|
// ============================
|
|
177
200
|
function listEvalRecords(params) {
|
package/dist/client.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"client.js","sourceRoot":"","sources":["../src/client.ts"],"names":[],"mappings":";;;
|
|
1
|
+
{"version":3,"file":"client.js","sourceRoot":"","sources":["../src/client.ts"],"names":[],"mappings":";;;AA+IA,kCAEC;AAED,8BAEC;AAED,oCAEC;AAED,oCAEC;AAED,oCAEC;AAED,gDAEC;AAED,gDAEC;AASD,oCAEC;AAED,gCAEC;AAED,sCAEC;AAED,sCAEC;AAMD,sCAEC;AAED,kCAEC;AAED,wCAEC;AAED,wCAEC;AAED,wCAEC;AAED,kDAEC;AAED,oDAEC;AAMD,sCAEC;AAED,wCAEC;AAED,wCAEC;AAED,wCAEC;AAMD,0CAEC;AAED,4CAEC;AAED,wDAEC;AAMD,8CAEC;AAED,8CAEC;AAED,4CAEC;AA3PD,2CAAoD;AAEpD,MAAa,QAAS,SAAQ,KAAK;IACjB,UAAU,CAAS;IACnB,OAAO,CAAU;IACjB,OAAO,CAAW;IAElC,YAAY,OAAe,EAAE,UAAkB,EAAE,OAAgB,EAAE,OAAiB;QAClF,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,UAAU,CAAC;QACvB,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;QAC7B,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;QACvB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;IACzB,CAAC;CACF;AAZD,4BAYC;AAED,MAAa,SAAU,SAAQ,KAAK;IAClC,YAAY,OAAO,GAAG,6FAA6F;QACjH,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,WAAW,CAAC;IAC1B,CAAC;CACF;AALD,8BAKC;AAID,SAAS,QAAQ,CAAC,IAAY,EAAE,IAAY,EAAE,MAAoB;IAChE,MAAM,SAAS,GAAG,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC9D,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,SAAS,EAAE,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,GAAG,CAAC,CAAC;IACvE,IAAI,MAAM,EAAE,CAAC;QACX,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;YAClD,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,KAAK,EAAE,EAAE,CAAC;gBAC1D,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;YAC3C,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC,QAAQ,EAAE,CAAC;AACxB,CAAC;AAED,KAAK,UAAU,OAAO,CACpB,MAAc,EACd,IAAY,EACZ,UAKI,EAAE;IAEN,MAAM,OAAO,GAAG,IAAA,sBAAU,GAAE,CAAC;IAC7B,IAAI,CAAC,OAAO,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CACb,mJAAmJ,CACpJ,CAAC;IACJ,CAAC;IAED,MAAM,MAAM,GAAG,IAAA,qBAAS,GAAE,CAAC;IAC3B,IAAI,CAAC,MAAM,IAAI,CAAC,OAAO,CAAC,QAAQ,EAAE,CAAC;QACjC,MAAM,IAAI,SAAS,EAAE,CAAC;IACxB,CAAC;IAED,MAAM,OAAO,GAA2B;QACtC,MAAM,EAAE,OAAO,CAAC,MAAM,IAAI,kBAAkB;KAC7C,CAAC;IAEF,IAAI,OAAO,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;QAC/B,OAAO,CAAC,cAAc,CAAC,GAAG,kBAAkB,CAAC;IAC/C,CAAC;IAED,IAAI,MAAM,EAAE,CAAC;QACX,OAAO,CAAC,aAAa,GAAG,UAAU,MAAM,EAAE,CAAC;IAC7C,CAAC;IAED,MAAM,YAAY,GAAgB,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;IACtD,IAAI,OAAO,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;QAC/B,YAAY,CAAC,IAAI,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IACnD,CAAC;IAED,IAAI,QAAkB,CAAC;IACvB,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,KAAK,CAAC,QAAQ,CAAC,OAAO,EAAE,IAAI,EAAE,OAAO,CAAC,KAAK,CAAC,EAAE,YAAY,CAAC,CAAC;IAC/E,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,OAAO,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACjE,MAAM,IAAI,QAAQ,CAAC,kBAAkB,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;IACrD,CAAC;IAED,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;QACvD,MAAM,IAAI,SAAS,CAAC,+BAA+B,QAAQ,CAAC,MAAM,wBAAwB,CAAC,CAAC;IAC9F,CAAC;IAED,MAAM,WAAW,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,IAAI,EAAE,CAAC;IAC/D,IAAI,IAAa,CAAC;IAClB,IAAI,WAAW,CAAC,QAAQ,CAAC,kBAAkB,CAAC,EAAE,CAAC;QAC7C,IAAI,CAAC;YACH,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QAC/B,CAAC;QAAC,MAAM,CAAC;YACP,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QAC/B,CAAC;IACH,CAAC;SAAM,CAAC;QACN,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;IAC/B,CAAC;IAED,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,MAAM,OAAO,GAAG,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,IAAI,CAAC,CAAC,CAAE,IAAyB,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5F,MAAM,IAAI,QAAQ,CAChB,OAAO,CAAC,OAAO,IAAI,QAAQ,QAAQ,CAAC,MAAM,IAAI,QAAQ,CAAC,UAAU,EAAE,EACnE,QAAQ,CAAC,MAAM,EACf,OAAO,CAAC,IAAI,EACZ,OAAO,CAAC,OAAO,CAChB,CAAC;IACJ,CAAC;IAED,OAAO,IAAS,CAAC;AACnB,CAAC;AAED,+BAA+B;AAC/B,UAAU;AACV,+BAA+B;AAE/B,SAAgB,WAAW,CAAC,MAA0B;IACpD,OAAO,OAAO,CAAC,KAAK,EAAE,kBAAkB,EAAE,EAAE,KAAK,EAAE,MAAiC,EAAE,CAAC,CAAC;AAC1F,CAAC;AAED,SAAgB,SAAS,CAAC,QAAgB;IACxC,OAAO,OAAO,CAAC,KAAK,EAAE,oBAAoB,kBAAkB,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;AAC5E,CAAC;AAED,SAAgB,YAAY,CAAC,OAA4B;IACvD,OAAO,OAAO,CAAC,MAAM,EAAE,kBAAkB,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;AAChE,CAAC;AAED,SAAgB,YAAY,CAAC,QAAgB,EAAE,OAA4B;IACzE,OAAO,OAAO,CAAC,OAAO,EAAE,oBAAoB,kBAAkB,CAAC,QAAQ,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;AACjG,CAAC;AAED,SAAgB,YAAY,CAAC,QAAgB;IAC3C,OAAO,OAAO,CAAC,QAAQ,EAAE,oBAAoB,kBAAkB,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;AAC/E,CAAC;AAED,SAAgB,kBAAkB,CAAC,QAAgB;IACjD,OAAO,OAAO,CAAC,KAAK,EAAE,oBAAoB,kBAAkB,CAAC,QAAQ,CAAC,aAAa,CAAC,CAAC;AACvF,CAAC;AAED,SAAgB,kBAAkB,CAAC,QAAgB;IACjD,OAAO,OAAO,CAAC,KAAK,EAAE,oBAAoB,kBAAkB,CAAC,QAAQ,CAAC,YAAY,EAAE,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC,CAAC;AAC/G,CAAC;AAED,0CAA0C;AAC7B,QAAA,YAAY,GAAG,kBAAkB,CAAC;AAE/C,+BAA+B;AAC/B,WAAW;AACX,+BAA+B;AAE/B,SAAgB,YAAY,CAAC,MAA2B;IACtD,OAAO,OAAO,CAAC,KAAK,EAAE,mBAAmB,EAAE,EAAE,KAAK,EAAE,MAAiC,EAAE,CAAC,CAAC;AAC3F,CAAC;AAED,SAAgB,UAAU,CAAC,EAAU;IACnC,OAAO,OAAO,CAAC,KAAK,EAAE,qBAAqB,kBAAkB,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;AACvE,CAAC;AAED,SAAgB,aAAa,CAAC,OAA6B;IACzD,OAAO,OAAO,CAAC,MAAM,EAAE,mBAAmB,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;AACjE,CAAC;AAED,SAAgB,aAAa,CAAC,EAAU;IACtC,OAAO,OAAO,CAAC,QAAQ,EAAE,qBAAqB,kBAAkB,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;AAC1E,CAAC;AAED,+BAA+B;AAC/B,aAAa;AACb,+BAA+B;AAE/B,SAAgB,aAAa,CAAC,MAA4B;IACxD,OAAO,OAAO,CAAC,KAAK,EAAE,qBAAqB,EAAE,EAAE,KAAK,EAAE,MAAiC,EAAE,CAAC,CAAC;AAC7F,CAAC;AAED,SAAgB,WAAW,CAAC,EAAU;IACpC,OAAO,OAAO,CAAC,KAAK,EAAE,uBAAuB,kBAAkB,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;AACzE,CAAC;AAED,SAAgB,cAAc,CAAC,OAA8B;IAC3D,OAAO,OAAO,CAAC,MAAM,EAAE,qBAAqB,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;AACnE,CAAC;AAED,SAAgB,cAAc,CAAC,EAAU,EAAE,OAA8B;IACvE,OAAO,OAAO,CAAC,OAAO,EAAE,uBAAuB,kBAAkB,CAAC,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;AAC9F,CAAC;AAED,SAAgB,cAAc,CAAC,EAAU;IACvC,OAAO,OAAO,CAAC,QAAQ,EAAE,uBAAuB,kBAAkB,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;AAC5E,CAAC;AAED,SAAgB,mBAAmB,CAAC,EAAU;IAC5C,OAAO,OAAO,CAAC,KAAK,EAAE,uBAAuB,kBAAkB,CAAC,EAAE,CAAC,UAAU,CAAC,CAAC;AACjF,CAAC;AAED,SAAgB,oBAAoB,CAAC,EAAU;IAC7C,OAAO,OAAO,CAAC,KAAK,EAAE,uBAAuB,kBAAkB,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC;AACpF,CAAC;AAED,+BAA+B;AAC/B,aAAa;AACb,+BAA+B;AAE/B,SAAgB,aAAa,CAAC,MAA4B;IACxD,OAAO,OAAO,CAAC,KAAK,EAAE,qBAAqB,EAAE,EAAE,KAAK,EAAE,MAAiC,EAAE,CAAC,CAAC;AAC7F,CAAC;AAED,SAAgB,cAAc,CAAC,OAA8B;IAC3D,OAAO,OAAO,CAAC,MAAM,EAAE,qBAAqB,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;AACnE,CAAC;AAED,SAAgB,cAAc,CAAC,EAAU,EAAE,OAA8B;IACvE,OAAO,OAAO,CAAC,OAAO,EAAE,uBAAuB,kBAAkB,CAAC,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;AAC9F,CAAC;AAED,SAAgB,cAAc,CAAC,EAAU;IACvC,OAAO,OAAO,CAAC,QAAQ,EAAE,uBAAuB,kBAAkB,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;AAC5E,CAAC;AAED,+BAA+B;AAC/B,eAAe;AACf,+BAA+B;AAE/B,SAAgB,eAAe,CAAC,MAA6B;IAC3D,OAAO,OAAO,CAAC,KAAK,EAAE,uBAAuB,EAAE,EAAE,KAAK,EAAE,MAAgC,EAAE,CAAC,CAAC;AAC9F,CAAC;AAED,SAAgB,gBAAgB,CAAC,EAAU,EAAE,OAAgC;IAC3E,OAAO,OAAO,CAAC,OAAO,EAAE,yBAAyB,kBAAkB,CAAC,EAAE,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;AAChG,CAAC;AAED,SAAgB,sBAAsB,CAAC,OAAsC;IAC3E,OAAO,OAAO,CAAC,MAAM,EAAE,6BAA6B,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;AAC3E,CAAC;AAED,+BAA+B;AAC/B,iBAAiB;AACjB,+BAA+B;AAE/B,SAAgB,iBAAiB,CAAC,MAAgC;IAChE,OAAO,OAAO,CAAC,KAAK,EAAE,wBAAwB,EAAE,EAAE,KAAK,EAAE,MAAiC,EAAE,CAAC,CAAC;AAChG,CAAC;AAED,SAAgB,iBAAiB,CAAC,OAAe;IAC/C,OAAO,OAAO,CAAC,MAAM,EAAE,0BAA0B,kBAAkB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;AACxF,CAAC;AAED,SAAgB,gBAAgB,CAAC,OAAe;IAC9C,OAAO,OAAO,CAAC,MAAM,EAAE,0BAA0B,kBAAkB,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;AACvF,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval-cases.d.ts","sourceRoot":"","sources":["../../src/commands/eval-cases.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAwIpC,wBAAgB,oBAAoB,IAAI,OAAO,CA+G9C"}
|
|
@@ -0,0 +1,273 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
|
|
3
|
+
if (k2 === undefined) k2 = k;
|
|
4
|
+
var desc = Object.getOwnPropertyDescriptor(m, k);
|
|
5
|
+
if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
|
|
6
|
+
desc = { enumerable: true, get: function() { return m[k]; } };
|
|
7
|
+
}
|
|
8
|
+
Object.defineProperty(o, k2, desc);
|
|
9
|
+
}) : (function(o, m, k, k2) {
|
|
10
|
+
if (k2 === undefined) k2 = k;
|
|
11
|
+
o[k2] = m[k];
|
|
12
|
+
}));
|
|
13
|
+
var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
|
|
14
|
+
Object.defineProperty(o, "default", { enumerable: true, value: v });
|
|
15
|
+
}) : function(o, v) {
|
|
16
|
+
o["default"] = v;
|
|
17
|
+
});
|
|
18
|
+
var __importStar = (this && this.__importStar) || (function () {
|
|
19
|
+
var ownKeys = function(o) {
|
|
20
|
+
ownKeys = Object.getOwnPropertyNames || function (o) {
|
|
21
|
+
var ar = [];
|
|
22
|
+
for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
|
|
23
|
+
return ar;
|
|
24
|
+
};
|
|
25
|
+
return ownKeys(o);
|
|
26
|
+
};
|
|
27
|
+
return function (mod) {
|
|
28
|
+
if (mod && mod.__esModule) return mod;
|
|
29
|
+
var result = {};
|
|
30
|
+
if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
|
|
31
|
+
__setModuleDefault(result, mod);
|
|
32
|
+
return result;
|
|
33
|
+
};
|
|
34
|
+
})();
|
|
35
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
36
|
+
exports.makeEvalCasesCommand = makeEvalCasesCommand;
|
|
37
|
+
const commander_1 = require("commander");
|
|
38
|
+
const node_fs_1 = require("node:fs");
|
|
39
|
+
const client = __importStar(require("../client.js"));
|
|
40
|
+
const output_js_1 = require("../output.js");
|
|
41
|
+
const utils_js_1 = require("../utils.js");
|
|
42
|
+
const EVAL_CASE_STATUSES = ['untested', 'passed', 'failed'];
|
|
43
|
+
function parseEvalCaseStatus(value) {
|
|
44
|
+
if (value === undefined)
|
|
45
|
+
return undefined;
|
|
46
|
+
if (EVAL_CASE_STATUSES.includes(value))
|
|
47
|
+
return value;
|
|
48
|
+
throw new Error(`Invalid status "${value}". Allowed: ${EVAL_CASE_STATUSES.join(', ')}`);
|
|
49
|
+
}
|
|
50
|
+
function parseExtras(options) {
|
|
51
|
+
if (options.clearExtras)
|
|
52
|
+
return null;
|
|
53
|
+
if (options.extrasJson !== undefined && options.extrasFile !== undefined) {
|
|
54
|
+
throw new Error('Use only one of --extras-json or --extras-file');
|
|
55
|
+
}
|
|
56
|
+
const input = options.extrasFile !== undefined
|
|
57
|
+
? (0, node_fs_1.readFileSync)(options.extrasFile, 'utf8')
|
|
58
|
+
: options.extrasJson;
|
|
59
|
+
if (input === undefined)
|
|
60
|
+
return undefined;
|
|
61
|
+
const parsed = JSON.parse(input);
|
|
62
|
+
if (!parsed || typeof parsed !== 'object' || Array.isArray(parsed)) {
|
|
63
|
+
throw new Error('extras must be a JSON object');
|
|
64
|
+
}
|
|
65
|
+
return parsed;
|
|
66
|
+
}
|
|
67
|
+
function nullableString(value, clear) {
|
|
68
|
+
if (clear)
|
|
69
|
+
return null;
|
|
70
|
+
return value;
|
|
71
|
+
}
|
|
72
|
+
function buildCreatePayload(options) {
|
|
73
|
+
if (!options.evalTaskId)
|
|
74
|
+
throw new Error('Missing required option --eval-task-id');
|
|
75
|
+
if (!options.instruction)
|
|
76
|
+
throw new Error('Missing required option --instruction');
|
|
77
|
+
const payload = {
|
|
78
|
+
evalTaskId: options.evalTaskId,
|
|
79
|
+
instruction: options.instruction,
|
|
80
|
+
};
|
|
81
|
+
const expectedResult = nullableString(options.expectedResult, options.clearExpectedResult);
|
|
82
|
+
if (expectedResult !== undefined)
|
|
83
|
+
payload.expectedResult = expectedResult;
|
|
84
|
+
const evaluationCriteria = nullableString(options.evaluationCriteria, options.clearEvaluationCriteria);
|
|
85
|
+
if (evaluationCriteria !== undefined)
|
|
86
|
+
payload.evaluationCriteria = evaluationCriteria;
|
|
87
|
+
const actualResult = nullableString(options.actualResult, options.clearActualResult);
|
|
88
|
+
if (actualResult !== undefined)
|
|
89
|
+
payload.actualResult = actualResult;
|
|
90
|
+
const artifactLink = nullableString(options.artifactLink, options.clearArtifactLink);
|
|
91
|
+
if (artifactLink !== undefined)
|
|
92
|
+
payload.artifactLink = artifactLink;
|
|
93
|
+
const taskLink = nullableString(options.taskLink, options.clearTaskLink);
|
|
94
|
+
if (taskLink !== undefined)
|
|
95
|
+
payload.taskLink = taskLink;
|
|
96
|
+
const executionStartTime = nullableString(options.executionStartTime, options.clearExecutionStartTime);
|
|
97
|
+
if (executionStartTime !== undefined)
|
|
98
|
+
payload.executionStartTime = executionStartTime;
|
|
99
|
+
const executionEndTime = nullableString(options.executionEndTime, options.clearExecutionEndTime);
|
|
100
|
+
if (executionEndTime !== undefined)
|
|
101
|
+
payload.executionEndTime = executionEndTime;
|
|
102
|
+
const extras = parseExtras(options);
|
|
103
|
+
if (extras !== undefined)
|
|
104
|
+
payload.extras = extras;
|
|
105
|
+
const status = parseEvalCaseStatus(options.status);
|
|
106
|
+
if (status !== undefined)
|
|
107
|
+
payload.status = status;
|
|
108
|
+
return payload;
|
|
109
|
+
}
|
|
110
|
+
function buildUpdatePayload(options) {
|
|
111
|
+
const payload = {};
|
|
112
|
+
if (options.instruction !== undefined)
|
|
113
|
+
payload.instruction = options.instruction;
|
|
114
|
+
const expectedResult = nullableString(options.expectedResult, options.clearExpectedResult);
|
|
115
|
+
if (expectedResult !== undefined)
|
|
116
|
+
payload.expectedResult = expectedResult;
|
|
117
|
+
const evaluationCriteria = nullableString(options.evaluationCriteria, options.clearEvaluationCriteria);
|
|
118
|
+
if (evaluationCriteria !== undefined)
|
|
119
|
+
payload.evaluationCriteria = evaluationCriteria;
|
|
120
|
+
const actualResult = nullableString(options.actualResult, options.clearActualResult);
|
|
121
|
+
if (actualResult !== undefined)
|
|
122
|
+
payload.actualResult = actualResult;
|
|
123
|
+
const artifactLink = nullableString(options.artifactLink, options.clearArtifactLink);
|
|
124
|
+
if (artifactLink !== undefined)
|
|
125
|
+
payload.artifactLink = artifactLink;
|
|
126
|
+
const taskLink = nullableString(options.taskLink, options.clearTaskLink);
|
|
127
|
+
if (taskLink !== undefined)
|
|
128
|
+
payload.taskLink = taskLink;
|
|
129
|
+
const executionStartTime = nullableString(options.executionStartTime, options.clearExecutionStartTime);
|
|
130
|
+
if (executionStartTime !== undefined)
|
|
131
|
+
payload.executionStartTime = executionStartTime;
|
|
132
|
+
const executionEndTime = nullableString(options.executionEndTime, options.clearExecutionEndTime);
|
|
133
|
+
if (executionEndTime !== undefined)
|
|
134
|
+
payload.executionEndTime = executionEndTime;
|
|
135
|
+
const extras = parseExtras(options);
|
|
136
|
+
if (extras !== undefined)
|
|
137
|
+
payload.extras = extras;
|
|
138
|
+
const status = parseEvalCaseStatus(options.status);
|
|
139
|
+
if (status !== undefined)
|
|
140
|
+
payload.status = status;
|
|
141
|
+
if (Object.keys(payload).length === 0) {
|
|
142
|
+
throw new Error('Provide at least one field to update.');
|
|
143
|
+
}
|
|
144
|
+
return payload;
|
|
145
|
+
}
|
|
146
|
+
function toEvalCaseRows(cases) {
|
|
147
|
+
return cases.map((item) => ({
|
|
148
|
+
id: item.id,
|
|
149
|
+
evalTaskId: item.evalTaskId,
|
|
150
|
+
instruction: item.instruction,
|
|
151
|
+
expectedResult: item.expectedResult || '',
|
|
152
|
+
evaluationCriteria: item.evaluationCriteria || '',
|
|
153
|
+
actualResult: item.actualResult || '',
|
|
154
|
+
status: item.status || '',
|
|
155
|
+
artifactLink: item.artifactLink || '',
|
|
156
|
+
taskLink: item.taskLink || '',
|
|
157
|
+
executionStartTime: item.executionStartTime || '',
|
|
158
|
+
executionEndTime: item.executionEndTime || '',
|
|
159
|
+
}));
|
|
160
|
+
}
|
|
161
|
+
function makeEvalCasesCommand() {
|
|
162
|
+
const evalCases = new commander_1.Command('eval-cases').description('Manage evaluation cases');
|
|
163
|
+
evalCases
|
|
164
|
+
.command('list')
|
|
165
|
+
.description('List evaluation cases by task, expert, or status')
|
|
166
|
+
.option('--eval-task-id <id>', 'Filter by evaluation task ID')
|
|
167
|
+
.option('--expert-id <id>', 'Filter by expert business ID')
|
|
168
|
+
.option('--status <status>', `Case status: ${EVAL_CASE_STATUSES.join(', ')}`)
|
|
169
|
+
.action(async (options) => {
|
|
170
|
+
const globalOpts = evalCases.optsWithGlobals();
|
|
171
|
+
try {
|
|
172
|
+
const params = {
|
|
173
|
+
evalTaskId: options.evalTaskId,
|
|
174
|
+
expertId: options.expertId,
|
|
175
|
+
status: parseEvalCaseStatus(options.status),
|
|
176
|
+
};
|
|
177
|
+
const cases = await client.listEvalCases(params);
|
|
178
|
+
if (globalOpts.quiet)
|
|
179
|
+
return;
|
|
180
|
+
const fmt = globalOpts.format === 'json' || !process.stdout.isTTY ? 'json' : 'csv';
|
|
181
|
+
if (fmt === 'json') {
|
|
182
|
+
(0, output_js_1.printJson)(cases, globalOpts);
|
|
183
|
+
return;
|
|
184
|
+
}
|
|
185
|
+
if (cases.length === 0) {
|
|
186
|
+
console.log('No evaluation cases found.');
|
|
187
|
+
return;
|
|
188
|
+
}
|
|
189
|
+
(0, output_js_1.printCsv)(toEvalCaseRows(cases), globalOpts);
|
|
190
|
+
}
|
|
191
|
+
catch (err) {
|
|
192
|
+
(0, output_js_1.printError)(err, globalOpts);
|
|
193
|
+
process.exitCode = utils_js_1.EXIT_CODES.GENERAL_ERROR;
|
|
194
|
+
}
|
|
195
|
+
});
|
|
196
|
+
evalCases
|
|
197
|
+
.command('create')
|
|
198
|
+
.description('Create an evaluation case')
|
|
199
|
+
.requiredOption('--eval-task-id <id>', 'Evaluation task ID')
|
|
200
|
+
.requiredOption('--instruction <text>', 'Instruction to execute')
|
|
201
|
+
.option('--expected-result <text>', 'Expected result')
|
|
202
|
+
.option('--evaluation-criteria <text>', 'Evaluation criteria')
|
|
203
|
+
.option('--actual-result <text>', 'Actual result')
|
|
204
|
+
.option('--artifact-link <url>', 'Artifact URL')
|
|
205
|
+
.option('--task-link <url>', 'External task URL')
|
|
206
|
+
.option('--extras-json <json>', 'Extras JSON object')
|
|
207
|
+
.option('--extras-file <path>', 'Read extras JSON object from file')
|
|
208
|
+
.option('--status <status>', `Case status: ${EVAL_CASE_STATUSES.join(', ')}`)
|
|
209
|
+
.option('--execution-start-time <time>', 'Execution start time')
|
|
210
|
+
.option('--execution-end-time <time>', 'Execution end time')
|
|
211
|
+
.action(async (options) => {
|
|
212
|
+
const globalOpts = evalCases.optsWithGlobals();
|
|
213
|
+
try {
|
|
214
|
+
const item = await client.createEvalCase(buildCreatePayload(options));
|
|
215
|
+
(0, output_js_1.printItem)(item, globalOpts);
|
|
216
|
+
}
|
|
217
|
+
catch (err) {
|
|
218
|
+
(0, output_js_1.printError)(err, globalOpts);
|
|
219
|
+
process.exitCode = utils_js_1.EXIT_CODES.GENERAL_ERROR;
|
|
220
|
+
}
|
|
221
|
+
});
|
|
222
|
+
evalCases
|
|
223
|
+
.command('update')
|
|
224
|
+
.description('Partially update an evaluation case')
|
|
225
|
+
.argument('<id>', 'Evaluation case ID')
|
|
226
|
+
.option('--instruction <text>', 'Instruction to execute')
|
|
227
|
+
.option('--expected-result <text>', 'Expected result')
|
|
228
|
+
.option('--clear-expected-result', 'Set expected result to null')
|
|
229
|
+
.option('--evaluation-criteria <text>', 'Evaluation criteria')
|
|
230
|
+
.option('--clear-evaluation-criteria', 'Set evaluation criteria to null')
|
|
231
|
+
.option('--actual-result <text>', 'Actual result')
|
|
232
|
+
.option('--clear-actual-result', 'Set actual result to null')
|
|
233
|
+
.option('--artifact-link <url>', 'Artifact URL')
|
|
234
|
+
.option('--clear-artifact-link', 'Set artifact link to null')
|
|
235
|
+
.option('--task-link <url>', 'External task URL')
|
|
236
|
+
.option('--clear-task-link', 'Set task link to null')
|
|
237
|
+
.option('--extras-json <json>', 'Extras JSON object')
|
|
238
|
+
.option('--extras-file <path>', 'Read extras JSON object from file')
|
|
239
|
+
.option('--clear-extras', 'Set extras to null')
|
|
240
|
+
.option('--status <status>', `Case status: ${EVAL_CASE_STATUSES.join(', ')}`)
|
|
241
|
+
.option('--execution-start-time <time>', 'Execution start time')
|
|
242
|
+
.option('--clear-execution-start-time', 'Set execution start time to null')
|
|
243
|
+
.option('--execution-end-time <time>', 'Execution end time')
|
|
244
|
+
.option('--clear-execution-end-time', 'Set execution end time to null')
|
|
245
|
+
.action(async (id, options) => {
|
|
246
|
+
const globalOpts = evalCases.optsWithGlobals();
|
|
247
|
+
try {
|
|
248
|
+
const item = await client.updateEvalCase(id, buildUpdatePayload(options));
|
|
249
|
+
(0, output_js_1.printItem)(item, globalOpts);
|
|
250
|
+
}
|
|
251
|
+
catch (err) {
|
|
252
|
+
(0, output_js_1.printError)(err, globalOpts);
|
|
253
|
+
process.exitCode = utils_js_1.EXIT_CODES.GENERAL_ERROR;
|
|
254
|
+
}
|
|
255
|
+
});
|
|
256
|
+
evalCases
|
|
257
|
+
.command('delete')
|
|
258
|
+
.description('Delete an evaluation case')
|
|
259
|
+
.argument('<id>', 'Evaluation case ID')
|
|
260
|
+
.action(async (id) => {
|
|
261
|
+
const globalOpts = evalCases.optsWithGlobals();
|
|
262
|
+
try {
|
|
263
|
+
const result = await client.deleteEvalCase(id);
|
|
264
|
+
(0, output_js_1.printItem)(result, globalOpts);
|
|
265
|
+
}
|
|
266
|
+
catch (err) {
|
|
267
|
+
(0, output_js_1.printError)(err, globalOpts);
|
|
268
|
+
process.exitCode = utils_js_1.EXIT_CODES.GENERAL_ERROR;
|
|
269
|
+
}
|
|
270
|
+
});
|
|
271
|
+
return evalCases;
|
|
272
|
+
}
|
|
273
|
+
//# sourceMappingURL=eval-cases.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval-cases.js","sourceRoot":"","sources":["../../src/commands/eval-cases.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAwIA,oDA+GC;AAvPD,yCAAoC;AACpC,qCAAuC;AACvC,qDAAuC;AACvC,4CAA0E;AAC1E,0CAAyC;AAGzC,MAAM,kBAAkB,GAAqB,CAAC,UAAU,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAC;AA+B9E,SAAS,mBAAmB,CAAC,KAAyB;IACpD,IAAI,KAAK,KAAK,SAAS;QAAE,OAAO,SAAS,CAAC;IAC1C,IAAI,kBAAkB,CAAC,QAAQ,CAAC,KAAuB,CAAC;QAAE,OAAO,KAAuB,CAAC;IACzF,MAAM,IAAI,KAAK,CAAC,mBAAmB,KAAK,eAAe,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAC1F,CAAC;AAED,SAAS,WAAW,CAAC,OAAmF;IACtG,IAAI,OAAO,CAAC,WAAW;QAAE,OAAO,IAAI,CAAC;IACrC,IAAI,OAAO,CAAC,UAAU,KAAK,SAAS,IAAI,OAAO,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QACzE,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;IACpE,CAAC;IACD,MAAM,KAAK,GAAG,OAAO,CAAC,UAAU,KAAK,SAAS;QAC5C,CAAC,CAAC,IAAA,sBAAY,EAAC,OAAO,CAAC,UAAU,EAAE,MAAM,CAAC;QAC1C,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC;IACvB,IAAI,KAAK,KAAK,SAAS;QAAE,OAAO,SAAS,CAAC;IAC1C,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAY,CAAC;IAC5C,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QACnE,MAAM,IAAI,KAAK,CAAC,8BAA8B,CAAC,CAAC;IAClD,CAAC;IACD,OAAO,MAAoB,CAAC;AAC9B,CAAC;AAED,SAAS,cAAc,CAAC,KAAyB,EAAE,KAA0B;IAC3E,IAAI,KAAK;QAAE,OAAO,IAAI,CAAC;IACvB,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,kBAAkB,CAAC,OAAgC;IAC1D,IAAI,CAAC,OAAO,CAAC,UAAU;QAAE,MAAM,IAAI,KAAK,CAAC,wCAAwC,CAAC,CAAC;IACnF,IAAI,CAAC,OAAO,CAAC,WAAW;QAAE,MAAM,IAAI,KAAK,CAAC,uCAAuC,CAAC,CAAC;IACnF,MAAM,OAAO,GAA0B;QACrC,UAAU,EAAE,OAAO,CAAC,UAAU;QAC9B,WAAW,EAAE,OAAO,CAAC,WAAW;KACjC,CAAC;IACF,MAAM,cAAc,GAAG,cAAc,CAAC,OAAO,CAAC,cAAc,EAAE,OAAO,CAAC,mBAAmB,CAAC,CAAC;IAC3F,IAAI,cAAc,KAAK,SAAS;QAAE,OAAO,CAAC,cAAc,GAAG,cAAc,CAAC;IAC1E,MAAM,kBAAkB,GAAG,cAAc,CAAC,OAAO,CAAC,kBAAkB,EAAE,OAAO,CAAC,uBAAuB,CAAC,CAAC;IACvG,IAAI,kBAAkB,KAAK,SAAS;QAAE,OAAO,CAAC,kBAAkB,GAAG,kBAAkB,CAAC;IACtF,MAAM,YAAY,GAAG,cAAc,CAAC,OAAO,CAAC,YAAY,EAAE,OAAO,CAAC,iBAAiB,CAAC,CAAC;IACrF,IAAI,YAAY,KAAK,SAAS;QAAE,OAAO,CAAC,YAAY,GAAG,YAAY,CAAC;IACpE,MAAM,YAAY,GAAG,cAAc,CAAC,OAAO,CAAC,YAAY,EAAE,OAAO,CAAC,iBAAiB,CAAC,CAAC;IACrF,IAAI,YAAY,KAAK,SAAS;QAAE,OAAO,CAAC,YAAY,GAAG,YAAY,CAAC;IACpE,MAAM,QAAQ,GAAG,cAAc,CAAC,OAAO,CAAC,QAAQ,EAAE,OAAO,CAAC,aAAa,CAAC,CAAC;IACzE,IAAI,QAAQ,KAAK,SAAS;QAAE,OAAO,CAAC,QAAQ,GAAG,QAAQ,CAAC;IACxD,MAAM,kBAAkB,GAAG,cAAc,CAAC,OAAO,CAAC,kBAAkB,EAAE,OAAO,CAAC,uBAAuB,CAAC,CAAC;IACvG,IAAI,kBAAkB,KAAK,SAAS;QAAE,OAAO,CAAC,kBAAkB,GAAG,kBAAkB,CAAC;IACtF,MAAM,gBAAgB,GAAG,cAAc,CAAC,OAAO,CAAC,gBAAgB,EAAE,OAAO,CAAC,qBAAqB,CAAC,CAAC;IACjG,IAAI,gBAAgB,KAAK,SAAS;QAAE,OAAO,CAAC,gBAAgB,GAAG,gBAAgB,CAAC;IAChF,MAAM,MAAM,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IACpC,IAAI,MAAM,KAAK,SAAS;QAAE,OAAO,CAAC,MAAM,GAAG,MAAM,CAAC;IAClD,MAAM,MAAM,GAAG,mBAAmB,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IACnD,IAAI,MAAM,KAAK,SAAS;QAAE,OAAO,CAAC,MAAM,GAAG,MAAM,CAAC;IAClD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,kBAAkB,CAAC,OAAgC;IAC1D,MAAM,OAAO,GAA0B,EAAE,CAAC;IAC1C,IAAI,OAAO,CAAC,WAAW,KAAK,SAAS;QAAE,OAAO,CAAC,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC;IACjF,MAAM,cAAc,GAAG,cAAc,CAAC,OAAO,CAAC,cAAc,EAAE,OAAO,CAAC,mBAAmB,CAAC,CAAC;IAC3F,IAAI,cAAc,KAAK,SAAS;QAAE,OAAO,CAAC,cAAc,GAAG,cAAc,CAAC;IAC1E,MAAM,kBAAkB,GAAG,cAAc,CAAC,OAAO,CAAC,kBAAkB,EAAE,OAAO,CAAC,uBAAuB,CAAC,CAAC;IACvG,IAAI,kBAAkB,KAAK,SAAS;QAAE,OAAO,CAAC,kBAAkB,GAAG,kBAAkB,CAAC;IACtF,MAAM,YAAY,GAAG,cAAc,CAAC,OAAO,CAAC,YAAY,EAAE,OAAO,CAAC,iBAAiB,CAAC,CAAC;IACrF,IAAI,YAAY,KAAK,SAAS;QAAE,OAAO,CAAC,YAAY,GAAG,YAAY,CAAC;IACpE,MAAM,YAAY,GAAG,cAAc,CAAC,OAAO,CAAC,YAAY,EAAE,OAAO,CAAC,iBAAiB,CAAC,CAAC;IACrF,IAAI,YAAY,KAAK,SAAS;QAAE,OAAO,CAAC,YAAY,GAAG,YAAY,CAAC;IACpE,MAAM,QAAQ,GAAG,cAAc,CAAC,OAAO,CAAC,QAAQ,EAAE,OAAO,CAAC,aAAa,CAAC,CAAC;IACzE,IAAI,QAAQ,KAAK,SAAS;QAAE,OAAO,CAAC,QAAQ,GAAG,QAAQ,CAAC;IACxD,MAAM,kBAAkB,GAAG,cAAc,CAAC,OAAO,CAAC,kBAAkB,EAAE,OAAO,CAAC,uBAAuB,CAAC,CAAC;IACvG,IAAI,kBAAkB,KAAK,SAAS;QAAE,OAAO,CAAC,kBAAkB,GAAG,kBAAkB,CAAC;IACtF,MAAM,gBAAgB,GAAG,cAAc,CAAC,OAAO,CAAC,gBAAgB,EAAE,OAAO,CAAC,qBAAqB,CAAC,CAAC;IACjG,IAAI,gBAAgB,KAAK,SAAS;QAAE,OAAO,CAAC,gBAAgB,GAAG,gBAAgB,CAAC;IAChF,MAAM,MAAM,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IACpC,IAAI,MAAM,KAAK,SAAS;QAAE,OAAO,CAAC,MAAM,GAAG,MAAM,CAAC;IAClD,MAAM,MAAM,GAAG,mBAAmB,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IACnD,IAAI,MAAM,KAAK,SAAS;QAAE,OAAO,CAAC,MAAM,GAAG,MAAM,CAAC;IAClD,IAAI,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,uCAAuC,CAAC,CAAC;IAC3D,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,cAAc,CAAC,KAAuD;IAC7E,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;QAC1B,EAAE,EAAE,IAAI,CAAC,EAAE;QACX,UAAU,EAAE,IAAI,CAAC,UAAU;QAC3B,WAAW,EAAE,IAAI,CAAC,WAAW;QAC7B,cAAc,EAAE,IAAI,CAAC,cAAc,IAAI,EAAE;QACzC,kBAAkB,EAAE,IAAI,CAAC,kBAAkB,IAAI,EAAE;QACjD,YAAY,EAAE,IAAI,CAAC,YAAY,IAAI,EAAE;QACrC,MAAM,EAAE,IAAI,CAAC,MAAM,IAAI,EAAE;QACzB,YAAY,EAAE,IAAI,CAAC,YAAY,IAAI,EAAE;QACrC,QAAQ,EAAE,IAAI,CAAC,QAAQ,IAAI,EAAE;QAC7B,kBAAkB,EAAE,IAAI,CAAC,kBAAkB,IAAI,EAAE;QACjD,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,IAAI,EAAE;KAC9C,CAAC,CAAC,CAAC;AACN,CAAC;AAED,SAAgB,oBAAoB;IAClC,MAAM,SAAS,GAAG,IAAI,mBAAO,CAAC,YAAY,CAAC,CAAC,WAAW,CAAC,yBAAyB,CAAC,CAAC;IAEnF,SAAS;SACN,OAAO,CAAC,MAAM,CAAC;SACf,WAAW,CAAC,kDAAkD,CAAC;SAC/D,MAAM,CAAC,qBAAqB,EAAE,8BAA8B,CAAC;SAC7D,MAAM,CAAC,kBAAkB,EAAE,8BAA8B,CAAC;SAC1D,MAAM,CAAC,mBAAmB,EAAE,gBAAgB,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;SAC5E,MAAM,CAAC,KAAK,EAAE,OAA4B,EAAE,EAAE;QAC7C,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,MAAM,GAAwB;gBAClC,UAAU,EAAE,OAAO,CAAC,UAAU;gBAC9B,QAAQ,EAAE,OAAO,CAAC,QAAQ;gBAC1B,MAAM,EAAE,mBAAmB,CAAC,OAAO,CAAC,MAAM,CAAC;aAC5C,CAAC;YACF,MAAM,KAAK,GAAG,MAAM,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;YACjD,IAAI,UAAU,CAAC,KAAK;gBAAE,OAAO;YAC7B,MAAM,GAAG,GAAG,UAAU,CAAC,MAAM,KAAK,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC;YACnF,IAAI,GAAG,KAAK,MAAM,EAAE,CAAC;gBACnB,IAAA,qBAAS,EAAC,KAAK,EAAE,UAAU,CAAC,CAAC;gBAC7B,OAAO;YACT,CAAC;YACD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACvB,OAAO,CAAC,GAAG,CAAC,4BAA4B,CAAC,CAAC;gBAC1C,OAAO;YACT,CAAC;YACD,IAAA,oBAAQ,EAAC,cAAc,CAAC,KAAK,CAAC,EAAE,UAAU,CAAC,CAAC;QAC9C,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,SAAS;SACN,OAAO,CAAC,QAAQ,CAAC;SACjB,WAAW,CAAC,2BAA2B,CAAC;SACxC,cAAc,CAAC,qBAAqB,EAAE,oBAAoB,CAAC;SAC3D,cAAc,CAAC,sBAAsB,EAAE,wBAAwB,CAAC;SAChE,MAAM,CAAC,0BAA0B,EAAE,iBAAiB,CAAC;SACrD,MAAM,CAAC,8BAA8B,EAAE,qBAAqB,CAAC;SAC7D,MAAM,CAAC,wBAAwB,EAAE,eAAe,CAAC;SACjD,MAAM,CAAC,uBAAuB,EAAE,cAAc,CAAC;SAC/C,MAAM,CAAC,mBAAmB,EAAE,mBAAmB,CAAC;SAChD,MAAM,CAAC,sBAAsB,EAAE,oBAAoB,CAAC;SACpD,MAAM,CAAC,sBAAsB,EAAE,mCAAmC,CAAC;SACnE,MAAM,CAAC,mBAAmB,EAAE,gBAAgB,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;SAC5E,MAAM,CAAC,+BAA+B,EAAE,sBAAsB,CAAC;SAC/D,MAAM,CAAC,6BAA6B,EAAE,oBAAoB,CAAC;SAC3D,MAAM,CAAC,KAAK,EAAE,OAAgC,EAAE,EAAE;QACjD,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC,kBAAkB,CAAC,OAAO,CAAC,CAAC,CAAC;YACtE,IAAA,qBAAS,EAAC,IAA0C,EAAE,UAAU,CAAC,CAAC;QACpE,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,SAAS;SACN,OAAO,CAAC,QAAQ,CAAC;SACjB,WAAW,CAAC,qCAAqC,CAAC;SAClD,QAAQ,CAAC,MAAM,EAAE,oBAAoB,CAAC;SACtC,MAAM,CAAC,sBAAsB,EAAE,wBAAwB,CAAC;SACxD,MAAM,CAAC,0BAA0B,EAAE,iBAAiB,CAAC;SACrD,MAAM,CAAC,yBAAyB,EAAE,6BAA6B,CAAC;SAChE,MAAM,CAAC,8BAA8B,EAAE,qBAAqB,CAAC;SAC7D,MAAM,CAAC,6BAA6B,EAAE,iCAAiC,CAAC;SACxE,MAAM,CAAC,wBAAwB,EAAE,eAAe,CAAC;SACjD,MAAM,CAAC,uBAAuB,EAAE,2BAA2B,CAAC;SAC5D,MAAM,CAAC,uBAAuB,EAAE,cAAc,CAAC;SAC/C,MAAM,CAAC,uBAAuB,EAAE,2BAA2B,CAAC;SAC5D,MAAM,CAAC,mBAAmB,EAAE,mBAAmB,CAAC;SAChD,MAAM,CAAC,mBAAmB,EAAE,uBAAuB,CAAC;SACpD,MAAM,CAAC,sBAAsB,EAAE,oBAAoB,CAAC;SACpD,MAAM,CAAC,sBAAsB,EAAE,mCAAmC,CAAC;SACnE,MAAM,CAAC,gBAAgB,EAAE,oBAAoB,CAAC;SAC9C,MAAM,CAAC,mBAAmB,EAAE,gBAAgB,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;SAC5E,MAAM,CAAC,+BAA+B,EAAE,sBAAsB,CAAC;SAC/D,MAAM,CAAC,8BAA8B,EAAE,kCAAkC,CAAC;SAC1E,MAAM,CAAC,6BAA6B,EAAE,oBAAoB,CAAC;SAC3D,MAAM,CAAC,4BAA4B,EAAE,gCAAgC,CAAC;SACtE,MAAM,CAAC,KAAK,EAAE,EAAU,EAAE,OAAgC,EAAE,EAAE;QAC7D,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC,EAAE,EAAE,kBAAkB,CAAC,OAAO,CAAC,CAAC,CAAC;YAC1E,IAAA,qBAAS,EAAC,IAA0C,EAAE,UAAU,CAAC,CAAC;QACpE,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,SAAS;SACN,OAAO,CAAC,QAAQ,CAAC;SACjB,WAAW,CAAC,2BAA2B,CAAC;SACxC,QAAQ,CAAC,MAAM,EAAE,oBAAoB,CAAC;SACtC,MAAM,CAAC,KAAK,EAAE,EAAU,EAAE,EAAE;QAC3B,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC,EAAE,CAAC,CAAC;YAC/C,IAAA,qBAAS,EAAC,MAA4C,EAAE,UAAU,CAAC,CAAC;QACtE,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,OAAO,SAAS,CAAC;AACnB,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"eval-tasks.d.ts","sourceRoot":"","sources":["../../src/commands/eval-tasks.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;
|
|
1
|
+
{"version":3,"file":"eval-tasks.d.ts","sourceRoot":"","sources":["../../src/commands/eval-tasks.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAoDpC,wBAAgB,oBAAoB,IAAI,OAAO,CAqJ9C"}
|
|
@@ -46,6 +46,21 @@ function parseEvalTaskStatus(value) {
|
|
|
46
46
|
return value;
|
|
47
47
|
throw new Error(`Invalid status "${value}". Allowed: ${EVAL_TASK_STATUSES.join(', ')}`);
|
|
48
48
|
}
|
|
49
|
+
function buildUpdateEvalTaskPayload(options) {
|
|
50
|
+
const payload = {};
|
|
51
|
+
if (options.clearPackageReport)
|
|
52
|
+
payload.packageReport = null;
|
|
53
|
+
else if (options.packageReport !== undefined)
|
|
54
|
+
payload.packageReport = options.packageReport;
|
|
55
|
+
if (options.clearCaseReport)
|
|
56
|
+
payload.caseReport = null;
|
|
57
|
+
else if (options.caseReport !== undefined)
|
|
58
|
+
payload.caseReport = options.caseReport;
|
|
59
|
+
if (Object.keys(payload).length === 0) {
|
|
60
|
+
throw new Error('Provide at least one report option: --package-report, --case-report, --clear-package-report, or --clear-case-report');
|
|
61
|
+
}
|
|
62
|
+
return payload;
|
|
63
|
+
}
|
|
49
64
|
function toEvalTaskRows(items) {
|
|
50
65
|
return items.map((task) => ({
|
|
51
66
|
id: task.id,
|
|
@@ -58,6 +73,8 @@ function toEvalTaskRows(items) {
|
|
|
58
73
|
failed: task.resultSummary.failed,
|
|
59
74
|
warning: task.resultSummary.warning,
|
|
60
75
|
cases: `${task.caseSummary.passed}/${task.caseSummary.total}`,
|
|
76
|
+
packageReport: task.packageReport || '',
|
|
77
|
+
caseReport: task.caseReport || '',
|
|
61
78
|
startedAt: task.startedAt || '',
|
|
62
79
|
completedAt: task.completedAt || '',
|
|
63
80
|
}));
|
|
@@ -136,6 +153,25 @@ function makeEvalTasksCommand() {
|
|
|
136
153
|
process.exitCode = utils_js_1.EXIT_CODES.GENERAL_ERROR;
|
|
137
154
|
}
|
|
138
155
|
});
|
|
156
|
+
evalTasks
|
|
157
|
+
.command('update')
|
|
158
|
+
.description('Update evaluation task report links')
|
|
159
|
+
.argument('<id>', 'Evaluation task ID')
|
|
160
|
+
.option('--package-report <url>', 'Package evaluation report URL')
|
|
161
|
+
.option('--case-report <url>', 'Case evaluation report URL')
|
|
162
|
+
.option('--clear-package-report', 'Clear package evaluation report URL')
|
|
163
|
+
.option('--clear-case-report', 'Clear case evaluation report URL')
|
|
164
|
+
.action(async (id, options) => {
|
|
165
|
+
const globalOpts = evalTasks.optsWithGlobals();
|
|
166
|
+
try {
|
|
167
|
+
const item = await client.updateEvalTask(id, buildUpdateEvalTaskPayload(options));
|
|
168
|
+
(0, output_js_1.printItem)(item, globalOpts);
|
|
169
|
+
}
|
|
170
|
+
catch (err) {
|
|
171
|
+
(0, output_js_1.printError)(err, globalOpts);
|
|
172
|
+
process.exitCode = utils_js_1.EXIT_CODES.GENERAL_ERROR;
|
|
173
|
+
}
|
|
174
|
+
});
|
|
139
175
|
evalTasks
|
|
140
176
|
.command('delete')
|
|
141
177
|
.description('Delete an evaluation task')
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"eval-tasks.js","sourceRoot":"","sources":["../../src/commands/eval-tasks.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"eval-tasks.js","sourceRoot":"","sources":["../../src/commands/eval-tasks.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAoDA,oDAqJC;AAzMD,yCAAoC;AACpC,qDAAuC;AACvC,4CAA0E;AAC1E,0CAAsD;AAGtD,MAAM,kBAAkB,GAAqB,CAAC,SAAS,EAAE,aAAa,EAAE,QAAQ,EAAE,mBAAmB,EAAE,QAAQ,CAAC,CAAC;AAEjH,SAAS,mBAAmB,CAAC,KAAyB;IACpD,IAAI,KAAK,KAAK,SAAS;QAAE,OAAO,SAAS,CAAC;IAC1C,IAAI,kBAAkB,CAAC,QAAQ,CAAC,KAAuB,CAAC;QAAE,OAAO,KAAuB,CAAC;IACzF,MAAM,IAAI,KAAK,CAAC,mBAAmB,KAAK,eAAe,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAC1F,CAAC;AASD,SAAS,0BAA0B,CAAC,OAA8B;IAChE,MAAM,OAAO,GAA0B,EAAE,CAAC;IAC1C,IAAI,OAAO,CAAC,kBAAkB;QAAE,OAAO,CAAC,aAAa,GAAG,IAAI,CAAC;SACxD,IAAI,OAAO,CAAC,aAAa,KAAK,SAAS;QAAE,OAAO,CAAC,aAAa,GAAG,OAAO,CAAC,aAAa,CAAC;IAC5F,IAAI,OAAO,CAAC,eAAe;QAAE,OAAO,CAAC,UAAU,GAAG,IAAI,CAAC;SAClD,IAAI,OAAO,CAAC,UAAU,KAAK,SAAS;QAAE,OAAO,CAAC,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC;IACnF,IAAI,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,qHAAqH,CAAC,CAAC;IACzI,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,cAAc,CAAC,KAAgE;IACtF,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;QAC1B,EAAE,EAAE,IAAI,CAAC,EAAE;QACX,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,SAAS,EAAE,IAAI,CAAC,SAAS;QACzB,SAAS,EAAE,IAAI,CAAC,SAAS;QACzB,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,MAAM,EAAE,IAAI,CAAC,MAAM;QACnB,OAAO,EAAE,GAAG,IAAI,CAAC,aAAa,CAAC,MAAM,IAAI,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE;QACnE,MAAM,EAAE,IAAI,CAAC,aAAa,CAAC,MAAM;QACjC,OAAO,EAAE,IAAI,CAAC,aAAa,CAAC,OAAO;QACnC,KAAK,EAAE,GAAG,IAAI,CAAC,WAAW,CAAC,MAAM,IAAI,IAAI,CAAC,WAAW,CAAC,KAAK,EAAE;QAC7D,aAAa,EAAE,IAAI,CAAC,aAAa,IAAI,EAAE;QACvC,UAAU,EAAE,IAAI,CAAC,UAAU,IAAI,EAAE;QACjC,SAAS,EAAE,IAAI,CAAC,SAAS,IAAI,EAAE;QAC/B,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,EAAE;KACpC,CAAC,CAAC,CAAC;AACN,CAAC;AAED,SAAgB,oBAAoB;IAClC,MAAM,SAAS,GAAG,IAAI,mBAAO,CAAC,YAAY,CAAC,CAAC,WAAW,CAAC,yBAAyB,CAAC,CAAC;IAEnF,SAAS;SACN,OAAO,CAAC,MAAM,CAAC;SACf,WAAW,CAAC,uCAAuC,CAAC;SACpD,MAAM,CAAC,YAAY,EAAE,uBAAuB,EAAE,GAAG,CAAC;SAClD,MAAM,CAAC,iBAAiB,EAAE,gBAAgB,EAAE,IAAI,CAAC;SACjD,MAAM,CAAC,kBAAkB,EAAE,8BAA8B,CAAC;SAC1D,MAAM,CAAC,mBAAmB,EAAE,sBAAsB,CAAC;SACnD,MAAM,CAAC,mBAAmB,EAAE,qBAAqB,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;SACjF,MAAM,CAAC,KAAK,EAAE,OAAqG,EAAE,EAAE;QACtH,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,MAAM,GAAwB;gBAClC,IAAI,EAAE,IAAA,sBAAW,EAAC,OAAO,CAAC,IAAI,CAAC;gBAC/B,QAAQ,EAAE,IAAA,sBAAW,EAAC,OAAO,CAAC,QAAQ,CAAC;gBACvC,QAAQ,EAAE,OAAO,CAAC,QAAQ;gBAC1B,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,MAAM,EAAE,mBAAmB,CAAC,OAAO,CAAC,MAAM,CAAC;aAC5C,CAAC;YACF,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;YAClD,IAAI,UAAU,CAAC,KAAK;gBAAE,OAAO;YAC7B,MAAM,GAAG,GAAG,UAAU,CAAC,MAAM,KAAK,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC;YACnF,IAAI,GAAG,KAAK,MAAM,EAAE,CAAC;gBACnB,IAAA,qBAAS,EAAC,MAAM,EAAE,UAAU,CAAC,CAAC;gBAC9B,OAAO;YACT,CAAC;YACD,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAC9B,OAAO,CAAC,GAAG,CAAC,4BAA4B,CAAC,CAAC;gBAC1C,OAAO;YACT,CAAC;YACD,IAAA,oBAAQ,EAAC,cAAc,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,UAAU,CAAC,CAAC;YACnD,OAAO,CAAC,GAAG,CAAC,UAAU,MAAM,CAAC,IAAI,IAAI,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,GAAG,MAAM,CAAC,QAAQ,CAAC,aAAa,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;QAC7G,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,SAAS;SACN,OAAO,CAAC,KAAK,CAAC;SACd,WAAW,CAAC,kEAAkE,CAAC;SAC/E,QAAQ,CAAC,MAAM,EAAE,oBAAoB,CAAC;SACtC,MAAM,CAAC,KAAK,EAAE,EAAU,EAAE,EAAE;QAC3B,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC;YAC1C,IAAA,qBAAS,EAAC,IAA0C,EAAE,UAAU,CAAC,CAAC;QACpE,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,SAAS;SACN,OAAO,CAAC,QAAQ,CAAC;SACjB,WAAW,CAAC,yCAAyC,CAAC;SACtD,cAAc,CAAC,mBAAmB,EAAE,YAAY,CAAC;SACjD,MAAM,CAAC,eAAe,EAAE,gDAAgD,CAAC;SACzE,MAAM,CAAC,KAAK,EAAE,OAA6C,EAAE,EAAE;QAC9D,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC;gBACvC,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,SAAS;aAChC,CAAC,CAAC;YACH,IAAA,qBAAS,EAAC,IAA0C,EAAE,UAAU,CAAC,CAAC;QACpE,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,SAAS;SACN,OAAO,CAAC,QAAQ,CAAC;SACjB,WAAW,CAAC,qCAAqC,CAAC;SAClD,QAAQ,CAAC,MAAM,EAAE,oBAAoB,CAAC;SACtC,MAAM,CAAC,wBAAwB,EAAE,+BAA+B,CAAC;SACjE,MAAM,CAAC,qBAAqB,EAAE,4BAA4B,CAAC;SAC3D,MAAM,CAAC,wBAAwB,EAAE,qCAAqC,CAAC;SACvE,MAAM,CAAC,qBAAqB,EAAE,kCAAkC,CAAC;SACjE,MAAM,CAAC,KAAK,EAAE,EAAU,EAAE,OAA8B,EAAE,EAAE;QAC3D,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC,EAAE,EAAE,0BAA0B,CAAC,OAAO,CAAC,CAAC,CAAC;YAClF,IAAA,qBAAS,EAAC,IAA0C,EAAE,UAAU,CAAC,CAAC;QACpE,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,SAAS;SACN,OAAO,CAAC,QAAQ,CAAC;SACjB,WAAW,CAAC,2BAA2B,CAAC;SACxC,QAAQ,CAAC,MAAM,EAAE,oBAAoB,CAAC;SACtC,MAAM,CAAC,KAAK,EAAE,EAAU,EAAE,EAAE;QAC3B,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC,EAAE,CAAC,CAAC;YAC/C,IAAA,qBAAS,EAAC,MAA4C,EAAE,UAAU,CAAC,CAAC;QACtE,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,SAAS;SACN,OAAO,CAAC,SAAS,CAAC;SAClB,WAAW,CAAC,kFAAkF,CAAC;SAC/F,QAAQ,CAAC,MAAM,EAAE,oBAAoB,CAAC;SACtC,MAAM,CAAC,KAAK,EAAE,EAAU,EAAE,EAAE;QAC3B,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,EAAE,CAAC,CAAC;YACrD,IAAI,UAAU,CAAC,KAAK;gBAAE,OAAO;YAC7B,MAAM,GAAG,GAAG,UAAU,CAAC,MAAM,KAAK,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC;YACnF,IAAI,GAAG,KAAK,MAAM,EAAE,CAAC;gBACnB,IAAA,qBAAS,EAAC,OAAO,EAAE,UAAU,CAAC,CAAC;gBAC/B,OAAO;YACT,CAAC;YACD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACzB,OAAO,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;gBAC5C,OAAO;YACT,CAAC;YACD,IAAA,oBAAQ,EAAC,OAAO,EAAE,UAAU,CAAC,CAAC;QAChC,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,SAAS;SACN,OAAO,CAAC,YAAY,CAAC;SACrB,WAAW,CAAC,iDAAiD,CAAC;SAC9D,QAAQ,CAAC,MAAM,EAAE,oBAAoB,CAAC;SACtC,MAAM,CAAC,KAAK,EAAE,EAAU,EAAE,EAAE;QAC3B,MAAM,UAAU,GAAG,SAAS,CAAC,eAAe,EAAiB,CAAC;QAC9D,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,EAAE,CAAC,CAAC;YACnD,IAAA,qBAAS,EAAC,IAAI,EAAE,UAAU,CAAC,CAAC;QAC9B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,sBAAU,EAAC,GAAG,EAAE,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,QAAQ,GAAG,qBAAU,CAAC,aAAa,CAAC;QAC9C,CAAC;IACH,CAAC,CAAC,CAAC;IAEL,OAAO,SAAS,CAAC;AACnB,CAAC"}
|
package/dist/index.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AASpC,wBAAgB,aAAa,IAAI,OAAO,CAmBvC"}
|
package/dist/index.js
CHANGED
|
@@ -7,13 +7,14 @@ const experts_js_1 = require("./commands/experts.js");
|
|
|
7
7
|
const versions_js_1 = require("./commands/versions.js");
|
|
8
8
|
const eval_records_js_1 = require("./commands/eval-records.js");
|
|
9
9
|
const eval_tasks_js_1 = require("./commands/eval-tasks.js");
|
|
10
|
+
const eval_cases_js_1 = require("./commands/eval-cases.js");
|
|
10
11
|
const agent_accounts_js_1 = require("./commands/agent-accounts.js");
|
|
11
12
|
function createProgram() {
|
|
12
13
|
const program = new commander_1.Command();
|
|
13
14
|
program
|
|
14
15
|
.name('expert_eval')
|
|
15
16
|
.description('AI-friendly CLI for ExpertEval API')
|
|
16
|
-
.version('2.0.
|
|
17
|
+
.version('2.0.2')
|
|
17
18
|
.option('--format <type>', 'Output format: json, csv, table, or auto (default: auto)', 'auto')
|
|
18
19
|
.option('-q, --quiet', 'Suppress output (exit codes only)', false);
|
|
19
20
|
program.addCommand((0, auth_js_1.makeAuthCommand)());
|
|
@@ -21,6 +22,7 @@ function createProgram() {
|
|
|
21
22
|
program.addCommand((0, versions_js_1.makeVersionsCommand)());
|
|
22
23
|
program.addCommand((0, eval_records_js_1.makeEvalRecordsCommand)());
|
|
23
24
|
program.addCommand((0, eval_tasks_js_1.makeEvalTasksCommand)());
|
|
25
|
+
program.addCommand((0, eval_cases_js_1.makeEvalCasesCommand)());
|
|
24
26
|
program.addCommand((0, agent_accounts_js_1.makeAgentAccountsCommand)());
|
|
25
27
|
return program;
|
|
26
28
|
}
|
package/dist/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";;
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";;AASA,sCAmBC;AA5BD,yCAAoC;AACpC,gDAAqD;AACrD,sDAA2D;AAC3D,wDAA6D;AAC7D,gEAAoE;AACpE,4DAAgE;AAChE,4DAAgE;AAChE,oEAAwE;AAExE,SAAgB,aAAa;IAC3B,MAAM,OAAO,GAAG,IAAI,mBAAO,EAAE,CAAC;IAE9B,OAAO;SACJ,IAAI,CAAC,aAAa,CAAC;SACnB,WAAW,CAAC,oCAAoC,CAAC;SACjD,OAAO,CAAC,OAAO,CAAC;SAChB,MAAM,CAAC,iBAAiB,EAAE,0DAA0D,EAAE,MAAM,CAAC;SAC7F,MAAM,CAAC,aAAa,EAAE,mCAAmC,EAAE,KAAK,CAAC,CAAC;IAErE,OAAO,CAAC,UAAU,CAAC,IAAA,yBAAe,GAAE,CAAC,CAAC;IACtC,OAAO,CAAC,UAAU,CAAC,IAAA,+BAAkB,GAAE,CAAC,CAAC;IACzC,OAAO,CAAC,UAAU,CAAC,IAAA,iCAAmB,GAAE,CAAC,CAAC;IAC1C,OAAO,CAAC,UAAU,CAAC,IAAA,wCAAsB,GAAE,CAAC,CAAC;IAC7C,OAAO,CAAC,UAAU,CAAC,IAAA,oCAAoB,GAAE,CAAC,CAAC;IAC3C,OAAO,CAAC,UAAU,CAAC,IAAA,oCAAoB,GAAE,CAAC,CAAC;IAC3C,OAAO,CAAC,UAAU,CAAC,IAAA,4CAAwB,GAAE,CAAC,CAAC;IAE/C,OAAO,OAAO,CAAC;AACjB,CAAC"}
|
package/dist/types.d.ts
CHANGED
|
@@ -2,6 +2,7 @@ export type JsonObject = Record<string, unknown>;
|
|
|
2
2
|
export type ExpertStatus = 'developing' | 'adjusting' | 'published' | 'paused';
|
|
3
3
|
export type EvalTaskStatus = 'pending' | 'in_progress' | 'failed' | 'changes_requested' | 'passed';
|
|
4
4
|
export type EvalRecordStatus = 'untested' | 'passed' | 'warning' | 'failed';
|
|
5
|
+
export type EvalCaseStatus = 'untested' | 'passed' | 'failed';
|
|
5
6
|
export type EvalRecordLevel = 'PASS' | 'WARNING' | 'FAIL' | string;
|
|
6
7
|
export type AgentAccountStatus = 'ready' | 'running';
|
|
7
8
|
export interface Expert {
|
|
@@ -77,7 +78,7 @@ export interface EvalCase {
|
|
|
77
78
|
artifactLink?: string | null;
|
|
78
79
|
taskLink?: string | null;
|
|
79
80
|
extras?: JsonObject | null;
|
|
80
|
-
status?:
|
|
81
|
+
status?: EvalCaseStatus;
|
|
81
82
|
executionStartTime?: string | null;
|
|
82
83
|
executionEndTime?: string | null;
|
|
83
84
|
createdAt: string;
|
|
@@ -93,6 +94,8 @@ export interface EvalTask {
|
|
|
93
94
|
resultSummary: EvalResultSummary;
|
|
94
95
|
caseSummary: EvalCaseSummary;
|
|
95
96
|
evaluator?: string | null;
|
|
97
|
+
packageReport?: string | null;
|
|
98
|
+
caseReport?: string | null;
|
|
96
99
|
startedAt?: string | null;
|
|
97
100
|
completedAt?: string | null;
|
|
98
101
|
createdAt: string;
|
|
@@ -158,6 +161,35 @@ export interface CreateEvalTaskPayload {
|
|
|
158
161
|
versionId: string;
|
|
159
162
|
name?: string | null;
|
|
160
163
|
}
|
|
164
|
+
export interface UpdateEvalTaskPayload {
|
|
165
|
+
packageReport?: string | null;
|
|
166
|
+
caseReport?: string | null;
|
|
167
|
+
}
|
|
168
|
+
export interface CreateEvalCasePayload {
|
|
169
|
+
evalTaskId: string;
|
|
170
|
+
instruction: string;
|
|
171
|
+
expectedResult?: string | null;
|
|
172
|
+
evaluationCriteria?: string | null;
|
|
173
|
+
actualResult?: string | null;
|
|
174
|
+
artifactLink?: string | null;
|
|
175
|
+
taskLink?: string | null;
|
|
176
|
+
extras?: JsonObject | null;
|
|
177
|
+
status?: EvalCaseStatus;
|
|
178
|
+
executionStartTime?: string | null;
|
|
179
|
+
executionEndTime?: string | null;
|
|
180
|
+
}
|
|
181
|
+
export interface UpdateEvalCasePayload {
|
|
182
|
+
instruction?: string;
|
|
183
|
+
expectedResult?: string | null;
|
|
184
|
+
evaluationCriteria?: string | null;
|
|
185
|
+
actualResult?: string | null;
|
|
186
|
+
artifactLink?: string | null;
|
|
187
|
+
taskLink?: string | null;
|
|
188
|
+
extras?: JsonObject | null;
|
|
189
|
+
status?: EvalCaseStatus;
|
|
190
|
+
executionStartTime?: string | null;
|
|
191
|
+
executionEndTime?: string | null;
|
|
192
|
+
}
|
|
161
193
|
export interface UpdateEvalRecordPayload {
|
|
162
194
|
status?: EvalRecordStatus;
|
|
163
195
|
comment?: string | null;
|
|
@@ -189,6 +221,11 @@ export interface ListEvalTasksParams {
|
|
|
189
221
|
export interface ListEvalRecordsParams {
|
|
190
222
|
evalTaskId: string;
|
|
191
223
|
}
|
|
224
|
+
export interface ListEvalCasesParams {
|
|
225
|
+
evalTaskId?: string;
|
|
226
|
+
expertId?: string;
|
|
227
|
+
status?: EvalCaseStatus;
|
|
228
|
+
}
|
|
192
229
|
export interface ListAgentAccountsParams {
|
|
193
230
|
status?: AgentAccountStatus;
|
|
194
231
|
owner?: string;
|
package/dist/types.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAIA,MAAM,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAEjD,MAAM,MAAM,YAAY,GAAG,YAAY,GAAG,WAAW,GAAG,WAAW,GAAG,QAAQ,CAAC;AAC/E,MAAM,MAAM,cAAc,GAAG,SAAS,GAAG,aAAa,GAAG,QAAQ,GAAG,mBAAmB,GAAG,QAAQ,CAAC;AACnG,MAAM,MAAM,gBAAgB,GAAG,UAAU,GAAG,QAAQ,GAAG,SAAS,GAAG,QAAQ,CAAC;AAC5E,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG,SAAS,GAAG,MAAM,GAAG,MAAM,CAAC;AACnE,MAAM,MAAM,kBAAkB,GAAG,OAAO,GAAG,SAAS,CAAC;AAErD,MAAM,WAAW,MAAM;IACrB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,YAAY,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,SAAS,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,SAAS,CAAC,EAAE,UAAU,EAAE,GAAG,IAAI,CAAC;IAChC,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,OAAO;IACtB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,SAAS,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,UAAU;IACzB,EAAE,EAAE,MAAM,CAAC;IACX,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,KAAK,CAAC,EAAE,eAAe,GAAG,IAAI,CAAC;IAC/B,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,MAAM,EAAE,gBAAgB,CAAC;IACzB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,MAAM,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,MAAM,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC;IAC3B,MAAM,CAAC,EAAE,
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAIA,MAAM,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;AAEjD,MAAM,MAAM,YAAY,GAAG,YAAY,GAAG,WAAW,GAAG,WAAW,GAAG,QAAQ,CAAC;AAC/E,MAAM,MAAM,cAAc,GAAG,SAAS,GAAG,aAAa,GAAG,QAAQ,GAAG,mBAAmB,GAAG,QAAQ,CAAC;AACnG,MAAM,MAAM,gBAAgB,GAAG,UAAU,GAAG,QAAQ,GAAG,SAAS,GAAG,QAAQ,CAAC;AAC5E,MAAM,MAAM,cAAc,GAAG,UAAU,GAAG,QAAQ,GAAG,QAAQ,CAAC;AAC9D,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG,SAAS,GAAG,MAAM,GAAG,MAAM,CAAC;AACnE,MAAM,MAAM,kBAAkB,GAAG,OAAO,GAAG,SAAS,CAAC;AAErD,MAAM,WAAW,MAAM;IACrB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,YAAY,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,SAAS,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,SAAS,CAAC,EAAE,UAAU,EAAE,GAAG,IAAI,CAAC;IAChC,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,OAAO;IACtB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,SAAS,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,UAAU;IACzB,EAAE,EAAE,MAAM,CAAC;IACX,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,KAAK,CAAC,EAAE,eAAe,GAAG,IAAI,CAAC;IAC/B,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,MAAM,EAAE,gBAAgB,CAAC;IACzB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,MAAM,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,MAAM,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC;IAC3B,MAAM,CAAC,EAAE,cAAc,CAAC;IACxB,kBAAkB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACnC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,cAAc,CAAC;IACvB,aAAa,EAAE,iBAAiB,CAAC;IACjC,WAAW,EAAE,eAAe,CAAC;IAC7B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,OAAO,CAAC,EAAE,UAAU,EAAE,CAAC;IACvB,KAAK,CAAC,EAAE,QAAQ,EAAE,CAAC;CACpB;AAED,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,MAAM,EAAE,kBAAkB,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB;AAMD,MAAM,WAAW,iBAAiB,CAAC,CAAC;IAClC,KAAK,EAAE,CAAC,EAAE,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,EAAE,MAAM,CAAC;CAClB;AAMD,MAAM,WAAW,SAAS;IACxB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,YAAY,GAAG,MAAM,GAAG,KAAK,GAAG,OAAO,GAAG,MAAM,CAAC;AAE7D,MAAM,WAAW,aAAa;IAC5B,MAAM,CAAC,EAAE,YAAY,CAAC;IACtB,KAAK,CAAC,EAAE,OAAO,CAAC;CACjB;AAED,MAAM,WAAW,gBAAgB;IAC/B,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,OAAO,CAAC;CACnB;AAMD,MAAM,WAAW,mBAAmB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,MAAM,CAAC,EAAE,YAAY,CAAC;CACvB;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,MAAM,CAAC,EAAE,YAAY,CAAC;CACvB;AAED,MAAM,WAAW,oBAAoB;IACnC,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,SAAS,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,SAAS,CAAC,EAAE,UAAU,EAAE,GAAG,IAAI,CAAC;CACjC;AAED,MAAM,WAAW,qBAAqB;IACpC,SAAS,EAAE,MAAM,CAAC;IAClB,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACtB;AAED,MAAM,WAAW,qBAAqB;IACpC,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC5B;AAED,MAAM,WAAW,qBAAqB;IACpC,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,kBAAkB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACnC,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,MAAM,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC;IAC3B,MAAM,CAAC,EAAE,cAAc,CAAC;IACxB,kBAAkB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACnC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAClC;AAED,MAAM,WAAW,qBAAqB;IACpC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,kBAAkB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACnC,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,MAAM,CAAC,EAAE,UAAU,GAAG,IAAI,CAAC;IAC3B,MAAM,CAAC,EAAE,cAAc,CAAC;IACxB,kBAAkB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACnC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAClC;AAED,MAAM,WAAW,uBAAuB;IACtC,MAAM,CAAC,EAAE,gBAAgB,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC1B;AAED,MAAM,WAAW,6BAA6B;IAC5C,OAAO,EAAE,KAAK,CAAC,uBAAuB,GAAG;QAAE,EAAE,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CAC1D;AAMD,MAAM,WAAW,iBAAiB;IAChC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,YAAY,CAAC;CACvB;AAED,MAAM,WAAW,kBAAkB;IACjC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,mBAAmB;IAClC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,cAAc,CAAC;CACzB;AAED,MAAM,WAAW,qBAAqB;IACpC,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,mBAAmB;IAClC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,MAAM,CAAC,EAAE,cAAc,CAAC;CACzB;AAED,MAAM,WAAW,uBAAuB;IACtC,MAAM,CAAC,EAAE,kBAAkB,CAAC;IAC5B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB"}
|
package/package.json
CHANGED
package/skill/SKILL.md
CHANGED
|
@@ -1,16 +1,47 @@
|
|
|
1
1
|
---
|
|
2
2
|
name: expert-eval-cli
|
|
3
|
-
version: "
|
|
3
|
+
version: "2.0.2"
|
|
4
4
|
description: >
|
|
5
|
-
ExpertEval CLI
|
|
6
|
-
|
|
7
|
-
|
|
5
|
+
ExpertEval CLI 工具技能：帮助用户安装、更新和使用 expert_eval_cli，完成 API Key 认证、
|
|
6
|
+
环境切换、专家管理、版本包上传与解析、评测任务创建/查询与报告链接维护、评测用例管理、评测记录单条或批量回写、
|
|
7
|
+
评测账号领取/释放等操作。适用于 ExpertEval、专家评估、评测任务、评测记录、版本管理、
|
|
8
|
+
agent.json/agent.yml/skills.json 上传、评测账号管理等场景。
|
|
8
9
|
---
|
|
9
10
|
|
|
10
11
|
# ExpertEval CLI 技能
|
|
11
12
|
|
|
12
13
|
本技能帮助用户和 AI Agent 使用 `expert_eval_cli` 命令行工具与 ExpertEval API 交互。
|
|
13
14
|
|
|
15
|
+
## 能力介绍
|
|
16
|
+
|
|
17
|
+
当用户需要操作 ExpertEval 平台时，使用本技能调用 `expert_eval` CLI 完成端到端任务。核心能力包括：
|
|
18
|
+
|
|
19
|
+
- **安装与升级**：安装、更新 `expert_eval_cli`，并校验 `expert_eval --version`。
|
|
20
|
+
- **认证管理**：通过环境变量或本地配置文件保存 API Key，支持查看认证状态与登出。
|
|
21
|
+
- **环境切换**：支持线上环境和开发环境 Base URL，通过环境变量或登录参数切换。
|
|
22
|
+
- **专家管理**：查询、创建、更新、删除专家；支持按关键词和状态筛选；可拉取专家原始 `agent.json` / `agent.yml`。
|
|
23
|
+
- **版本管理**：查询、获取、创建、删除版本；支持上传 ZIP 包，也支持直接传入解析后的 `agent.json`、`agent.yml`、`skills.json`。
|
|
24
|
+
- **评测任务管理**：查询、创建、获取详情、更新 Package/Case 评测报告飞书云文档链接、删除评测任务；详情包含版本、记录、用例与报告链接信息；可拉取任务对应的被测 Agent Card。
|
|
25
|
+
- **评测用例管理**：按评测任务、专家业务 ID、状态查询用例；创建、增量更新或删除用例；支持回填实际结果、产物链接、任务链接、扩展信息与执行时间。
|
|
26
|
+
- **评测记录回写**：按任务查询评测记录，单条更新状态/评论/证据，或通过 JSON/文件批量更新记录。
|
|
27
|
+
- **评测账号管理**：查询 ready/running 账号，领取评测账号并在完成后释放回 ready。
|
|
28
|
+
- **Agent 友好输出**：非 TTY 默认 JSON 输出，支持 `--format json|csv|table|auto` 与 `--quiet`，便于自动化解析和流水线集成。
|
|
29
|
+
|
|
30
|
+
## 适用场景
|
|
31
|
+
|
|
32
|
+
- 用户提到 ExpertEval、专家评估、专家/版本/评测任务/评测记录/评测账号。
|
|
33
|
+
- 用户需要创建或维护专家条目、上传专家版本包、发起评测、查看评测进度。
|
|
34
|
+
- 用户需要读取评测任务记录，并把评测结论、评论、证据链接回写到平台。
|
|
35
|
+
- 用户需要为自动化 Agent/CI 脚本配置 ExpertEval API Key、Base URL 或 JSON 输出。
|
|
36
|
+
|
|
37
|
+
## 使用原则
|
|
38
|
+
|
|
39
|
+
1. 优先使用 `--format json` 获取机器可读输出，除非用户明确需要表格或 CSV。
|
|
40
|
+
2. 需要认证时优先使用 `EXPERT_EVAL_API_KEY` 环境变量；交互式长期使用可通过 `expert_eval auth login` 写入本地配置。
|
|
41
|
+
3. 上传版本时，如已有本地 ZIP，优先使用 `--package-zip-file`；如已有解析产物，可直接使用 `--agent-json-file`、`--agent-yaml-file`、`--skill-json-file`。
|
|
42
|
+
4. 回写评测记录前，应先查询记录 ID 和现有状态，避免误更新。
|
|
43
|
+
5. 对批量更新，优先使用 `--records-file`，便于审计和重试。
|
|
44
|
+
|
|
14
45
|
## 安装与更新
|
|
15
46
|
|
|
16
47
|
### 首次安装
|
|
@@ -105,24 +136,54 @@ expert_eval --format json experts list
|
|
|
105
136
|
expert_eval experts list | cat
|
|
106
137
|
```
|
|
107
138
|
|
|
139
|
+
全局选项：
|
|
140
|
+
|
|
141
|
+
```bash
|
|
142
|
+
expert_eval --format json # 强制 JSON 输出
|
|
143
|
+
expert_eval --format csv # 强制 CSV 输出
|
|
144
|
+
expert_eval --format table # 强制表格输出
|
|
145
|
+
expert_eval --format auto # 自动选择输出格式（默认）
|
|
146
|
+
expert_eval -q # 静默模式，仅依赖 exit code
|
|
147
|
+
```
|
|
148
|
+
|
|
108
149
|
### 常用操作速查
|
|
109
150
|
|
|
110
151
|
| 操作 | 命令 |
|
|
111
152
|
|---|---|
|
|
112
153
|
| 列出专家 | `expert_eval experts list` |
|
|
154
|
+
| 按条件查询专家 | `expert_eval experts list --page 1 --page-size 50 --keyword "AI" --status published` |
|
|
155
|
+
| 获取专家详情 | `expert_eval experts get exp-001` |
|
|
113
156
|
| 创建专家 | `expert_eval experts create --expert-id exp-001 --name "AI安全专家" --owner 1847292357012580 --status developing` |
|
|
114
157
|
| 更新专家 | `expert_eval experts update exp-001 --name "新名称" --status published` |
|
|
158
|
+
| 清空专家 owner/description | `expert_eval experts update exp-001 --clear-owner --clear-description` |
|
|
115
159
|
| 查看专家卡/YAML | `expert_eval experts agent-card exp-001` / `expert_eval experts agent-yaml exp-001` |
|
|
116
160
|
| 删除专家 | `expert_eval experts delete <expert-id>` |
|
|
117
161
|
| 列出版本 | `expert_eval versions list --expert-id exp-001` |
|
|
118
|
-
|
|
|
162
|
+
| 获取版本详情 | `expert_eval versions get <version-id>` |
|
|
163
|
+
| 上传 ZIP 创建版本 | `expert_eval versions create --expert-id exp-001 --package-zip-file ./pkg.zip --package-name pkg.zip` |
|
|
164
|
+
| 用解析产物创建版本 | `expert_eval versions create --expert-id exp-001 --agent-json-file ./agent.json --agent-yaml-file ./agent.yml --skill-json-file ./skills.json` |
|
|
165
|
+
| 删除版本 | `expert_eval versions delete <version-id>` |
|
|
166
|
+
| 查询评测任务 | `expert_eval eval-tasks list --expert-id exp-001 --status in_progress` |
|
|
119
167
|
| 创建评测任务 | `expert_eval eval-tasks create --version-id v-1 --name "v1 评测"` |
|
|
120
168
|
| 获取任务详情 | `expert_eval eval-tasks get <task-id>` |
|
|
169
|
+
| 更新任务报告链接 | `expert_eval eval-tasks update <task-id> --package-report https://example.feishu.cn/docx/package-report --case-report https://example.feishu.cn/docx/case-report` |
|
|
170
|
+
| 清空任务报告链接 | `expert_eval eval-tasks update <task-id> --clear-package-report --clear-case-report` |
|
|
121
171
|
| 获取任务记录 | `expert_eval eval-tasks records <task-id>` |
|
|
172
|
+
| 获取任务 Agent Card | `expert_eval eval-tasks agent-card <task-id>` |
|
|
173
|
+
| 删除评测任务 | `expert_eval eval-tasks delete <task-id>` |
|
|
174
|
+
| 查询评测用例 | `expert_eval eval-cases list --eval-task-id task-1 --status untested` |
|
|
175
|
+
| 按专家查询用例 | `expert_eval eval-cases list --expert-id exp-001` |
|
|
176
|
+
| 创建评测用例 | `expert_eval eval-cases create --eval-task-id task-1 --instruction "查询年假政策" --expected-result "准确回答年假天数" --evaluation-criteria "答案符合政策" --status untested` |
|
|
177
|
+
| 更新评测用例 | `expert_eval eval-cases update <case-id> --status passed --actual-result "已完成" --artifact-link https://example.com/report.xlsx` |
|
|
178
|
+
| 清空用例字段 | `expert_eval eval-cases update <case-id> --clear-actual-result --clear-artifact-link --clear-extras` |
|
|
179
|
+
| 删除评测用例 | `expert_eval eval-cases delete <case-id>` |
|
|
122
180
|
| 列出评审记录 | `expert_eval eval-records list --eval-task-id task-1` |
|
|
123
181
|
| 修改评审记录 | `expert_eval eval-records update <record-id> --status passed --comment "通过"` |
|
|
182
|
+
| 清空记录评论/证据 | `expert_eval eval-records update <record-id> --clear-comment --clear-evidence` |
|
|
124
183
|
| 批量修改记录 | `expert_eval eval-records batch --records-file ./records.json` |
|
|
125
184
|
| 查询评测账号 | `expert_eval agent-accounts list --status ready` |
|
|
185
|
+
| 领取评测账号 | `expert_eval agent-accounts start agent-001` |
|
|
186
|
+
| 释放评测账号 | `expert_eval agent-accounts done agent-001` |
|
|
126
187
|
|
|
127
188
|
### 分页查询
|
|
128
189
|
|
|
@@ -130,6 +191,36 @@ expert_eval experts list | cat
|
|
|
130
191
|
|
|
131
192
|
```bash
|
|
132
193
|
expert_eval experts list --page 2 --page-size 20
|
|
194
|
+
expert_eval versions list --page 1 --page-size 50 --expert-id exp-001
|
|
195
|
+
expert_eval eval-tasks list --page 1 --page-size 50 --version-id v-1
|
|
196
|
+
```
|
|
197
|
+
|
|
198
|
+
### 状态枚举
|
|
199
|
+
|
|
200
|
+
- 专家状态：`developing`、`adjusting`、`published`、`paused`
|
|
201
|
+
- 评测任务状态：`pending`、`in_progress`、`failed`、`changes_requested`、`passed`
|
|
202
|
+
- 评测用例状态：`untested`、`passed`、`failed`
|
|
203
|
+
- 评测记录状态：`untested`、`passed`、`warning`、`failed`
|
|
204
|
+
- 评测账号状态：`ready`、`running`
|
|
205
|
+
|
|
206
|
+
### 批量回写记录格式
|
|
207
|
+
|
|
208
|
+
`eval-records batch` 支持 JSON 数组或 `{ "records": [...] }`：
|
|
209
|
+
|
|
210
|
+
```json
|
|
211
|
+
[
|
|
212
|
+
{
|
|
213
|
+
"id": "record-1",
|
|
214
|
+
"status": "passed",
|
|
215
|
+
"comment": "符合标准"
|
|
216
|
+
},
|
|
217
|
+
{
|
|
218
|
+
"id": "record-2",
|
|
219
|
+
"status": "failed",
|
|
220
|
+
"comment": "未满足验收标准",
|
|
221
|
+
"evidence": "https://example.com/evidence.png"
|
|
222
|
+
}
|
|
223
|
+
]
|
|
133
224
|
```
|
|
134
225
|
|
|
135
226
|
### 错误处理
|
|
@@ -82,6 +82,15 @@ Body: `{ versionId, name? }`
|
|
|
82
82
|
### `GET /openapi/eval-tasks/{id}`
|
|
83
83
|
Get task detail, including version raw content, all evaluation records and cases.
|
|
84
84
|
|
|
85
|
+
Task detail/list items may include report links:
|
|
86
|
+
- `packageReport` (string | null): Package evaluation report Lark document URL.
|
|
87
|
+
- `caseReport` (string | null): Case evaluation report Lark document URL.
|
|
88
|
+
|
|
89
|
+
### `PATCH /openapi/eval-tasks/{id}`
|
|
90
|
+
Update evaluation task report links. Only `packageReport` and `caseReport` are supported. Values must be valid `https` URLs; pass `null` or an empty string to clear a report link.
|
|
91
|
+
|
|
92
|
+
Body: `{ packageReport?, caseReport? }`
|
|
93
|
+
|
|
85
94
|
### `DELETE /openapi/eval-tasks/{id}`
|
|
86
95
|
Delete an evaluation task. Response: `{ id }`
|
|
87
96
|
|
|
@@ -91,6 +100,41 @@ Get all records for an evaluation task, including the original standard content
|
|
|
91
100
|
### `GET /openapi/eval-tasks/{id}/agent-card`
|
|
92
101
|
Get the raw agent card for the version evaluated by this task.
|
|
93
102
|
|
|
103
|
+
## Eval Cases
|
|
104
|
+
|
|
105
|
+
### `GET /openapi/eval-cases`
|
|
106
|
+
List evaluation cases.
|
|
107
|
+
|
|
108
|
+
Query params:
|
|
109
|
+
- `evalTaskId` (string, optional)
|
|
110
|
+
- `expertId` (string, optional; expert business ID / agent ID)
|
|
111
|
+
- `status` (`untested` | `passed` | `failed`, optional)
|
|
112
|
+
|
|
113
|
+
Response: `EvalCase[]`
|
|
114
|
+
|
|
115
|
+
### `POST /openapi/eval-cases`
|
|
116
|
+
Create an evaluation case. The API returns `404` if the evaluation task does not exist.
|
|
117
|
+
|
|
118
|
+
Body fields include:
|
|
119
|
+
- `evalTaskId` (string, required)
|
|
120
|
+
- `instruction` (string, required)
|
|
121
|
+
- `expectedResult` (string, optional)
|
|
122
|
+
- `evaluationCriteria` (string, optional)
|
|
123
|
+
- `actualResult` (string, optional)
|
|
124
|
+
- `artifactLink` (string, optional)
|
|
125
|
+
- `taskLink` (string, optional)
|
|
126
|
+
- `extras` (object, optional)
|
|
127
|
+
- `status` (`untested` | `passed` | `failed`, optional)
|
|
128
|
+
- `executionStartTime` / `executionEndTime` (string, optional)
|
|
129
|
+
|
|
130
|
+
### `PATCH /openapi/eval-cases/{id}`
|
|
131
|
+
Partially update an evaluation case. Only provided fields are updated; an empty body returns `400`.
|
|
132
|
+
|
|
133
|
+
Body: any subset of `instruction`, `expectedResult`, `evaluationCriteria`, `actualResult`, `artifactLink`, `taskLink`, `extras`, `status`, `executionStartTime`, `executionEndTime`.
|
|
134
|
+
|
|
135
|
+
### `DELETE /openapi/eval-cases/{id}`
|
|
136
|
+
Delete an evaluation case. Response: `{ id }`
|
|
137
|
+
|
|
94
138
|
## Eval Records
|
|
95
139
|
|
|
96
140
|
### `GET /openapi/eval-records`
|