@dalmia/calibrate-mcp 0.0.24 → 0.0.26
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/bin/mcp-server.js +170 -136
- package/bin/mcp-server.js.map +24 -24
- package/esm/funcs/agentTestsGetBenchmark.js +6 -1
- package/esm/funcs/agentTestsGetBenchmark.js.map +1 -1
- package/esm/funcs/agentTestsGetRun.js +6 -1
- package/esm/funcs/agentTestsGetRun.js.map +1 -1
- package/esm/funcs/annotationTasksGetSummary.js +2 -0
- package/esm/funcs/annotationTasksGetSummary.js.map +1 -1
- package/esm/funcs/annotationTasksSetEvaluators.d.ts +16 -0
- package/esm/funcs/annotationTasksSetEvaluators.d.ts.map +1 -0
- package/esm/funcs/{annotationTasksLinkEvaluator.js → annotationTasksSetEvaluators.js} +8 -8
- package/esm/funcs/{annotationTasksLinkEvaluator.js.map → annotationTasksSetEvaluators.js.map} +1 -1
- package/esm/funcs/evaluatorsGet.js +5 -1
- package/esm/funcs/evaluatorsGet.js.map +1 -1
- package/esm/landing-page.js +1 -1
- package/esm/lib/config.d.ts +2 -2
- package/esm/lib/config.js +2 -2
- package/esm/mcp-server/mcp-server.js +1 -1
- package/esm/mcp-server/server.js +3 -3
- package/esm/mcp-server/server.js.map +1 -1
- package/esm/mcp-server/tools/annotationTasksSetEvaluators.d.ts +7 -0
- package/esm/mcp-server/tools/annotationTasksSetEvaluators.d.ts.map +1 -0
- package/esm/mcp-server/tools/annotationTasksSetEvaluators.js +38 -0
- package/esm/mcp-server/tools/annotationTasksSetEvaluators.js.map +1 -0
- package/esm/models/{evaluatordetailresponse.d.ts → evaluatordetailresponsecompact.d.ts} +17 -17
- package/esm/models/evaluatordetailresponsecompact.d.ts.map +1 -0
- package/esm/models/{evaluatordetailresponse.js → evaluatordetailresponsecompact.js} +14 -13
- package/esm/models/evaluatordetailresponsecompact.js.map +1 -0
- package/esm/models/evaluatorsetrequest.d.ts +6 -0
- package/esm/models/evaluatorsetrequest.d.ts.map +1 -0
- package/esm/models/evaluatorsetrequest.js +9 -0
- package/esm/models/evaluatorsetrequest.js.map +1 -0
- package/esm/models/evaluatorsetresponse.d.ts +9 -0
- package/esm/models/evaluatorsetresponse.d.ts.map +1 -0
- package/esm/models/evaluatorsetresponse.js +12 -0
- package/esm/models/evaluatorsetresponse.js.map +1 -0
- package/esm/models/{evaluatorversionresponse.d.ts → evaluatorversioncompact.d.ts} +4 -4
- package/esm/models/evaluatorversioncompact.d.ts.map +1 -0
- package/esm/models/{evaluatorversionresponse.js → evaluatorversioncompact.js} +3 -3
- package/esm/models/evaluatorversioncompact.js.map +1 -0
- package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.d.ts +2 -0
- package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.d.ts.map +1 -1
- package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.js +2 -0
- package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.js.map +1 -1
- package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts +2 -0
- package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts.map +1 -1
- package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.js +2 -0
- package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.js.map +1 -1
- package/esm/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.d.ts +3 -2
- package/esm/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.d.ts.map +1 -1
- package/esm/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.js +3 -2
- package/esm/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.js.map +1 -1
- package/esm/models/settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.d.ts +13 -0
- package/esm/models/settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.d.ts.map +1 -0
- package/esm/models/settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.js +19 -0
- package/esm/models/settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.js.map +1 -0
- package/esm/models/tasksummaryannotationtaskstaskuuidsummarygetop.d.ts +2 -0
- package/esm/models/tasksummaryannotationtaskstaskuuidsummarygetop.d.ts.map +1 -1
- package/esm/models/tasksummaryannotationtaskstaskuuidsummarygetop.js +2 -0
- package/esm/models/tasksummaryannotationtaskstaskuuidsummarygetop.js.map +1 -1
- package/esm/models/tasksummaryresponse.d.ts +2 -2
- package/esm/models/tasksummaryresponse.d.ts.map +1 -1
- package/esm/models/tasksummaryresponse.js +3 -2
- package/esm/models/tasksummaryresponse.js.map +1 -1
- package/esm/tool-names.js +2 -2
- package/esm/tool-names.js.map +1 -1
- package/package.json +1 -1
- package/src/funcs/agentTestsGetBenchmark.ts +6 -1
- package/src/funcs/agentTestsGetRun.ts +6 -1
- package/src/funcs/annotationTasksGetSummary.ts +2 -0
- package/src/funcs/{annotationTasksLinkEvaluator.ts → annotationTasksSetEvaluators.ts} +11 -11
- package/src/funcs/evaluatorsGet.ts +5 -1
- package/src/landing-page.ts +1 -1
- package/src/lib/config.ts +2 -2
- package/src/mcp-server/mcp-server.ts +1 -1
- package/src/mcp-server/server.ts +3 -3
- package/src/mcp-server/tools/annotationTasksSetEvaluators.ts +48 -0
- package/src/models/{evaluatordetailresponse.ts → evaluatordetailresponsecompact.ts} +29 -28
- package/src/models/evaluatorsetrequest.ts +14 -0
- package/src/models/evaluatorsetresponse.ts +28 -0
- package/src/models/{evaluatorversionresponse.ts → evaluatorversioncompact.ts} +5 -5
- package/src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts +8 -0
- package/src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts +8 -0
- package/src/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.ts +9 -5
- package/src/models/settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.ts +42 -0
- package/src/models/tasksummaryannotationtaskstaskuuidsummarygetop.ts +8 -0
- package/src/models/tasksummaryresponse.ts +4 -5
- package/src/tool-names.ts +2 -2
- package/esm/funcs/annotationTasksLinkEvaluator.d.ts +0 -16
- package/esm/funcs/annotationTasksLinkEvaluator.d.ts.map +0 -1
- package/esm/mcp-server/tools/annotationTasksLinkEvaluator.d.ts +0 -7
- package/esm/mcp-server/tools/annotationTasksLinkEvaluator.d.ts.map +0 -1
- package/esm/mcp-server/tools/annotationTasksLinkEvaluator.js +0 -34
- package/esm/mcp-server/tools/annotationTasksLinkEvaluator.js.map +0 -1
- package/esm/models/evaluatordetailresponse.d.ts.map +0 -1
- package/esm/models/evaluatordetailresponse.js.map +0 -1
- package/esm/models/evaluatorlinkrequest.d.ts +0 -6
- package/esm/models/evaluatorlinkrequest.d.ts.map +0 -1
- package/esm/models/evaluatorlinkrequest.js +0 -9
- package/esm/models/evaluatorlinkrequest.js.map +0 -1
- package/esm/models/evaluatorlinkresponse.d.ts +0 -6
- package/esm/models/evaluatorlinkresponse.d.ts.map +0 -1
- package/esm/models/evaluatorlinkresponse.js +0 -8
- package/esm/models/evaluatorlinkresponse.js.map +0 -1
- package/esm/models/evaluatorversionresponse.d.ts.map +0 -1
- package/esm/models/evaluatorversionresponse.js.map +0 -1
- package/esm/models/linkevaluatortotaskannotationtaskstaskuuidevaluatorspostop.d.ts +0 -13
- package/esm/models/linkevaluatortotaskannotationtaskstaskuuidevaluatorspostop.d.ts.map +0 -1
- package/esm/models/linkevaluatortotaskannotationtaskstaskuuidevaluatorspostop.js +0 -18
- package/esm/models/linkevaluatortotaskannotationtaskstaskuuidevaluatorspostop.js.map +0 -1
- package/src/mcp-server/tools/annotationTasksLinkEvaluator.ts +0 -44
- package/src/models/evaluatorlinkrequest.ts +0 -14
- package/src/models/evaluatorlinkresponse.ts +0 -12
- package/src/models/linkevaluatortotaskannotationtaskstaskuuidevaluatorspostop.ts +0 -42
|
@@ -0,0 +1,9 @@
|
|
|
1
|
+
import * as z from "zod";
|
|
2
|
+
export type EvaluatorSetResponse = {
|
|
3
|
+
message: string;
|
|
4
|
+
evaluator_ids: Array<string>;
|
|
5
|
+
linked: Array<string>;
|
|
6
|
+
unlinked: Array<string>;
|
|
7
|
+
};
|
|
8
|
+
export declare const EvaluatorSetResponse$zodSchema: z.ZodType<EvaluatorSetResponse>;
|
|
9
|
+
//# sourceMappingURL=evaluatorsetresponse.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluatorsetresponse.d.ts","sourceRoot":"","sources":["../../src/models/evaluatorsetresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,MAAM,MAAM,oBAAoB,GAAG;IACjC,OAAO,EAAE,MAAM,CAAC;IAChB,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IAC7B,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IACtB,QAAQ,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;CACzB,CAAC;AAEF,eAAO,MAAM,8BAA8B,EAAE,CAAC,CAAC,OAAO,CAAC,oBAAoB,CAcvE,CAAC"}
|
|
@@ -0,0 +1,12 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
import * as z from "zod";
|
|
5
|
+
export const EvaluatorSetResponse$zodSchema = z
|
|
6
|
+
.object({
|
|
7
|
+
evaluator_ids: z.array(z.string()).describe("The task's evaluator IDs after the update, in display order"),
|
|
8
|
+
linked: z.array(z.string()).describe("Evaluator IDs newly linked by this request"),
|
|
9
|
+
message: z.string().describe("Confirmation that the task's evaluators were updated"),
|
|
10
|
+
unlinked: z.array(z.string()).describe("Evaluator IDs unlinked by this request"),
|
|
11
|
+
});
|
|
12
|
+
//# sourceMappingURL=evaluatorsetresponse.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluatorsetresponse.js","sourceRoot":"","sources":["../../src/models/evaluatorsetresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AASzB,MAAM,CAAC,MAAM,8BAA8B,GAAoC,CAAC;KAC7E,MAAM,CAAC;IACN,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,CACzC,6DAA6D,CAC9D;IACD,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,CAClC,4CAA4C,CAC7C;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC1B,sDAAsD,CACvD;IACD,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,CACpC,wCAAwC,CACzC;CACF,CAAC,CAAC"}
|
|
@@ -1,14 +1,14 @@
|
|
|
1
1
|
import * as z from "zod";
|
|
2
2
|
import { OutputConfig } from "./outputconfig.js";
|
|
3
3
|
import { VariableSpec } from "./variablespec.js";
|
|
4
|
-
export type
|
|
4
|
+
export type EvaluatorVersionCompact = {
|
|
5
5
|
uuid: string;
|
|
6
6
|
version_number: number;
|
|
7
7
|
judge_model: string;
|
|
8
|
-
system_prompt
|
|
8
|
+
system_prompt?: string | null | undefined;
|
|
9
9
|
output_config?: OutputConfig | null | undefined;
|
|
10
10
|
variables?: Array<VariableSpec> | null | undefined;
|
|
11
11
|
created_at: string;
|
|
12
12
|
};
|
|
13
|
-
export declare const
|
|
14
|
-
//# sourceMappingURL=
|
|
13
|
+
export declare const EvaluatorVersionCompact$zodSchema: z.ZodType<EvaluatorVersionCompact>;
|
|
14
|
+
//# sourceMappingURL=evaluatorversioncompact.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluatorversioncompact.d.ts","sourceRoot":"","sources":["../../src/models/evaluatorversioncompact.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,YAAY,EAA0B,MAAM,mBAAmB,CAAC;AACzE,OAAO,EAAE,YAAY,EAA0B,MAAM,mBAAmB,CAAC;AAEzE,MAAM,MAAM,uBAAuB,GAAG;IACpC,IAAI,EAAE,MAAM,CAAC;IACb,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC1C,aAAa,CAAC,EAAE,YAAY,GAAG,IAAI,GAAG,SAAS,CAAC;IAChD,SAAS,CAAC,EAAE,KAAK,CAAC,YAAY,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACnD,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,eAAO,MAAM,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,uBAAuB,CAqBvB,CAAC"}
|
|
@@ -4,13 +4,13 @@
|
|
|
4
4
|
import * as z from "zod";
|
|
5
5
|
import { OutputConfig$zodSchema } from "./outputconfig.js";
|
|
6
6
|
import { VariableSpec$zodSchema } from "./variablespec.js";
|
|
7
|
-
export const
|
|
7
|
+
export const EvaluatorVersionCompact$zodSchema = z.object({
|
|
8
8
|
created_at: z.string().describe("When the version was created (ISO 8601 UTC)"),
|
|
9
9
|
judge_model: z.string().describe("The model that runs the judge, named the way its provider does, for example `openai/gpt-4.1` or `anthropic/claude-sonnet-4`"),
|
|
10
10
|
output_config: OutputConfig$zodSchema.nullable().optional().describe("The scale points and their labels. Required for a `rating` evaluator. A `binary` evaluator uses the default Correct/Wrong labels unless you set your own"),
|
|
11
|
-
system_prompt: z.string().describe("Judge system prompt, with `{{\"{{\"}}variable}}` placeholders unrendered"),
|
|
11
|
+
system_prompt: z.string().nullable().optional().describe("Judge system prompt, with `{{\"{{\"}}variable}}` placeholders unrendered"),
|
|
12
12
|
uuid: z.string().describe("Version ID"),
|
|
13
13
|
variables: z.array(VariableSpec$zodSchema).nullable().optional().describe("Declared prompt variables"),
|
|
14
14
|
version_number: z.int().describe("The version's number. The first version is 1, and it goes up by one for each new version of the evaluator"),
|
|
15
15
|
});
|
|
16
|
-
//# sourceMappingURL=
|
|
16
|
+
//# sourceMappingURL=evaluatorversioncompact.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluatorversioncompact.js","sourceRoot":"","sources":["../../src/models/evaluatorversioncompact.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAgB,sBAAsB,EAAE,MAAM,mBAAmB,CAAC;AACzE,OAAO,EAAgB,sBAAsB,EAAE,MAAM,mBAAmB,CAAC;AAYzE,MAAM,CAAC,MAAM,iCAAiC,GAE1C,CAAC,CAAC,MAAM,CAAC;IACX,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,6CAA6C,CAC9C;IACD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC9B,6HAA6H,CAC9H;IACD,aAAa,EAAE,sBAAsB,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClE,0JAA0J,CAC3J;IACD,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtD,0EAA0E,CAC3E;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,YAAY,CAAC;IACvC,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACvE,2BAA2B,CAC5B;IACD,cAAc,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,CAC9B,2GAA2G,CAC5G;CACF,CAAC,CAAC"}
|
|
@@ -3,6 +3,8 @@ import { HTTPValidationError } from "./httpvalidationerror.js";
|
|
|
3
3
|
import { TestRunStatusResponse } from "./testrunstatusresponse.js";
|
|
4
4
|
export type GetAgentTestRunStatusAgentTestsRunTaskIdGetRequest = {
|
|
5
5
|
task_id: string;
|
|
6
|
+
only_failed?: boolean | undefined;
|
|
7
|
+
compact?: boolean | undefined;
|
|
6
8
|
xAPIKey?: string | null | undefined;
|
|
7
9
|
};
|
|
8
10
|
export declare const GetAgentTestRunStatusAgentTestsRunTaskIdGetRequest$zodSchema: z.ZodType<GetAgentTestRunStatusAgentTestsRunTaskIdGetRequest>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getagenttestrunstatusagenttestsruntaskidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,qBAAqB,EAEtB,MAAM,4BAA4B,CAAC;AAEpC,MAAM,MAAM,kDAAkD,GAAG;IAC/D,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,4DAA4D,EACvE,CAAC,CAAC,OAAO,CAAC,kDAAkD,
|
|
1
|
+
{"version":3,"file":"getagenttestrunstatusagenttestsruntaskidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,qBAAqB,EAEtB,MAAM,4BAA4B,CAAC;AAEpC,MAAM,MAAM,kDAAkD,GAAG;IAC/D,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAClC,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,4DAA4D,EACvE,CAAC,CAAC,OAAO,CAAC,kDAAkD,CAS1D,CAAC;AAEL,MAAM,MAAM,mDAAmD,GAC3D,qBAAqB,GACrB,mBAAmB,CAAC;AAExB,eAAO,MAAM,6DAA6D,EACxE,CAAC,CAAC,OAAO,CAAC,mDAAmD,CAG3D,CAAC"}
|
|
@@ -5,6 +5,8 @@ import * as z from "zod";
|
|
|
5
5
|
import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
|
|
6
6
|
import { TestRunStatusResponse$zodSchema, } from "./testrunstatusresponse.js";
|
|
7
7
|
export const GetAgentTestRunStatusAgentTestsRunTaskIdGetRequest$zodSchema = z.object({
|
|
8
|
+
compact: z.boolean().default(false).describe("Return a compact response that omits heavy detail fields (`results.output`, `results.test_case`, `results.judge_results`, `results.reasoning`, `evaluators.output_config`), keeping only the lightweight decision fields. Omit for full detail"),
|
|
9
|
+
only_failed: z.boolean().default(false).describe("Return only failing test cases. Omit to return every case"),
|
|
8
10
|
task_id: z.string().describe("Test run to poll for status and results"),
|
|
9
11
|
xAPIKey: z.string().nullable().optional(),
|
|
10
12
|
});
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getagenttestrunstatusagenttestsruntaskidgetop.js","sourceRoot":"","sources":["../../src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,+BAA+B,GAChC,MAAM,4BAA4B,CAAC;
|
|
1
|
+
{"version":3,"file":"getagenttestrunstatusagenttestsruntaskidgetop.js","sourceRoot":"","sources":["../../src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,+BAA+B,GAChC,MAAM,4BAA4B,CAAC;AASpC,MAAM,CAAC,MAAM,4DAA4D,GACP,CAAC,CAAC,MAAM,CAAC;IACvE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gPAAgP,CACjP;IACD,WAAW,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC9C,2DAA2D,CAC5D;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yCAAyC,CAAC;IACvE,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAML,MAAM,CAAC,MAAM,6DAA6D,GACP,CAAC,CAAC,KAAK,CAAC;IACvE,+BAA+B;IAC/B,6BAA6B;CAC9B,CAAC,CAAC"}
|
|
@@ -3,6 +3,8 @@ import { BenchmarkStatusResponse } from "./benchmarkstatusresponse.js";
|
|
|
3
3
|
import { HTTPValidationError } from "./httpvalidationerror.js";
|
|
4
4
|
export type GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetRequest = {
|
|
5
5
|
task_id: string;
|
|
6
|
+
only_failed?: boolean | undefined;
|
|
7
|
+
compact?: boolean | undefined;
|
|
6
8
|
xAPIKey?: string | null | undefined;
|
|
7
9
|
};
|
|
8
10
|
export declare const GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetRequest$zodSchema: z.ZodType<GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetRequest>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,uBAAuB,EAExB,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC,MAAM,MAAM,qDAAqD,GAAG;IAClE,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,+DAA+D,EAC1E,CAAC,CAAC,OAAO,CAAC,qDAAqD,
|
|
1
|
+
{"version":3,"file":"getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,uBAAuB,EAExB,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC,MAAM,MAAM,qDAAqD,GAAG;IAClE,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAClC,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,+DAA+D,EAC1E,CAAC,CAAC,OAAO,CAAC,qDAAqD,CAW7D,CAAC;AAEL,MAAM,MAAM,sDAAsD,GAC9D,uBAAuB,GACvB,mBAAmB,CAAC;AAExB,eAAO,MAAM,gEAAgE,EAC3E,CAAC,CAAC,OAAO,CAAC,sDAAsD,CAG9D,CAAC"}
|
|
@@ -5,6 +5,8 @@ import * as z from "zod";
|
|
|
5
5
|
import { BenchmarkStatusResponse$zodSchema, } from "./benchmarkstatusresponse.js";
|
|
6
6
|
import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
|
|
7
7
|
export const GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetRequest$zodSchema = z.object({
|
|
8
|
+
compact: z.boolean().default(false).describe("Return a compact response that omits heavy detail fields (`model_results.test_results`, `evaluators.output_config`), keeping only the lightweight decision fields. Omit for full detail"),
|
|
9
|
+
only_failed: z.boolean().default(false).describe("Return only failing test cases for each model. Omit to return every case"),
|
|
8
10
|
task_id: z.string().describe("Benchmark run to poll for status and results"),
|
|
9
11
|
xAPIKey: z.string().nullable().optional(),
|
|
10
12
|
});
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getbenchmarkstatusagenttestsbenchmarktaskidgetop.js","sourceRoot":"","sources":["../../src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,iCAAiC,GAClC,MAAM,8BAA8B,CAAC;AACtC,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;
|
|
1
|
+
{"version":3,"file":"getbenchmarkstatusagenttestsbenchmarktaskidgetop.js","sourceRoot":"","sources":["../../src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,iCAAiC,GAClC,MAAM,8BAA8B,CAAC;AACtC,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AASlC,MAAM,CAAC,MAAM,+DAA+D,GACP,CAAC,CAAC,MAAM,CAAC;IAC1E,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,yLAAyL,CAC1L;IACD,WAAW,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC9C,0EAA0E,CAC3E;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC1B,8CAA8C,CAC/C;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAML,MAAM,CAAC,MAAM,gEAAgE,GACP,CAAC,CAAC,KAAK,CAAC;IAC1E,iCAAiC;IACjC,6BAA6B;CAC9B,CAAC,CAAC"}
|
|
@@ -1,11 +1,12 @@
|
|
|
1
1
|
import * as z from "zod";
|
|
2
|
-
import {
|
|
2
|
+
import { EvaluatorDetailResponseCompact } from "./evaluatordetailresponsecompact.js";
|
|
3
3
|
import { HTTPValidationError } from "./httpvalidationerror.js";
|
|
4
4
|
export type GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest = {
|
|
5
5
|
evaluator_uuid: string;
|
|
6
|
+
compact?: boolean | undefined;
|
|
6
7
|
xAPIKey?: string | null | undefined;
|
|
7
8
|
};
|
|
8
9
|
export declare const GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest$zodSchema: z.ZodType<GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest>;
|
|
9
|
-
export type GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetResponse =
|
|
10
|
+
export type GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetResponse = EvaluatorDetailResponseCompact | HTTPValidationError;
|
|
10
11
|
export declare const GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetResponse$zodSchema: z.ZodType<GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetResponse>;
|
|
11
12
|
//# sourceMappingURL=getevaluatorendpointevaluatorsevaluatoruuidgetop.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getevaluatorendpointevaluatorsevaluatoruuidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,
|
|
1
|
+
{"version":3,"file":"getevaluatorendpointevaluatorsevaluatoruuidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,8BAA8B,EAE/B,MAAM,qCAAqC,CAAC;AAC7C,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC,MAAM,MAAM,qDAAqD,GAAG;IAClE,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,+DAA+D,EAC1E,CAAC,CAAC,OAAO,CAAC,qDAAqD,CAM7D,CAAC;AAEL,MAAM,MAAM,sDAAsD,GAC9D,8BAA8B,GAC9B,mBAAmB,CAAC;AAExB,eAAO,MAAM,gEAAgE,EAC3E,CAAC,CAAC,OAAO,CAAC,sDAAsD,CAG9D,CAAC"}
|
|
@@ -2,14 +2,15 @@
|
|
|
2
2
|
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
3
|
*/
|
|
4
4
|
import * as z from "zod";
|
|
5
|
-
import {
|
|
5
|
+
import { EvaluatorDetailResponseCompact$zodSchema, } from "./evaluatordetailresponsecompact.js";
|
|
6
6
|
import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
|
|
7
7
|
export const GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest$zodSchema = z.object({
|
|
8
|
+
compact: z.boolean().default(false).describe("Return a compact response that omits heavy detail fields (`versions.system_prompt`, `versions.output_config`, `versions.variables`), keeping only the lightweight decision fields. Omit for full detail"),
|
|
8
9
|
evaluator_uuid: z.string().describe("Evaluator to retrieve"),
|
|
9
10
|
xAPIKey: z.string().nullable().optional(),
|
|
10
11
|
});
|
|
11
12
|
export const GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetResponse$zodSchema = z.union([
|
|
12
|
-
|
|
13
|
+
EvaluatorDetailResponseCompact$zodSchema,
|
|
13
14
|
HTTPValidationError$zodSchema,
|
|
14
15
|
]);
|
|
15
16
|
//# sourceMappingURL=getevaluatorendpointevaluatorsevaluatoruuidgetop.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getevaluatorendpointevaluatorsevaluatoruuidgetop.js","sourceRoot":"","sources":["../../src/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,
|
|
1
|
+
{"version":3,"file":"getevaluatorendpointevaluatorsevaluatoruuidgetop.js","sourceRoot":"","sources":["../../src/models/getevaluatorendpointevaluatorsevaluatoruuidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,wCAAwC,GACzC,MAAM,qCAAqC,CAAC;AAC7C,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAQlC,MAAM,CAAC,MAAM,+DAA+D,GACP,CAAC,CAAC,MAAM,CAAC;IAC1E,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,yMAAyM,CAC1M;IACD,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,uBAAuB,CAAC;IAC5D,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAML,MAAM,CAAC,MAAM,gEAAgE,GACP,CAAC,CAAC,KAAK,CAAC;IAC1E,wCAAwC;IACxC,6BAA6B;CAC9B,CAAC,CAAC"}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
import * as z from "zod";
|
|
2
|
+
import { EvaluatorSetRequest } from "./evaluatorsetrequest.js";
|
|
3
|
+
import { EvaluatorSetResponse } from "./evaluatorsetresponse.js";
|
|
4
|
+
import { HTTPValidationError } from "./httpvalidationerror.js";
|
|
5
|
+
export type SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutRequest = {
|
|
6
|
+
task_uuid: string;
|
|
7
|
+
xAPIKey?: string | null | undefined;
|
|
8
|
+
body: EvaluatorSetRequest;
|
|
9
|
+
};
|
|
10
|
+
export declare const SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutRequest$zodSchema: z.ZodType<SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutRequest>;
|
|
11
|
+
export type SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutResponse = EvaluatorSetResponse | HTTPValidationError;
|
|
12
|
+
export declare const SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutResponse$zodSchema: z.ZodType<SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutResponse>;
|
|
13
|
+
//# sourceMappingURL=settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.d.ts","sourceRoot":"","sources":["../../src/models/settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,oBAAoB,EAErB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC,MAAM,MAAM,4DAA4D,GAAG;IACzE,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,IAAI,EAAE,mBAAmB,CAAC;CAC3B,CAAC;AAEF,eAAO,MAAM,sEAAsE,EACjF,CAAC,CAAC,OAAO,CAAC,4DAA4D,CAKlE,CAAC;AAEP,MAAM,MAAM,6DAA6D,GACrE,oBAAoB,GACpB,mBAAmB,CAAC;AAExB,eAAO,MAAM,uEAAuE,EAClF,CAAC,CAAC,OAAO,CAAC,6DAA6D,CAInE,CAAC"}
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
import * as z from "zod";
|
|
5
|
+
import { EvaluatorSetRequest$zodSchema, } from "./evaluatorsetrequest.js";
|
|
6
|
+
import { EvaluatorSetResponse$zodSchema, } from "./evaluatorsetresponse.js";
|
|
7
|
+
import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
|
|
8
|
+
export const SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutRequest$zodSchema = z
|
|
9
|
+
.object({
|
|
10
|
+
body: EvaluatorSetRequest$zodSchema,
|
|
11
|
+
task_uuid: z.string().describe("Annotation task to act on"),
|
|
12
|
+
xAPIKey: z.string().nullable().optional(),
|
|
13
|
+
});
|
|
14
|
+
export const SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutResponse$zodSchema = z
|
|
15
|
+
.union([
|
|
16
|
+
EvaluatorSetResponse$zodSchema,
|
|
17
|
+
HTTPValidationError$zodSchema,
|
|
18
|
+
]);
|
|
19
|
+
//# sourceMappingURL=settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.js","sourceRoot":"","sources":["../../src/models/settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,8BAA8B,GAC/B,MAAM,2BAA2B,CAAC;AACnC,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAQlC,MAAM,CAAC,MAAM,sEAAsE,GACP,CAAC;KACxE,MAAM,CAAC;IACN,IAAI,EAAE,6BAA6B;IACnC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2BAA2B,CAAC;IAC3D,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAMP,MAAM,CAAC,MAAM,uEAAuE,GACP,CAAC;KACzE,KAAK,CAAC;IACL,8BAA8B;IAC9B,6BAA6B;CAC9B,CAAC,CAAC"}
|
|
@@ -21,11 +21,13 @@ export type TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest = {
|
|
|
21
21
|
task_uuid: string;
|
|
22
22
|
item_id?: string | null | undefined;
|
|
23
23
|
live_only?: boolean | undefined;
|
|
24
|
+
disagreement_only?: boolean | undefined;
|
|
24
25
|
q?: string | null | undefined;
|
|
25
26
|
sort_by?: string | undefined;
|
|
26
27
|
order?: Order | undefined;
|
|
27
28
|
limit?: number | undefined;
|
|
28
29
|
offset?: number | undefined;
|
|
30
|
+
compact?: boolean | undefined;
|
|
29
31
|
xAPIKey?: string | null | undefined;
|
|
30
32
|
};
|
|
31
33
|
export declare const TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest$zodSchema: z.ZodType<TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"tasksummaryannotationtaskstaskuuidsummarygetop.d.ts","sourceRoot":"","sources":["../../src/models/tasksummaryannotationtaskstaskuuidsummarygetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,eAAO,MAAM,KAAK;;;CAGR,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,KAAK,GAAG,UAAU,CAAC,OAAO,KAAK,CAAC,CAAC;AAE7C,eAAO,MAAM,eAAe;;;EAGC,CAAC;AAE9B,MAAM,MAAM,mDAAmD,GAAG;IAChE,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,SAAS,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAChC,CAAC,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7B,KAAK,CAAC,EAAE,KAAK,GAAG,SAAS,CAAC;IAC1B,KAAK,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,6DAA6D,EACxE,CAAC,CAAC,OAAO,CAAC,mDAAmD,
|
|
1
|
+
{"version":3,"file":"tasksummaryannotationtaskstaskuuidsummarygetop.d.ts","sourceRoot":"","sources":["../../src/models/tasksummaryannotationtaskstaskuuidsummarygetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,eAAO,MAAM,KAAK;;;CAGR,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,KAAK,GAAG,UAAU,CAAC,OAAO,KAAK,CAAC,CAAC;AAE7C,eAAO,MAAM,eAAe;;;EAGC,CAAC;AAE9B,MAAM,MAAM,mDAAmD,GAAG;IAChE,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,SAAS,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAChC,iBAAiB,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IACxC,CAAC,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7B,KAAK,CAAC,EAAE,KAAK,GAAG,SAAS,CAAC;IAC1B,KAAK,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,6DAA6D,EACxE,CAAC,CAAC,OAAO,CAAC,mDAAmD,CA0B3D,CAAC;AAEL,MAAM,MAAM,oDAAoD,GAC5D,mBAAmB,GACnB,mBAAmB,CAAC;AAExB,eAAO,MAAM,8DAA8D,EACzE,CAAC,CAAC,OAAO,CAAC,oDAAoD,CAG5D,CAAC"}
|
|
@@ -16,6 +16,8 @@ export const Order$zodSchema = z.enum([
|
|
|
16
16
|
"desc",
|
|
17
17
|
]).describe("Sort direction");
|
|
18
18
|
export const TaskSummaryAnnotationTasksTaskUuidSummaryGetRequest$zodSchema = z.object({
|
|
19
|
+
compact: z.boolean().default(false).describe("Return a compact response that omits heavy detail fields (`rows.payload`, `rows.evaluator_reasoning`, `rows.annotations.reasoning`, `evaluators.versions.system_prompt`, `evaluators.versions.output_config`, `evaluators.versions.variables`, `item_comments`), keeping only the lightweight decision fields. Omit for full detail"),
|
|
20
|
+
disagreement_only: z.boolean().default(false).describe("When true, keep only rows where the evaluator disagreed with at least one annotator"),
|
|
19
21
|
item_id: z.string().describe("Filter rows to a single item. The full task-wide annotator union is still returned in `annotators`").nullable().optional(),
|
|
20
22
|
limit: z.int().default(50).describe("Maximum number of items to return"),
|
|
21
23
|
live_only: z.boolean().default(false).describe("When true, emit only one row for each (item, evaluator) pair using the evaluator's live version. Versions other than the live one that have runs are excluded"),
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"tasksummaryannotationtaskstaskuuidsummarygetop.js","sourceRoot":"","sources":["../../src/models/tasksummaryannotationtaskstaskuuidsummarygetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,MAAM,CAAC,MAAM,KAAK,GAAG;IACnB,GAAG,EAAE,KAAK;IACV,IAAI,EAAE,MAAM;CACJ,CAAC;AAMX,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,IAAI,CAAC;IACpC,KAAK;IACL,MAAM;CACP,CAAC,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC;
|
|
1
|
+
{"version":3,"file":"tasksummaryannotationtaskstaskuuidsummarygetop.js","sourceRoot":"","sources":["../../src/models/tasksummaryannotationtaskstaskuuidsummarygetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,MAAM,CAAC,MAAM,KAAK,GAAG;IACnB,GAAG,EAAE,KAAK;IACV,IAAI,EAAE,MAAM;CACJ,CAAC;AAMX,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,IAAI,CAAC;IACpC,KAAK;IACL,MAAM;CACP,CAAC,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC;AAgB9B,MAAM,CAAC,MAAM,6DAA6D,GACP,CAAC,CAAC,MAAM,CAAC;IACxE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,qUAAqU,CACtU;IACD,iBAAiB,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CACpD,qFAAqF,CACtF;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC1B,oGAAoG,CACrG,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvB,KAAK,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,QAAQ,CAAC,mCAAmC,CAAC;IACxE,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,+JAA+J,CAChK;IACD,MAAM,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,QAAQ,CACjC,kDAAkD,CACnD;IACD,KAAK,EAAE,eAAe,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IACjE,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACpB,uEAAuE,CACxE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,QAAQ,CAChD,0BAA0B,CAC3B;IACD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2BAA2B,CAAC;IAC3D,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAML,MAAM,CAAC,MAAM,8DAA8D,GACP,CAAC,CAAC,KAAK,CAAC;IACxE,6BAA6B;IAC7B,6BAA6B;CAC9B,CAAC,CAAC"}
|
|
@@ -32,9 +32,9 @@ export type TaskSummaryResponse = {
|
|
|
32
32
|
rows: Array<{
|
|
33
33
|
[k: string]: any;
|
|
34
34
|
}>;
|
|
35
|
-
item_comments
|
|
35
|
+
item_comments?: {
|
|
36
36
|
[k: string]: any;
|
|
37
|
-
};
|
|
37
|
+
} | null | undefined;
|
|
38
38
|
pagination: PaginationMeta;
|
|
39
39
|
AdditionalProperties?: {
|
|
40
40
|
[k: string]: any;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"tasksummaryresponse.d.ts","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,cAAc,EAA4B,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,eAAO,MAAM,QAAQ;;;;;CAKX,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,QAAQ,GAAG,UAAU,CAAC,OAAO,QAAQ,CAAC,CAAC;AAEnD,eAAO,MAAM,kBAAkB;;;;;EAKA,CAAC;AAEhC,MAAM,MAAM,mBAAmB,GAAG;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,QAAQ,CAAC;IACpB,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,IAAI,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IAClC,aAAa,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC;
|
|
1
|
+
{"version":3,"file":"tasksummaryresponse.d.ts","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,cAAc,EAA4B,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,eAAO,MAAM,QAAQ;;;;;CAKX,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,QAAQ,GAAG,UAAU,CAAC,OAAO,QAAQ,CAAC,CAAC;AAEnD,eAAO,MAAM,kBAAkB;;;;;EAKA,CAAC;AAEhC,MAAM,MAAM,mBAAmB,GAAG;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,QAAQ,CAAC;IACpB,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,IAAI,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IAClC,aAAa,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,UAAU,EAAE,cAAc,CAAC;IAC3B,oBAAoB,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC;CAC7C,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,CAoBtE,CAAC;AAEJ,eAAO,MAAM,qCAAqC,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAe/D,CAAC"}
|
|
@@ -22,7 +22,8 @@ export const TaskType$zodSchema = z.enum([
|
|
|
22
22
|
export const TaskSummaryResponse$zodSchema = collectExtraKeys$(z.object({
|
|
23
23
|
annotators: z.array(z.record(z.string(), z.any())).describe("The annotators who labelled or commented on items in the task, stable across pages"),
|
|
24
24
|
evaluators: z.array(z.record(z.string(), z.any())).describe("The task's evaluators, stable across pages, each with its live rubric"),
|
|
25
|
-
item_comments: z.record(z.string(), z.any()).
|
|
25
|
+
item_comments: z.record(z.string(), z.any()).nullable().optional()
|
|
26
|
+
.describe("Annotator comments, keyed by item ID"),
|
|
26
27
|
pagination: PaginationMeta$zodSchema,
|
|
27
28
|
rows: z.array(z.record(z.string(), z.any())).describe("One row per item, evaluator, and version on this page, with the evaluator's value and each annotator's label"),
|
|
28
29
|
task_id: z.string().describe("ID of the task"),
|
|
@@ -32,7 +33,7 @@ export const TaskSummaryResponse$zodSchemaOutbound = z.object({
|
|
|
32
33
|
AdditionalProperties: z.record(z.string(), z.any()).optional(),
|
|
33
34
|
annotators: z.array(z.record(z.string(), z.any())),
|
|
34
35
|
evaluators: z.array(z.record(z.string(), z.any())),
|
|
35
|
-
item_comments: z.record(z.string(), z.any()),
|
|
36
|
+
item_comments: z.nullable(z.record(z.string(), z.any())).optional(),
|
|
36
37
|
pagination: PaginationMeta$zodSchema,
|
|
37
38
|
rows: z.array(z.record(z.string(), z.any())),
|
|
38
39
|
task_id: z.string(),
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"tasksummaryresponse.js","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,gBAAgB,IAAI,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAE1E,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,MAAM,CAAC,MAAM,QAAQ,GAAG;IACtB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,UAAU,EAAE,aAAa;IACzB,YAAY,EAAE,cAAc;CACpB,CAAC;AAMX,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,IAAI,CAAC;IACvC,KAAK;IACL,KAAK;IACL,aAAa;IACb,cAAc;CACf,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;AAahC,MAAM,CAAC,MAAM,6BAA6B,GACxC,iBAAiB,CACf,CAAC,CAAC,MAAM,CAAC;IACP,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,oFAAoF,CACrF;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,uEAAuE,CACxE;IACD,aAAa,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,
|
|
1
|
+
{"version":3,"file":"tasksummaryresponse.js","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,gBAAgB,IAAI,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAE1E,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,MAAM,CAAC,MAAM,QAAQ,GAAG;IACtB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,UAAU,EAAE,aAAa;IACzB,YAAY,EAAE,cAAc;CACpB,CAAC;AAMX,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,IAAI,CAAC;IACvC,KAAK;IACL,KAAK;IACL,aAAa;IACb,cAAc;CACf,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;AAahC,MAAM,CAAC,MAAM,6BAA6B,GACxC,iBAAiB,CACf,CAAC,CAAC,MAAM,CAAC;IACP,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,oFAAoF,CACrF;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,uEAAuE,CACxE;IACD,aAAa,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SAC/D,QAAQ,CAAC,sCAAsC,CAAC;IACnD,UAAU,EAAE,wBAAwB;IACpC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACnD,8GAA8G,CAC/G;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IAC9C,SAAS,EAAE,kBAAkB,CAAC,QAAQ,CAAC,kBAAkB,CAAC;CAC3D,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,EACpB,sBAAsB,EACtB,IAAI,CACL,CAAC;AAEJ,MAAM,CAAC,MAAM,qCAAqC,GAAmB,CAAC,CAAC,MAAM,CAAC;IAC5E,oBAAoB,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IAC9D,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAClD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAClD,aAAa,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;IACnE,UAAU,EAAE,wBAAwB;IACpC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAC5C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,SAAS,EAAE,kBAAkB;CAC9B,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,MAAM,EAAE,oBAAoB,EAAE,GAAG,IAAI,EAAE,GAAG,CAAC,CAAC;IAC5C,OAAO;QACL,GAAG,IAAI;QACP,GAAG,oBAAoB;KACxB,CAAC;AACJ,CAAC,CAAC,CAAC"}
|
package/esm/tool-names.js
CHANGED
|
@@ -105,8 +105,8 @@ export const toolNames = [
|
|
|
105
105
|
"description": "Get one annotation task with its evaluators, items, and labelling jobs.\n"
|
|
106
106
|
},
|
|
107
107
|
{
|
|
108
|
-
"name": "
|
|
109
|
-
"description": "
|
|
108
|
+
"name": "set-task-evaluators",
|
|
109
|
+
"description": "Replace the task's linked evaluators with exactly the given ordered\nset: new ids are linked, omitted ids are unlinked, and the list order\nsets their display position. Send an empty list to unlink all. Each\nevaluator must be one you created or a built-in default, and the task\nmust be in your workspace.\n"
|
|
110
110
|
},
|
|
111
111
|
{
|
|
112
112
|
"name": "add-annotation-items",
|
package/esm/tool-names.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"tool-names.js","sourceRoot":"","sources":["../src/tool-names.ts"],"names":[],"mappings":"AAAA,+BAA+B;AAC/B,MAAM,CAAC,MAAM,SAAS,GAAgD;IACpE;QACE,MAAM,EAAE,yBAAyB;QACjC,aAAa,EAAE,sNAAsN;KACtO;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,oQAAoQ;KACpR;IACD;QACE,MAAM,EAAE,cAAc;QACtB,aAAa,EAAE,kMAAkM;KAClN;IACD;QACE,MAAM,EAAE,aAAa;QACrB,aAAa,EAAE,iPAAiP;KACjQ;IACD;QACE,MAAM,EAAE,WAAW;QACnB,aAAa,EAAE,6KAA6K;KAC7L;IACD;QACE,MAAM,EAAE,cAAc;QACtB,aAAa,EAAE,kNAAkN;KAClO;IACD;QACE,MAAM,EAAE,mBAAmB;QAC3B,aAAa,EAAE,+KAA+K;KAC/L;IACD;QACE,MAAM,EAAE,aAAa;QACrB,aAAa,EAAE,+LAA+L;KAC/M;IACD;QACE,MAAM,EAAE,YAAY;QACpB,aAAa,EAAE,kKAAkK;KAClL;IACD;QACE,MAAM,EAAE,UAAU;QAClB,aAAa,EAAE,0KAA0K;KAC1L;IACD;QACE,MAAM,EAAE,aAAa;QACrB,aAAa,EAAE,yMAAyM;KACzN;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,2JAA2J;KAC3K;IACD;QACE,MAAM,EAAE,kBAAkB;QAC1B,aAAa,EAAE,sCAAsC;KACtD;IACD;QACE,MAAM,EAAE,sBAAsB;QAC9B,aAAa,EAAE,wGAAwG;KACxH;IACD;QACE,MAAM,EAAE,iBAAiB;QACzB,aAAa,EAAE,oSAAoS;KACpT;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,qUAAqU;KACrV;IACD;QACE,MAAM,EAAE,oBAAoB;QAC5B,aAAa,EAAE,wRAAwR;KACxS;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,wLAAwL;KACxM;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,uFAAuF;KACvG;IACD;QACE,MAAM,EAAE,kBAAkB;QAC1B,aAAa,EAAE,4HAA4H;KAC5I;IACD;QACE,MAAM,EAAE,iBAAiB;QACzB,aAAa,EAAE,oHAAoH;KACpI;IACD;QACE,MAAM,EAAE,eAAe;QACvB,aAAa,EAAE,oDAAoD;KACpE;IACD;QACE,MAAM,EAAE,0BAA0B;QAClC,aAAa,EAAE,iIAAiI;KACjJ;IACD;QACE,MAAM,EAAE,wBAAwB;QAChC,aAAa,EAAE,0GAA0G;KAC1H;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,4DAA4D;KAC5E;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,2EAA2E;KAC3F;IACD;QACE,MAAM,EAAE,
|
|
1
|
+
{"version":3,"file":"tool-names.js","sourceRoot":"","sources":["../src/tool-names.ts"],"names":[],"mappings":"AAAA,+BAA+B;AAC/B,MAAM,CAAC,MAAM,SAAS,GAAgD;IACpE;QACE,MAAM,EAAE,yBAAyB;QACjC,aAAa,EAAE,sNAAsN;KACtO;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,oQAAoQ;KACpR;IACD;QACE,MAAM,EAAE,cAAc;QACtB,aAAa,EAAE,kMAAkM;KAClN;IACD;QACE,MAAM,EAAE,aAAa;QACrB,aAAa,EAAE,iPAAiP;KACjQ;IACD;QACE,MAAM,EAAE,WAAW;QACnB,aAAa,EAAE,6KAA6K;KAC7L;IACD;QACE,MAAM,EAAE,cAAc;QACtB,aAAa,EAAE,kNAAkN;KAClO;IACD;QACE,MAAM,EAAE,mBAAmB;QAC3B,aAAa,EAAE,+KAA+K;KAC/L;IACD;QACE,MAAM,EAAE,aAAa;QACrB,aAAa,EAAE,+LAA+L;KAC/M;IACD;QACE,MAAM,EAAE,YAAY;QACpB,aAAa,EAAE,kKAAkK;KAClL;IACD;QACE,MAAM,EAAE,UAAU;QAClB,aAAa,EAAE,0KAA0K;KAC1L;IACD;QACE,MAAM,EAAE,aAAa;QACrB,aAAa,EAAE,yMAAyM;KACzN;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,2JAA2J;KAC3K;IACD;QACE,MAAM,EAAE,kBAAkB;QAC1B,aAAa,EAAE,sCAAsC;KACtD;IACD;QACE,MAAM,EAAE,sBAAsB;QAC9B,aAAa,EAAE,wGAAwG;KACxH;IACD;QACE,MAAM,EAAE,iBAAiB;QACzB,aAAa,EAAE,oSAAoS;KACpT;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,qUAAqU;KACrV;IACD;QACE,MAAM,EAAE,oBAAoB;QAC5B,aAAa,EAAE,wRAAwR;KACxS;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,wLAAwL;KACxM;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,uFAAuF;KACvG;IACD;QACE,MAAM,EAAE,kBAAkB;QAC1B,aAAa,EAAE,4HAA4H;KAC5I;IACD;QACE,MAAM,EAAE,iBAAiB;QACzB,aAAa,EAAE,oHAAoH;KACpI;IACD;QACE,MAAM,EAAE,eAAe;QACvB,aAAa,EAAE,oDAAoD;KACpE;IACD;QACE,MAAM,EAAE,0BAA0B;QAClC,aAAa,EAAE,iIAAiI;KACjJ;IACD;QACE,MAAM,EAAE,wBAAwB;QAChC,aAAa,EAAE,0GAA0G;KAC1H;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,4DAA4D;KAC5E;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,2EAA2E;KAC3F;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,sTAAsT;KACtU;IACD;QACE,MAAM,EAAE,sBAAsB;QAC9B,aAAa,EAAE,kFAAkF;KAClG;IACD;QACE,MAAM,EAAE,yBAAyB;QACjC,aAAa,EAAE,oDAAoD;KACpE;IACD;QACE,MAAM,EAAE,2BAA2B;QACnC,aAAa,EAAE,4IAA4I;KAC5J;IACD;QACE,MAAM,EAAE,8BAA8B;QACtC,aAAa,EAAE,2EAA2E;KAC3F;IACD;QACE,MAAM,EAAE,0BAA0B;QAClC,aAAa,EAAE,gGAAgG;KAChH;IACD;QACE,MAAM,EAAE,wBAAwB;QAChC,aAAa,EAAE,iGAAiG;KACjH;CACF,CAAC"}
|
package/package.json
CHANGED
|
@@ -3,7 +3,7 @@
|
|
|
3
3
|
*/
|
|
4
4
|
|
|
5
5
|
import { CalibrateMcpCore } from "../core.js";
|
|
6
|
-
import { encodeSimple } from "../lib/encodings.js";
|
|
6
|
+
import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
|
|
7
7
|
import { compactMap } from "../lib/primitives.js";
|
|
8
8
|
import { safeParse } from "../lib/schemas.js";
|
|
9
9
|
import { RequestOptions } from "../lib/sdks.js";
|
|
@@ -96,6 +96,10 @@ async function $do(
|
|
|
96
96
|
const path$ = pathToFunc("/agent-tests/benchmark/{task_id}")(
|
|
97
97
|
pathParams$,
|
|
98
98
|
);
|
|
99
|
+
const query$ = encodeFormQuery({
|
|
100
|
+
"compact": payload$.compact,
|
|
101
|
+
"only_failed": payload$.only_failed,
|
|
102
|
+
});
|
|
99
103
|
|
|
100
104
|
const headers$ = new Headers(compactMap({
|
|
101
105
|
Accept: "application/json",
|
|
@@ -132,6 +136,7 @@ async function $do(
|
|
|
132
136
|
baseURL: options?.serverURL,
|
|
133
137
|
path: path$,
|
|
134
138
|
headers: headers$,
|
|
139
|
+
query: query$,
|
|
135
140
|
body: body$,
|
|
136
141
|
userAgent: client$._options.userAgent,
|
|
137
142
|
timeoutMs: options?.timeoutMs || client$._options.timeoutMs
|
|
@@ -3,7 +3,7 @@
|
|
|
3
3
|
*/
|
|
4
4
|
|
|
5
5
|
import { CalibrateMcpCore } from "../core.js";
|
|
6
|
-
import { encodeSimple } from "../lib/encodings.js";
|
|
6
|
+
import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
|
|
7
7
|
import { compactMap } from "../lib/primitives.js";
|
|
8
8
|
import { safeParse } from "../lib/schemas.js";
|
|
9
9
|
import { RequestOptions } from "../lib/sdks.js";
|
|
@@ -96,6 +96,10 @@ async function $do(
|
|
|
96
96
|
const path$ = pathToFunc("/agent-tests/run/{task_id}")(
|
|
97
97
|
pathParams$,
|
|
98
98
|
);
|
|
99
|
+
const query$ = encodeFormQuery({
|
|
100
|
+
"compact": payload$.compact,
|
|
101
|
+
"only_failed": payload$.only_failed,
|
|
102
|
+
});
|
|
99
103
|
|
|
100
104
|
const headers$ = new Headers(compactMap({
|
|
101
105
|
Accept: "application/json",
|
|
@@ -132,6 +136,7 @@ async function $do(
|
|
|
132
136
|
baseURL: options?.serverURL,
|
|
133
137
|
path: path$,
|
|
134
138
|
headers: headers$,
|
|
139
|
+
query: query$,
|
|
135
140
|
body: body$,
|
|
136
141
|
userAgent: client$._options.userAgent,
|
|
137
142
|
timeoutMs: options?.timeoutMs || client$._options.timeoutMs
|
|
@@ -97,6 +97,8 @@ async function $do(
|
|
|
97
97
|
pathParams$,
|
|
98
98
|
);
|
|
99
99
|
const query$ = encodeFormQuery({
|
|
100
|
+
"compact": payload$.compact,
|
|
101
|
+
"disagreement_only": payload$.disagreement_only,
|
|
100
102
|
"item_id": payload$.item_id,
|
|
101
103
|
"limit": payload$.limit,
|
|
102
104
|
"live_only": payload$.live_only,
|
|
@@ -19,21 +19,21 @@ import {
|
|
|
19
19
|
} from "../models/errors/httpclienterrors.js";
|
|
20
20
|
import { SDKValidationError } from "../models/errors/sdkvalidationerror.js";
|
|
21
21
|
import {
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
} from "../models/
|
|
22
|
+
SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutRequest,
|
|
23
|
+
SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutRequest$zodSchema,
|
|
24
|
+
} from "../models/settaskevaluatorsannotationtaskstaskuuidevaluatorsputop.js";
|
|
25
25
|
import { APICall, APIPromise } from "../types/async.js";
|
|
26
26
|
import { Result } from "../types/fp.js";
|
|
27
27
|
|
|
28
28
|
/**
|
|
29
|
-
*
|
|
29
|
+
* Update task evaluators
|
|
30
30
|
*
|
|
31
31
|
* @remarks
|
|
32
|
-
*
|
|
32
|
+
* Replace a task's linked evaluators with the given ordered set, linking, unlinking, and reordering as needed
|
|
33
33
|
*/
|
|
34
|
-
export function
|
|
34
|
+
export function annotationTasksSetEvaluators(
|
|
35
35
|
client$: CalibrateMcpCore,
|
|
36
|
-
request:
|
|
36
|
+
request: SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutRequest,
|
|
37
37
|
options?: RequestOptions,
|
|
38
38
|
): APIPromise<
|
|
39
39
|
Result<
|
|
@@ -56,7 +56,7 @@ export function annotationTasksLinkEvaluator(
|
|
|
56
56
|
|
|
57
57
|
async function $do(
|
|
58
58
|
client$: CalibrateMcpCore,
|
|
59
|
-
request:
|
|
59
|
+
request: SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutRequest,
|
|
60
60
|
options?: RequestOptions,
|
|
61
61
|
): Promise<
|
|
62
62
|
[
|
|
@@ -76,7 +76,7 @@ async function $do(
|
|
|
76
76
|
const parsed$ = safeParse(
|
|
77
77
|
request,
|
|
78
78
|
(value$) =>
|
|
79
|
-
|
|
79
|
+
SetTaskEvaluatorsAnnotationTasksTaskUuidEvaluatorsPutRequest$zodSchema
|
|
80
80
|
.parse(value$),
|
|
81
81
|
"Input validation failed",
|
|
82
82
|
);
|
|
@@ -111,7 +111,7 @@ async function $do(
|
|
|
111
111
|
options: client$._options,
|
|
112
112
|
baseURL: options?.serverURL ?? client$._baseURL ?? "",
|
|
113
113
|
operationID:
|
|
114
|
-
"
|
|
114
|
+
"set_task_evaluators_annotation_tasks__task_uuid__evaluators_put",
|
|
115
115
|
oAuth2Scopes: null,
|
|
116
116
|
resolvedSecurity: requestSecurity,
|
|
117
117
|
securitySource: client$._options.security,
|
|
@@ -129,7 +129,7 @@ async function $do(
|
|
|
129
129
|
|
|
130
130
|
const requestRes = client$._createRequest(context, {
|
|
131
131
|
security: requestSecurity,
|
|
132
|
-
method: "
|
|
132
|
+
method: "PUT",
|
|
133
133
|
baseURL: options?.serverURL,
|
|
134
134
|
path: path$,
|
|
135
135
|
headers: headers$,
|
|
@@ -3,7 +3,7 @@
|
|
|
3
3
|
*/
|
|
4
4
|
|
|
5
5
|
import { CalibrateMcpCore } from "../core.js";
|
|
6
|
-
import { encodeSimple } from "../lib/encodings.js";
|
|
6
|
+
import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
|
|
7
7
|
import { compactMap } from "../lib/primitives.js";
|
|
8
8
|
import { safeParse } from "../lib/schemas.js";
|
|
9
9
|
import { RequestOptions } from "../lib/sdks.js";
|
|
@@ -96,6 +96,9 @@ async function $do(
|
|
|
96
96
|
const path$ = pathToFunc("/evaluators/{evaluator_uuid}")(
|
|
97
97
|
pathParams$,
|
|
98
98
|
);
|
|
99
|
+
const query$ = encodeFormQuery({
|
|
100
|
+
"compact": payload$.compact,
|
|
101
|
+
});
|
|
99
102
|
|
|
100
103
|
const headers$ = new Headers(compactMap({
|
|
101
104
|
Accept: "application/json",
|
|
@@ -132,6 +135,7 @@ async function $do(
|
|
|
132
135
|
baseURL: options?.serverURL,
|
|
133
136
|
path: path$,
|
|
134
137
|
headers: headers$,
|
|
138
|
+
query: query$,
|
|
135
139
|
body: body$,
|
|
136
140
|
userAgent: client$._options.userAgent,
|
|
137
141
|
timeoutMs: options?.timeoutMs || client$._options.timeoutMs
|
package/src/landing-page.ts
CHANGED
|
@@ -930,7 +930,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
|
|
|
930
930
|
<h1>Instructions</h1>
|
|
931
931
|
<p>One-click installation for Claude Desktop users</p>
|
|
932
932
|
<div class="instruction-item">
|
|
933
|
-
<a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.
|
|
933
|
+
<a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.26/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
|
|
934
934
|
📥 Download MCP Bundle
|
|
935
935
|
</a>
|
|
936
936
|
</div>
|
package/src/lib/config.ts
CHANGED
|
@@ -65,8 +65,8 @@ export function serverURLFromOptions(options: SDKOptions): URL | null {
|
|
|
65
65
|
export const SDK_METADATA = {
|
|
66
66
|
language: "typescript",
|
|
67
67
|
openapiDocVersion: "0.1.0",
|
|
68
|
-
sdkVersion: "0.0.
|
|
68
|
+
sdkVersion: "0.0.26",
|
|
69
69
|
genVersion: "2.915.1",
|
|
70
70
|
userAgent:
|
|
71
|
-
"speakeasy-sdk/mcp-typescript 0.0.
|
|
71
|
+
"speakeasy-sdk/mcp-typescript 0.0.26 2.915.1 0.1.0 @dalmia/calibrate-mcp",
|
|
72
72
|
} as const;
|
package/src/mcp-server/server.ts
CHANGED
|
@@ -38,8 +38,8 @@ import { tool$annotationTasksGet } from "./tools/annotationTasksGet.js";
|
|
|
38
38
|
import { tool$annotationTasksGetAgreement } from "./tools/annotationTasksGetAgreement.js";
|
|
39
39
|
import { tool$annotationTasksGetEvaluatorRun } from "./tools/annotationTasksGetEvaluatorRun.js";
|
|
40
40
|
import { tool$annotationTasksGetSummary } from "./tools/annotationTasksGetSummary.js";
|
|
41
|
-
import { tool$annotationTasksLinkEvaluator } from "./tools/annotationTasksLinkEvaluator.js";
|
|
42
41
|
import { tool$annotationTasksList } from "./tools/annotationTasksList.js";
|
|
42
|
+
import { tool$annotationTasksSetEvaluators } from "./tools/annotationTasksSetEvaluators.js";
|
|
43
43
|
import { tool$annotationTasksUpdateItems } from "./tools/annotationTasksUpdateItems.js";
|
|
44
44
|
import { tool$evaluatorsCreate } from "./tools/evaluatorsCreate.js";
|
|
45
45
|
import { tool$evaluatorsCreateVersion } from "./tools/evaluatorsCreateVersion.js";
|
|
@@ -64,7 +64,7 @@ export function createMCPServer(deps: {
|
|
|
64
64
|
}) {
|
|
65
65
|
const server = new McpServer({
|
|
66
66
|
name: "CalibrateMcp",
|
|
67
|
-
version: "0.0.
|
|
67
|
+
version: "0.0.26",
|
|
68
68
|
});
|
|
69
69
|
|
|
70
70
|
const getClient = deps.getSDK || (() =>
|
|
@@ -135,7 +135,7 @@ export function createMCPServer(deps: {
|
|
|
135
135
|
tool(tool$annotationTasksCreate);
|
|
136
136
|
tool(tool$annotationTasksList);
|
|
137
137
|
tool(tool$annotationTasksGet);
|
|
138
|
-
tool(tool$
|
|
138
|
+
tool(tool$annotationTasksSetEvaluators);
|
|
139
139
|
tool(tool$annotationTasksAddItems);
|
|
140
140
|
tool(tool$annotationTasksUpdateItems);
|
|
141
141
|
tool(tool$annotationTasksCreateEvaluatorRun);
|