@dalmia/calibrate-mcp 0.0.14 → 0.0.15

package/esm/mcp-server/tools/annotationTasksLinkEvaluator.js

@@ -0,0 +1,34 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { annotationTasksLinkEvaluator } from "../../funcs/annotationTasksLinkEvaluator.js";
+import { LinkEvaluatorToTaskAnnotationTasksTaskUuidEvaluatorsPostRequest$zodSchema } from "../../models/linkevaluatortotaskannotationtaskstaskuuidevaluatorspostop.js";
+import { formatResult } from "../tools.js";
+const args = {
+    request: LinkEvaluatorToTaskAnnotationTasksTaskUuidEvaluatorsPostRequest$zodSchema,
+};
+export const tool$annotationTasksLinkEvaluator = {
+    name: "link-evaluator-to-task",
+    description: `Link an evaluator to an annotation task, appending it to the display order.
+`,
+    scopes: ["write"],
+    annotations: {
+        "title": "Link evaluator to task",
+        "destructiveHint": false,
+        "idempotentHint": true,
+        "openWorldHint": true,
+        "readOnlyHint": false,
+    },
+    args,
+    tool: async (client, args, ctx) => {
+        const [result] = await annotationTasksLinkEvaluator(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
+        if (!result.ok) {
+            return {
+                content: [{ type: "text", text: result.error.message }],
+                isError: true,
+            };
+        }
+        return formatResult(result.value);
+    },
+};
+//# sourceMappingURL=annotationTasksLinkEvaluator.js.map

package/esm/mcp-server/tools/annotationTasksLinkEvaluator.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"annotationTasksLinkEvaluator.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/annotationTasksLinkEvaluator.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,4BAA4B,EAAE,MAAM,6CAA6C,CAAC;AAC3F,OAAO,EAAE,yEAAyE,EAAE,MAAM,4EAA4E,CAAC;AACvK,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EACL,yEAAyE;CAC5E,CAAC;AAEF,MAAM,CAAC,MAAM,iCAAiC,GAAgC;IAC5E,IAAI,EAAE,wBAAwB;IAC9B,WAAW,EACT;CACH;IACC,MAAM,EAAE,CAAC,OAAO,CAAC;IACjB,WAAW,EAAE;QACX,OAAO,EAAE,wBAAwB;QACjC,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,IAAI;QACtB,eAAe,EAAE,IAAI;QACrB,cAAc,EAAE,KAAK;KACtB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,4BAA4B,CACjD,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}

package/esm/mcp-server/tools/annotationTasksList.d.ts

@@ -0,0 +1,7 @@
+import { ToolDefinition } from "../tools.js";
+declare const args: {
+    request: import("zod").ZodOptional<import("zod").ZodType<import("../../models/listannotationtasksannotationtasksgetop.js").ListAnnotationTasksAnnotationTasksGetRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/listannotationtasksannotationtasksgetop.js").ListAnnotationTasksAnnotationTasksGetRequest, unknown>>>;
+};
+export declare const tool$annotationTasksList: ToolDefinition<typeof args>;
+export {};
+//# sourceMappingURL=annotationTasksList.d.ts.map

package/esm/mcp-server/tools/annotationTasksList.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"annotationTasksList.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/annotationTasksList.ts"],"names":[],"mappings":"AAMA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,wBAAwB,EAAE,cAAc,CAAC,OAAO,IAAI,CA6BhE,CAAC"}

package/esm/mcp-server/tools/annotationTasksList.js

@@ -0,0 +1,34 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { annotationTasksList } from "../../funcs/annotationTasksList.js";
+import { ListAnnotationTasksAnnotationTasksGetRequest$zodSchema } from "../../models/listannotationtasksannotationtasksgetop.js";
+import { formatResult } from "../tools.js";
+const args = {
+    request: ListAnnotationTasksAnnotationTasksGetRequest$zodSchema.optional(),
+};
+export const tool$annotationTasksList = {
+    name: "list-annotation-tasks",
+    description: `List your annotation tasks with their linked evaluators.
+`,
+    scopes: ["read"],
+    annotations: {
+        "title": "List annotation tasks",
+        "destructiveHint": false,
+        "idempotentHint": false,
+        "openWorldHint": true,
+        "readOnlyHint": true,
+    },
+    args,
+    tool: async (client, args, ctx) => {
+        const [result] = await annotationTasksList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
+        if (!result.ok) {
+            return {
+                content: [{ type: "text", text: result.error.message }],
+                isError: true,
+            };
+        }
+        return formatResult(result.value);
+    },
+};
+//# sourceMappingURL=annotationTasksList.js.map

package/esm/mcp-server/tools/annotationTasksList.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"annotationTasksList.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/annotationTasksList.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,mBAAmB,EAAE,MAAM,oCAAoC,CAAC;AACzE,OAAO,EAAE,sDAAsD,EAAE,MAAM,yDAAyD,CAAC;AACjI,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,sDAAsD,CAAC,QAAQ,EAAE;CAC3E,CAAC;AAEF,MAAM,CAAC,MAAM,wBAAwB,GAAgC;IACnE,IAAI,EAAE,uBAAuB;IAC7B,WAAW,EAAE;CACd;IACC,MAAM,EAAE,CAAC,MAAM,CAAC;IAChB,WAAW,EAAE;QACX,OAAO,EAAE,uBAAuB;QAChC,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,IAAI;QACrB,cAAc,EAAE,IAAI;KACrB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,mBAAmB,CACxC,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}

package/esm/mcp-server/tools/annotationTasksUpdateItems.d.ts

@@ -0,0 +1,7 @@
+import { ToolDefinition } from "../tools.js";
+declare const args: {
+    request: import("zod").ZodType<import("../../models/bulkupdateitemsannotationtaskstaskuuiditemsputop.js").BulkUpdateItemsAnnotationTasksTaskUuidItemsPutRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/bulkupdateitemsannotationtaskstaskuuiditemsputop.js").BulkUpdateItemsAnnotationTasksTaskUuidItemsPutRequest, unknown>>;
+};
+export declare const tool$annotationTasksUpdateItems: ToolDefinition<typeof args>;
+export {};
+//# sourceMappingURL=annotationTasksUpdateItems.d.ts.map

package/esm/mcp-server/tools/annotationTasksUpdateItems.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"annotationTasksUpdateItems.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/annotationTasksUpdateItems.ts"],"names":[],"mappings":"AAMA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,+BAA+B,EAAE,cAAc,CAAC,OAAO,IAAI,CA6BvE,CAAC"}

package/esm/mcp-server/tools/annotationTasksUpdateItems.js

@@ -0,0 +1,34 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { annotationTasksUpdateItems } from "../../funcs/annotationTasksUpdateItems.js";
+import { BulkUpdateItemsAnnotationTasksTaskUuidItemsPutRequest$zodSchema } from "../../models/bulkupdateitemsannotationtaskstaskuuiditemsputop.js";
+import { formatResult } from "../tools.js";
+const args = {
+    request: BulkUpdateItemsAnnotationTasksTaskUuidItemsPutRequest$zodSchema,
+};
+export const tool$annotationTasksUpdateItems = {
+    name: "update-annotation-items",
+    description: `Bulk-update item payloads in an annotation task.
+`,
+    scopes: ["write"],
+    annotations: {
+        "title": "Bulk update items",
+        "destructiveHint": false,
+        "idempotentHint": true,
+        "openWorldHint": true,
+        "readOnlyHint": false,
+    },
+    args,
+    tool: async (client, args, ctx) => {
+        const [result] = await annotationTasksUpdateItems(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
+        if (!result.ok) {
+            return {
+                content: [{ type: "text", text: result.error.message }],
+                isError: true,
+            };
+        }
+        return formatResult(result.value);
+    },
+};
+//# sourceMappingURL=annotationTasksUpdateItems.js.map

package/esm/mcp-server/tools/annotationTasksUpdateItems.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"annotationTasksUpdateItems.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/annotationTasksUpdateItems.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,0BAA0B,EAAE,MAAM,2CAA2C,CAAC;AACvF,OAAO,EAAE,+DAA+D,EAAE,MAAM,kEAAkE,CAAC;AACnJ,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,+DAA+D;CACzE,CAAC;AAEF,MAAM,CAAC,MAAM,+BAA+B,GAAgC;IAC1E,IAAI,EAAE,yBAAyB;IAC/B,WAAW,EAAE;CACd;IACC,MAAM,EAAE,CAAC,OAAO,CAAC;IACjB,WAAW,EAAE;QACX,OAAO,EAAE,mBAAmB;QAC5B,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,IAAI;QACtB,eAAe,EAAE,IAAI;QACrB,cAAc,EAAE,KAAK;KACtB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,0BAA0B,CAC/C,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}

package/esm/mcp-server/tools/evaluatorsCreate.d.ts

@@ -0,0 +1,7 @@
+import { ToolDefinition } from "../tools.js";
+declare const args: {
+    request: import("zod").ZodType<import("../../models/createevaluatorendpointevaluatorspostop.js").CreateEvaluatorEndpointEvaluatorsPostRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/createevaluatorendpointevaluatorspostop.js").CreateEvaluatorEndpointEvaluatorsPostRequest, unknown>>;
+};
+export declare const tool$evaluatorsCreate: ToolDefinition<typeof args>;
+export {};
+//# sourceMappingURL=evaluatorsCreate.d.ts.map

package/esm/mcp-server/tools/evaluatorsCreate.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluatorsCreate.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/evaluatorsCreate.ts"],"names":[],"mappings":"AAMA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,qBAAqB,EAAE,cAAc,CAAC,OAAO,IAAI,CA8B7D,CAAC"}

package/esm/mcp-server/tools/evaluatorsCreate.js

@@ -0,0 +1,34 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { evaluatorsCreate } from "../../funcs/evaluatorsCreate.js";
+import { CreateEvaluatorEndpointEvaluatorsPostRequest$zodSchema } from "../../models/createevaluatorendpointevaluatorspostop.js";
+import { formatResult } from "../tools.js";
+const args = {
+    request: CreateEvaluatorEndpointEvaluatorsPostRequest$zodSchema,
+};
+export const tool$evaluatorsCreate = {
+    name: "create-evaluator",
+    description: `Create an evaluator and its first version, which is set live. An evaluator judges an output against a prompt and rubric.
+`,
+    scopes: ["write"],
+    annotations: {
+        "title": "Create evaluator",
+        "destructiveHint": false,
+        "idempotentHint": false,
+        "openWorldHint": true,
+        "readOnlyHint": false,
+    },
+    args,
+    tool: async (client, args, ctx) => {
+        const [result] = await evaluatorsCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
+        if (!result.ok) {
+            return {
+                content: [{ type: "text", text: result.error.message }],
+                isError: true,
+            };
+        }
+        return formatResult(result.value);
+    },
+};
+//# sourceMappingURL=evaluatorsCreate.js.map

package/esm/mcp-server/tools/evaluatorsCreate.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluatorsCreate.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evaluatorsCreate.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,gBAAgB,EAAE,MAAM,iCAAiC,CAAC;AACnE,OAAO,EAAE,sDAAsD,EAAE,MAAM,yDAAyD,CAAC;AACjI,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,sDAAsD;CAChE,CAAC;AAEF,MAAM,CAAC,MAAM,qBAAqB,GAAgC;IAChE,IAAI,EAAE,kBAAkB;IACxB,WAAW,EACT;CACH;IACC,MAAM,EAAE,CAAC,OAAO,CAAC;IACjB,WAAW,EAAE;QACX,OAAO,EAAE,kBAAkB;QAC3B,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,IAAI;QACrB,cAAc,EAAE,KAAK;KACtB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,gBAAgB,CACrC,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}

package/esm/mcp-server/tools/evaluatorsCreateVersion.d.ts

@@ -0,0 +1,7 @@
+import { ToolDefinition } from "../tools.js";
+declare const args: {
+    request: import("zod").ZodType<import("../../models/createversionevaluatorsevaluatoruuidversionspostop.js").CreateVersionEvaluatorsEvaluatorUuidVersionsPostRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/createversionevaluatorsevaluatoruuidversionspostop.js").CreateVersionEvaluatorsEvaluatorUuidVersionsPostRequest, unknown>>;
+};
+export declare const tool$evaluatorsCreateVersion: ToolDefinition<typeof args>;
+export {};
+//# sourceMappingURL=evaluatorsCreateVersion.d.ts.map

package/esm/mcp-server/tools/evaluatorsCreateVersion.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluatorsCreateVersion.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/evaluatorsCreateVersion.ts"],"names":[],"mappings":"AAMA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,4BAA4B,EAAE,cAAc,CAAC,OAAO,IAAI,CA8BpE,CAAC"}

package/esm/mcp-server/tools/evaluatorsCreateVersion.js

@@ -0,0 +1,34 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { evaluatorsCreateVersion } from "../../funcs/evaluatorsCreateVersion.js";
+import { CreateVersionEvaluatorsEvaluatorUuidVersionsPostRequest$zodSchema } from "../../models/createversionevaluatorsevaluatoruuidversionspostop.js";
+import { formatResult } from "../tools.js";
+const args = {
+    request: CreateVersionEvaluatorsEvaluatorUuidVersionsPostRequest$zodSchema,
+};
+export const tool$evaluatorsCreateVersion = {
+    name: "create-evaluator-version",
+    description: `Add a new version to an evaluator you created, optionally making it live. The variable names must match the previous version.
+`,
+    scopes: ["write"],
+    annotations: {
+        "title": "Create evaluator version",
+        "destructiveHint": false,
+        "idempotentHint": false,
+        "openWorldHint": true,
+        "readOnlyHint": false,
+    },
+    args,
+    tool: async (client, args, ctx) => {
+        const [result] = await evaluatorsCreateVersion(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
+        if (!result.ok) {
+            return {
+                content: [{ type: "text", text: result.error.message }],
+                isError: true,
+            };
+        }
+        return formatResult(result.value);
+    },
+};
+//# sourceMappingURL=evaluatorsCreateVersion.js.map

package/esm/mcp-server/tools/evaluatorsCreateVersion.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluatorsCreateVersion.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evaluatorsCreateVersion.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,uBAAuB,EAAE,MAAM,wCAAwC,CAAC;AACjF,OAAO,EAAE,iEAAiE,EAAE,MAAM,oEAAoE,CAAC;AACvJ,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,iEAAiE;CAC3E,CAAC;AAEF,MAAM,CAAC,MAAM,4BAA4B,GAAgC;IACvE,IAAI,EAAE,0BAA0B;IAChC,WAAW,EACT;CACH;IACC,MAAM,EAAE,CAAC,OAAO,CAAC;IACjB,WAAW,EAAE;QACX,OAAO,EAAE,0BAA0B;QACnC,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,IAAI;QACrB,cAAc,EAAE,KAAK;KACtB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,uBAAuB,CAC5C,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}

package/esm/mcp-server/tools/evaluatorsGet.d.ts

@@ -0,0 +1,7 @@
+import { ToolDefinition } from "../tools.js";
+declare const args: {
+    request: import("zod").ZodType<import("../../models/getevaluatorendpointevaluatorsevaluatoruuidgetop.js").GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/getevaluatorendpointevaluatorsevaluatoruuidgetop.js").GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest, unknown>>;
+};
+export declare const tool$evaluatorsGet: ToolDefinition<typeof args>;
+export {};
+//# sourceMappingURL=evaluatorsGet.d.ts.map

package/esm/mcp-server/tools/evaluatorsGet.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluatorsGet.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/evaluatorsGet.ts"],"names":[],"mappings":"AAMA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,kBAAkB,EAAE,cAAc,CAAC,OAAO,IAAI,CA6B1D,CAAC"}

package/esm/mcp-server/tools/evaluatorsGet.js

@@ -0,0 +1,34 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { evaluatorsGet } from "../../funcs/evaluatorsGet.js";
+import { GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest$zodSchema } from "../../models/getevaluatorendpointevaluatorsevaluatoruuidgetop.js";
+import { formatResult } from "../tools.js";
+const args = {
+    request: GetEvaluatorEndpointEvaluatorsEvaluatorUuidGetRequest$zodSchema,
+};
+export const tool$evaluatorsGet = {
+    name: "get-evaluator",
+    description: `Get one evaluator with its full version history.
+`,
+    scopes: ["read"],
+    annotations: {
+        "title": "Get evaluator",
+        "destructiveHint": false,
+        "idempotentHint": false,
+        "openWorldHint": true,
+        "readOnlyHint": true,
+    },
+    args,
+    tool: async (client, args, ctx) => {
+        const [result] = await evaluatorsGet(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
+        if (!result.ok) {
+            return {
+                content: [{ type: "text", text: result.error.message }],
+                isError: true,
+            };
+        }
+        return formatResult(result.value);
+    },
+};
+//# sourceMappingURL=evaluatorsGet.js.map

package/esm/mcp-server/tools/evaluatorsGet.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluatorsGet.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evaluatorsGet.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,8BAA8B,CAAC;AAC7D,OAAO,EAAE,+DAA+D,EAAE,MAAM,kEAAkE,CAAC;AACnJ,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,+DAA+D;CACzE,CAAC;AAEF,MAAM,CAAC,MAAM,kBAAkB,GAAgC;IAC7D,IAAI,EAAE,eAAe;IACrB,WAAW,EAAE;CACd;IACC,MAAM,EAAE,CAAC,MAAM,CAAC;IAChB,WAAW,EAAE;QACX,OAAO,EAAE,eAAe;QACxB,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,IAAI;QACrB,cAAc,EAAE,IAAI;KACrB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,aAAa,CAClC,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}

package/esm/mcp-server/tools/evaluatorsList.d.ts

@@ -0,0 +1,7 @@
+import { ToolDefinition } from "../tools.js";
+declare const args: {
+    request: import("zod").ZodOptional<import("zod").ZodType<import("../../models/listevaluatorsevaluatorsgetop.js").ListEvaluatorsEvaluatorsGetRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/listevaluatorsevaluatorsgetop.js").ListEvaluatorsEvaluatorsGetRequest, unknown>>>;
+};
+export declare const tool$evaluatorsList: ToolDefinition<typeof args>;
+export {};
+//# sourceMappingURL=evaluatorsList.d.ts.map

package/esm/mcp-server/tools/evaluatorsList.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluatorsList.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/evaluatorsList.ts"],"names":[],"mappings":"AAMA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,mBAAmB,EAAE,cAAc,CAAC,OAAO,IAAI,CA8B3D,CAAC"}

package/esm/mcp-server/tools/evaluatorsList.js

@@ -0,0 +1,34 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import { evaluatorsList } from "../../funcs/evaluatorsList.js";
+import { ListEvaluatorsEvaluatorsGetRequest$zodSchema } from "../../models/listevaluatorsevaluatorsgetop.js";
+import { formatResult } from "../tools.js";
+const args = {
+    request: ListEvaluatorsEvaluatorsGetRequest$zodSchema.optional(),
+};
+export const tool$evaluatorsList = {
+    name: "list-evaluators",
+    description: `List your evaluators. Optionally filter by what they judge or their modality, and include the built-in defaults.
+`,
+    scopes: ["read"],
+    annotations: {
+        "title": "List evaluators",
+        "destructiveHint": false,
+        "idempotentHint": false,
+        "openWorldHint": true,
+        "readOnlyHint": true,
+    },
+    args,
+    tool: async (client, args, ctx) => {
+        const [result] = await evaluatorsList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
+        if (!result.ok) {
+            return {
+                content: [{ type: "text", text: result.error.message }],
+                isError: true,
+            };
+        }
+        return formatResult(result.value);
+    },
+};
+//# sourceMappingURL=evaluatorsList.js.map

package/esm/mcp-server/tools/evaluatorsList.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluatorsList.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evaluatorsList.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,+BAA+B,CAAC;AAC/D,OAAO,EAAE,4CAA4C,EAAE,MAAM,+CAA+C,CAAC;AAC7G,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,4CAA4C,CAAC,QAAQ,EAAE;CACjE,CAAC;AAEF,MAAM,CAAC,MAAM,mBAAmB,GAAgC;IAC9D,IAAI,EAAE,iBAAiB;IACvB,WAAW,EACT;CACH;IACC,MAAM,EAAE,CAAC,MAAM,CAAC;IAChB,WAAW,EAAE;QACX,OAAO,EAAE,iBAAiB;QAC1B,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,IAAI;QACrB,cAAc,EAAE,IAAI;KACrB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,cAAc,CACnC,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}

package/esm/models/agentcreate.js

@@ -17,7 +17,7 @@ export const AgentCreateType$zodSchema = z.enum([
     "connection",
 ]).describe("- `agent`: built inside Calibrate\n- `connection`: your existing agent connected to Calibrate");
 export const AgentCreate$zodSchema = z.object({
-    config: z.record(z.string(), z.any()).nullable().optional().describe("Agent behavioral config. The keys depend on `type`.\n\n**`type=agent`** (built inside Calibrate):\n- `system_prompt` (string): the agent's instructions\n- `llm.model` (string): `provider/model`, e.g. `openai/gpt-4.1` or `google/gemini-2.5-flash`\n- `stt.provider` (string): `deepgram`, `openai`, `cartesia`, `elevenlabs`, `google`, `sarvam`, or `smallest`\n- `tts.provider` (string): `cartesia`, `openai`, `google`, `elevenlabs`, `sarvam`, or `smallest`\n- `settings.agent_speaks_first` (bool), `settings.max_assistant_turns` (int)\n- `system_tools.end_call` (bool, optional): let the agent end the call\n- `data_extraction_fields` (array, optional): `[{name, type, description, required}]`\n\n```json\n{\n  \"system_prompt\": \"You are a helpful support agent.\",\n  \"llm\": {\"model\": \"openai/gpt-4.1\"},\n  \"stt\": {\"provider\": \"deepgram\"},\n  \"tts\": {\"provider\": \"elevenlabs\"},\n  \"settings\": {\"agent_speaks_first\": true, \"max_assistant_turns\": 50}\n}\n```\n\n**`type=connection`** (your own HTTP endpoint):\n- `agent_url` (string, required): public HTTPS endpoint the agent is called at\n- `agent_headers` (object, optional): headers sent on each request, e.g. auth\n- `benchmark_provider` (string, optional): `openrouter` (default), `openai`, `google`, `anthropic`, `meta-llama`, `mistralai`, `deepseek`, `x-ai`, `cohere`, `qwen`, or `ai21`\n\n```json\n{\n  \"agent_url\": \"https://api.example.com/agent\",\n  \"agent_headers\": {\"Authorization\": \"Bearer <token>\"},\n  \"benchmark_provider\": \"openrouter\"\n}\n```\n\nFor `type=agent`, omitted keys inherit managed defaults (omit `config` entirely to use all defaults). For `type=connection`, `config` is stored as-is and must contain `agent_url`."),
+    config: z.record(z.string(), z.any()).nullable().optional().describe("Agent behavioral config. The keys depend on `type`.\n\n**`type=agent`**, built inside Calibrate:\n- `system_prompt`: the agent's instructions\n- `llm.model`: `provider/model`, e.g. `openai/gpt-4.1` or `google/gemini-2.5-flash`\n- `stt.provider`: `deepgram`, `openai`, `cartesia`, `elevenlabs`, `google`, `sarvam`, or `smallest`\n- `tts.provider`: `cartesia`, `openai`, `google`, `elevenlabs`, `sarvam`, or `smallest`\n- `settings.agent_speaks_first`, `settings.max_assistant_turns`\n- `system_tools.end_call`: let the agent end the call\n- `data_extraction_fields`: `[{name, type, description, required}]`\n\n```json\n{\n  \"system_prompt\": \"You are a helpful support agent.\",\n  \"llm\": {\"model\": \"openai/gpt-4.1\"},\n  \"stt\": {\"provider\": \"deepgram\"},\n  \"tts\": {\"provider\": \"elevenlabs\"},\n  \"settings\": {\"agent_speaks_first\": true, \"max_assistant_turns\": 50}\n}\n```\n\n**`type=connection`**, your own HTTP endpoint:\n- `agent_url`: public HTTP(S) endpoint your agent is called at\n- `agent_headers`: headers sent on each request, e.g. auth\n- `benchmark_provider`: `openrouter` by default. Other values: `openai`, `google`, `anthropic`, `meta-llama`, `mistralai`, `deepseek`, `x-ai`, `cohere`, `qwen`, or `ai21`\n\n```json\n{\n  \"agent_url\": \"https://api.example.com/agent\",\n  \"agent_headers\": {\"Authorization\": \"Bearer <token>\"},\n  \"benchmark_provider\": \"openrouter\"\n}\n```\n\nFor `type=agent`, omitted keys inherit managed defaults. Omit `config` entirely to use all defaults. For `type=connection`, `config` is stored as-is and must contain `agent_url`"),
     name: z.string().describe("Agent name, unique within the workspace"),
     type: AgentCreateType$zodSchema.default("agent").describe("- `agent`: built inside Calibrate\n- `connection`: your existing agent connected to Calibrate"),
 });

package/esm/models/agentcreate.js.map

@@ -1 +1 @@
-{"version":3,"file":"agentcreate.js","sourceRoot":"","sources":["../../src/models/agentcreate.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB;;;;;GAKG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG;IAC7B,KAAK,EAAE,OAAO;IACd,UAAU,EAAE,YAAY;CAChB,CAAC;AASX,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,IAAI,CAAC;IAC9C,OAAO;IACP,YAAY;CACb,CAAC,CAAC,QAAQ,CACT,+FAA+F,CAChG,CAAC;AAQF,MAAM,CAAC,MAAM,qBAAqB,GAA2B,CAAC,CAAC,MAAM,CAAC;IACpE,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClE,msDAAmsD,CACpsD;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yCAAyC,CAAC;IACpE,IAAI,EAAE,yBAAyB,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,QAAQ,CACvD,+FAA+F,CAChG;CACF,CAAC,CAAC"}
+{"version":3,"file":"agentcreate.js","sourceRoot":"","sources":["../../src/models/agentcreate.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB;;;;;GAKG;AACH,MAAM,CAAC,MAAM,eAAe,GAAG;IAC7B,KAAK,EAAE,OAAO;IACd,UAAU,EAAE,YAAY;CAChB,CAAC;AASX,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,IAAI,CAAC;IAC9C,OAAO;IACP,YAAY;CACb,CAAC,CAAC,QAAQ,CACT,+FAA+F,CAChG,CAAC;AAQF,MAAM,CAAC,MAAM,qBAAqB,GAA2B,CAAC,CAAC,MAAM,CAAC;IACpE,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClE,okDAAokD,CACrkD;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yCAAyC,CAAC;IACpE,IAAI,EAAE,yBAAyB,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,QAAQ,CACvD,+FAA+F,CAChG;CACF,CAAC,CAAC"}

package/esm/models/agenttestruncreateresponse.d.ts

@@ -1,14 +1,14 @@
 import * as z from "zod";
 import { ClosedEnum } from "../types/enums.js";
 /**
- * Current status of the test run
+ * Current status of the run
  */
 export declare const AgentTestRunCreateResponseStatus: {
     readonly Queued: "queued";
     readonly InProgress: "in_progress";
 };
 /**
- * Current status of the test run
+ * Current status of the run
  */
 export type AgentTestRunCreateResponseStatus = ClosedEnum<typeof AgentTestRunCreateResponseStatus>;
 export declare const AgentTestRunCreateResponseStatus$zodSchema: z.ZodEnum<{

package/esm/models/agenttestruncreateresponse.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"agenttestruncreateresponse.d.ts","sourceRoot":"","sources":["../../src/models/agenttestruncreateresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C;;GAEG;AACH,eAAO,MAAM,gCAAgC;;;CAGnC,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,gCAAgC,GAAG,UAAU,CACvD,OAAO,gCAAgC,CACxC,CAAC;AAEF,eAAO,MAAM,0CAA0C;;;EAGV,CAAC;AAE9C,MAAM,MAAM,0BAA0B,GAAG;IACvC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,gCAAgC,CAAC;CAC1C,CAAC;AAEF,eAAO,MAAM,oCAAoC,EAAE,CAAC,CAAC,OAAO,CAC1D,0BAA0B,CAM1B,CAAC"}
+{"version":3,"file":"agenttestruncreateresponse.d.ts","sourceRoot":"","sources":["../../src/models/agenttestruncreateresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C;;GAEG;AACH,eAAO,MAAM,gCAAgC;;;CAGnC,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,gCAAgC,GAAG,UAAU,CACvD,OAAO,gCAAgC,CACxC,CAAC;AAEF,eAAO,MAAM,0CAA0C;;;EAGf,CAAC;AAEzC,MAAM,MAAM,0BAA0B,GAAG;IACvC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,gCAAgC,CAAC;CAC1C,CAAC;AAEF,eAAO,MAAM,oCAAoC,EAAE,CAAC,CAAC,OAAO,CAC1D,0BAA0B,CAQ1B,CAAC"}

package/esm/models/agenttestruncreateresponse.js

@@ -3,7 +3,7 @@
  */
 import * as z from "zod";
 /**
- * Current status of the test run
+ * Current status of the run
  */
 export const AgentTestRunCreateResponseStatus = {
     Queued: "queued",
@@ -12,9 +12,9 @@ export const AgentTestRunCreateResponseStatus = {
 export const AgentTestRunCreateResponseStatus$zodSchema = z.enum([
     "queued",
     "in_progress",
-]).describe("Current status of the test run");
+]).describe("Current status of the run");
 export const AgentTestRunCreateResponse$zodSchema = z.object({
-    status: AgentTestRunCreateResponseStatus$zodSchema.describe("Current status of the test run"),
-    task_id: z.string().describe("Test run job ID. Poll for status and results"),
+    status: AgentTestRunCreateResponseStatus$zodSchema.describe("Current status of the run"),
+    task_id: z.string().describe("Test run job ID. Poll it for status and results"),
 });
 //# sourceMappingURL=agenttestruncreateresponse.js.map

package/esm/models/agenttestruncreateresponse.js.map

@@ -1 +1 @@
-{"version":3,"file":"agenttestruncreateresponse.js","sourceRoot":"","sources":["../../src/models/agenttestruncreateresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB;;GAEG;AACH,MAAM,CAAC,MAAM,gCAAgC,GAAG;IAC9C,MAAM,EAAE,QAAQ;IAChB,UAAU,EAAE,aAAa;CACjB,CAAC;AAQX,MAAM,CAAC,MAAM,0CAA0C,GAAG,CAAC,CAAC,IAAI,CAAC;IAC/D,QAAQ;IACR,aAAa;CACd,CAAC,CAAC,QAAQ,CAAC,gCAAgC,CAAC,CAAC;AAO9C,MAAM,CAAC,MAAM,oCAAoC,GAE7C,CAAC,CAAC,MAAM,CAAC;IACX,MAAM,EAAE,0CAA0C,CAAC,QAAQ,CACzD,gCAAgC,CACjC;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,8CAA8C,CAAC;CAC7E,CAAC,CAAC"}
+{"version":3,"file":"agenttestruncreateresponse.js","sourceRoot":"","sources":["../../src/models/agenttestruncreateresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB;;GAEG;AACH,MAAM,CAAC,MAAM,gCAAgC,GAAG;IAC9C,MAAM,EAAE,QAAQ;IAChB,UAAU,EAAE,aAAa;CACjB,CAAC;AAQX,MAAM,CAAC,MAAM,0CAA0C,GAAG,CAAC,CAAC,IAAI,CAAC;IAC/D,QAAQ;IACR,aAAa;CACd,CAAC,CAAC,QAAQ,CAAC,2BAA2B,CAAC,CAAC;AAOzC,MAAM,CAAC,MAAM,oCAAoC,GAE7C,CAAC,CAAC,MAAM,CAAC;IACX,MAAM,EAAE,0CAA0C,CAAC,QAAQ,CACzD,2BAA2B,CAC5B;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC1B,iDAAiD,CAClD;CACF,CAAC,CAAC"}

package/esm/models/agenttestrunlistitem.d.ts

@@ -0,0 +1,59 @@
+import * as z from "zod";
+import { ClosedEnum } from "../types/enums.js";
+import { ModelResult } from "./modelresult.js";
+import { TaskStatus } from "./taskstatus.js";
+import { TestCaseResult } from "./testcaseresult.js";
+import { TestRunEvaluator } from "./testrunevaluator.js";
+/**
+ * What kind of run this is:
+ *
+ * @remarks
+ * - `llm-unit-test`: a single run of the agent's tests
+ * - `llm-benchmark`: a multi-model comparison
+ */
+export declare const AgentTestRunListItemType: {
+    readonly LlmUnitTest: "llm-unit-test";
+    readonly LlmBenchmark: "llm-benchmark";
+};
+/**
+ * What kind of run this is:
+ *
+ * @remarks
+ * - `llm-unit-test`: a single run of the agent's tests
+ * - `llm-benchmark`: a multi-model comparison
+ */
+export type AgentTestRunListItemType = ClosedEnum<typeof AgentTestRunListItemType>;
+export declare const AgentTestRunListItemType$zodSchema: z.ZodEnum<{
+    "llm-unit-test": "llm-unit-test";
+    "llm-benchmark": "llm-benchmark";
+}>;
+export type AgentTestRunListItem = {
+    uuid: string;
+    name: string;
+    status: TaskStatus;
+    type: AgentTestRunListItemType;
+    updated_at: string;
+    evaluators?: Array<TestRunEvaluator> | null | undefined;
+    total_tests?: number | null | undefined;
+    passed?: number | null | undefined;
+    failed?: number | null | undefined;
+    results?: Array<TestCaseResult> | null | undefined;
+    latency_ms?: {
+        [k: string]: any;
+    } | null | undefined;
+    cost?: {
+        [k: string]: any;
+    } | null | undefined;
+    total_tokens?: {
+        [k: string]: any;
+    } | null | undefined;
+    model_results?: Array<ModelResult> | null | undefined;
+    leaderboard_summary?: Array<{
+        [k: string]: any;
+    }> | null | undefined;
+    error?: boolean | undefined;
+    is_public?: boolean | undefined;
+    share_token?: string | null | undefined;
+};
+export declare const AgentTestRunListItem$zodSchema: z.ZodType<AgentTestRunListItem>;
+//# sourceMappingURL=agenttestrunlistitem.d.ts.map

package/esm/models/agenttestrunlistitem.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agenttestrunlistitem.d.ts","sourceRoot":"","sources":["../../src/models/agenttestrunlistitem.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAAE,cAAc,EAA4B,MAAM,qBAAqB,CAAC;AAC/E,OAAO,EACL,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAE/B;;;;;;GAMG;AACH,eAAO,MAAM,wBAAwB;;;CAG3B,CAAC;AACX;;;;;;GAMG;AACH,MAAM,MAAM,wBAAwB,GAAG,UAAU,CAC/C,OAAO,wBAAwB,CAChC,CAAC;AAEF,eAAO,MAAM,kCAAkC;;;EAK9C,CAAC;AAEF,MAAM,MAAM,oBAAoB,GAAG;IACjC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,UAAU,CAAC;IACnB,IAAI,EAAE,wBAAwB,CAAC;IAC/B,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACxC,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACnC,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACnC,OAAO,CAAC,EAAE,KAAK,CAAC,cAAc,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACnD,UAAU,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IACrD,IAAI,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IAC/C,YAAY,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IACvD,aAAa,CAAC,EAAE,KAAK,CAAC,WAAW,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACtD,mBAAmB,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACrE,KAAK,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC5B,SAAS,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAChC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACzC,CAAC;AAEF,eAAO,MAAM,8BAA8B,EAAE,CAAC,CAAC,OAAO,CAAC,oBAAoB,CAmDvE,CAAC"}

package/esm/models/agenttestrunlistitem.js

@@ -0,0 +1,48 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import * as z from "zod";
+import { ModelResult$zodSchema } from "./modelresult.js";
+import { TaskStatus$zodSchema } from "./taskstatus.js";
+import { TestCaseResult$zodSchema } from "./testcaseresult.js";
+import { TestRunEvaluator$zodSchema, } from "./testrunevaluator.js";
+/**
+ * What kind of run this is:
+ *
+ * @remarks
+ * - `llm-unit-test`: a single run of the agent's tests
+ * - `llm-benchmark`: a multi-model comparison
+ */
+export const AgentTestRunListItemType = {
+    LlmUnitTest: "llm-unit-test",
+    LlmBenchmark: "llm-benchmark",
+};
+export const AgentTestRunListItemType$zodSchema = z.enum([
+    "llm-unit-test",
+    "llm-benchmark",
+]).describe("What kind of run this is:\n- `llm-unit-test`: a single run of the agent's tests\n- `llm-benchmark`: a multi-model comparison");
+export const AgentTestRunListItem$zodSchema = z
+    .object({
+    cost: z.record(z.string(), z.any()).nullable().optional().describe("Aggregated cost as `{mean, min, max, count}` (USD)"),
+    error: z.boolean().default(false).describe("True if the run failed"),
+    evaluators: z.array(TestRunEvaluator$zodSchema).nullable().optional()
+        .describe("The evaluators used in this run. Each verdict in `judge_results` links to one of these by `evaluator_uuid`"),
+    failed: z.int().nullable().optional().describe("Number of test cases that failed"),
+    is_public: z.boolean().default(false).describe("Whether the run is shared publicly"),
+    latency_ms: z.record(z.string(), z.any()).nullable().optional().describe("Aggregated latency in milliseconds, as `{p50, p95, p99, count}`"),
+    leaderboard_summary: z.array(z.record(z.string(), z.any())).nullable()
+        .optional().describe("Leaderboard comparing the models, one row per model. Columns vary by benchmark: a `model` column plus pass/fail counts, latency, cost, and one score column per evaluator, keyed by evaluator name"),
+    model_results: z.array(ModelResult$zodSchema).nullable().optional()
+        .describe("Results for each model in a benchmark run"),
+    name: z.string().describe("Display name, such as `Run 1` for a unit test or `Benchmark 1` for a benchmark"),
+    passed: z.int().nullable().optional().describe("Number of test cases that passed"),
+    results: z.array(TestCaseResult$zodSchema).nullable().optional().describe("Results for each test case"),
+    share_token: z.string().nullable().optional().describe("Token for building the public share URL"),
+    status: TaskStatus$zodSchema,
+    total_tests: z.int().nullable().optional().describe("Total number of test cases"),
+    total_tokens: z.record(z.string(), z.any()).nullable().optional().describe("Aggregated token usage as `{mean, min, max, count}`"),
+    type: AgentTestRunListItemType$zodSchema.describe("What kind of run this is:\n- `llm-unit-test`: a single run of the agent's tests\n- `llm-benchmark`: a multi-model comparison"),
+    updated_at: z.string().describe("When the run was last updated (ISO 8601 UTC)"),
+    uuid: z.string().describe("Test run job ID"),
+});
+//# sourceMappingURL=agenttestrunlistitem.js.map

package/esm/models/agenttestrunlistitem.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agenttestrunlistitem.js","sourceRoot":"","sources":["../../src/models/agenttestrunlistitem.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAC/E,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAE/B;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG;IACtC,WAAW,EAAE,eAAe;IAC5B,YAAY,EAAE,eAAe;CACrB,CAAC;AAYX,MAAM,CAAC,MAAM,kCAAkC,GAAG,CAAC,CAAC,IAAI,CAAC;IACvD,eAAe;IACf,eAAe;CAChB,CAAC,CAAC,QAAQ,CACT,8HAA8H,CAC/H,CAAC;AAuBF,MAAM,CAAC,MAAM,8BAA8B,GAAoC,CAAC;KAC7E,MAAM,CAAC;IACN,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAChE,oDAAoD,CACrD;IACD,KAAK,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,wBAAwB,CAAC;IACpE,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,0BAA0B,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SAClE,QAAQ,CACP,4GAA4G,CAC7G;IACH,MAAM,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5C,kCAAkC,CACnC;IACD,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,oCAAoC,CACrC;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtE,iEAAiE,CAClE;IACD,mBAAmB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;SACnE,QAAQ,EAAE,CAAC,QAAQ,CAClB,oMAAoM,CACrM;IACH,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SAChE,QAAQ,CAAC,2CAA2C,CAAC;IACxD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACvB,gFAAgF,CACjF;IACD,MAAM,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5C,kCAAkC,CACnC;IACD,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,wBAAwB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACvE,4BAA4B,CAC7B;IACD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,yCAAyC,CAC1C;IACD,MAAM,EAAE,oBAAoB;IAC5B,WAAW,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACjD,4BAA4B,CAC7B;IACD,YAAY,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACxE,qDAAqD,CACtD;IACD,IAAI,EAAE,kCAAkC,CAAC,QAAQ,CAC/C,8HAA8H,CAC/H;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,8CAA8C,CAC/C;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC;CAC7C,CAAC,CAAC"}

package/esm/models/agenttestrunsresponse.d.ts

@@ -0,0 +1,7 @@
+import * as z from "zod";
+import { AgentTestRunListItem } from "./agenttestrunlistitem.js";
+export type AgentTestRunsResponse = {
+    runs: Array<AgentTestRunListItem>;
+};
+export declare const AgentTestRunsResponse$zodSchema: z.ZodType<AgentTestRunsResponse>;
+//# sourceMappingURL=agenttestrunsresponse.d.ts.map

package/esm/models/agenttestrunsresponse.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agenttestrunsresponse.d.ts","sourceRoot":"","sources":["../../src/models/agenttestrunsresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,oBAAoB,EAErB,MAAM,2BAA2B,CAAC;AAEnC,MAAM,MAAM,qBAAqB,GAAG;IAAE,IAAI,EAAE,KAAK,CAAC,oBAAoB,CAAC,CAAA;CAAE,CAAC;AAE1E,eAAO,MAAM,+BAA+B,EAAE,CAAC,CAAC,OAAO,CAAC,qBAAqB,CAKzE,CAAC"}

package/esm/models/agenttestrunsresponse.js

@@ -0,0 +1,9 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import * as z from "zod";
+import { AgentTestRunListItem$zodSchema, } from "./agenttestrunlistitem.js";
+export const AgentTestRunsResponse$zodSchema = z.object({
+    runs: z.array(AgentTestRunListItem$zodSchema).describe("Test runs for the agent, most recently created first"),
+});
+//# sourceMappingURL=agenttestrunsresponse.js.map

package/esm/models/agenttestrunsresponse.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agenttestrunsresponse.js","sourceRoot":"","sources":["../../src/models/agenttestrunsresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,8BAA8B,GAC/B,MAAM,2BAA2B,CAAC;AAInC,MAAM,CAAC,MAAM,+BAA+B,GAC1C,CAAC,CAAC,MAAM,CAAC;IACP,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,8BAA8B,CAAC,CAAC,QAAQ,CACpD,sDAAsD,CACvD;CACF,CAAC,CAAC"}

package/esm/models/agenttestscreate.d.ts

@@ -0,0 +1,7 @@
+import * as z from "zod";
+export type AgentTestsCreate = {
+    agent_uuid: string;
+    test_uuids: Array<string>;
+};
+export declare const AgentTestsCreate$zodSchema: z.ZodType<AgentTestsCreate>;
+//# sourceMappingURL=agenttestscreate.d.ts.map

package/esm/models/agenttestscreate.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agenttestscreate.d.ts","sourceRoot":"","sources":["../../src/models/agenttestscreate.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,MAAM,MAAM,gBAAgB,GAAG;IAC7B,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;CAC3B,CAAC;AAEF,eAAO,MAAM,0BAA0B,EAAE,CAAC,CAAC,OAAO,CAAC,gBAAgB,CAOlE,CAAC"}

package/esm/models/agenttestscreate.js

@@ -0,0 +1,9 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import * as z from "zod";
+export const AgentTestsCreate$zodSchema = z.object({
+    agent_uuid: z.string().describe("Agent to link tests to"),
+    test_uuids: z.array(z.string()).describe("Tests to link. Any that are already linked are skipped"),
+});
+//# sourceMappingURL=agenttestscreate.js.map

package/esm/models/agenttestscreate.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agenttestscreate.js","sourceRoot":"","sources":["../../src/models/agenttestscreate.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAOzB,MAAM,CAAC,MAAM,0BAA0B,GAAgC,CAAC,CAAC,MAAM,CAC7E;IACE,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,wBAAwB,CAAC;IACzD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,CACtC,wDAAwD,CACzD;CACF,CACF,CAAC"}