@dalmia/calibrate-mcp 0.0.21 → 0.0.23

package/esm/models/testlistresponse.js

@@ -0,0 +1,33 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+import * as z from "zod";
+import { TestListConfig$zodSchema } from "./testlistconfig.js";
+/**
+ * What the test judges:
+ *
+ * @remarks
+ *
+ * - `response`: judges the generated reply
+ * - `tool_call`: diffs the generated tool calls
+ * - `conversation`: judges the full conversation
+ */
+export const TestListResponseType = {
+    Response: "response",
+    ToolCall: "tool_call",
+    Conversation: "conversation",
+};
+export const TestListResponseType$zodSchema = z.enum([
+    "response",
+    "tool_call",
+    "conversation",
+]).describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n");
+export const TestListResponse$zodSchema = z.object({
+    config: TestListConfig$zodSchema.nullable().optional().describe("Trimmed config carrying only the test's description. Fetch the test by ID for the full config and evaluators"),
+    created_at: z.string().describe("When the test was created (ISO 8601 UTC)"),
+    name: z.string().describe("Name of the test"),
+    type: TestListResponseType$zodSchema.describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n"),
+    updated_at: z.string().describe("When the test was last updated (ISO 8601 UTC)"),
+    uuid: z.string().describe("Unique ID for the test"),
+}).describe("Trimmed test shape for list/index endpoints. The full config and hydrated\nevaluators live on the detail endpoint (`GET /tests/{uuid}`); the list keeps\nonly what list and attach-dropdown views render.");
+//# sourceMappingURL=testlistresponse.js.map

package/esm/models/testlistresponse.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"testlistresponse.js","sourceRoot":"","sources":["../../src/models/testlistresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAE/E;;;;;;;;GAQG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAAG;IAClC,QAAQ,EAAE,UAAU;IACpB,QAAQ,EAAE,WAAW;IACrB,YAAY,EAAE,cAAc;CACpB,CAAC;AAYX,MAAM,CAAC,MAAM,8BAA8B,GAAG,CAAC,CAAC,IAAI,CAAC;IACnD,UAAU;IACV,WAAW;IACX,cAAc;CACf,CAAC,CAAC,QAAQ,CACT,oKAAoK,CACrK,CAAC;AAkBF,MAAM,CAAC,MAAM,0BAA0B,GAAgC,CAAC,CAAC,MAAM,CAC7E;IACE,MAAM,EAAE,wBAAwB,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC7D,8GAA8G,CAC/G;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,0CAA0C,CAAC;IAC3E,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,kBAAkB,CAAC;IAC7C,IAAI,EAAE,8BAA8B,CAAC,QAAQ,CAC3C,oKAAoK,CACrK;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,+CAA+C,CAChD;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,wBAAwB,CAAC;CACpD,CACF,CAAC,QAAQ,CACR,2MAA2M,CAC5M,CAAC"}

package/esm/models/{routersteststestresponse.d.ts → testresponse.d.ts}

@@ -9,7 +9,7 @@ import { ClosedEnum } from "../types/enums.js";
  * - `tool_call`: diffs the generated tool calls
  * - `conversation`: judges the full conversation
  */
-export declare const RoutersTestsTestResponseType: {
+export declare const TestResponseType: {
     readonly Response: "response";
     readonly ToolCall: "tool_call";
     readonly Conversation: "conversation";
@@ -23,16 +23,16 @@ export declare const RoutersTestsTestResponseType: {
  * - `tool_call`: diffs the generated tool calls
  * - `conversation`: judges the full conversation
  */
-export type RoutersTestsTestResponseType = ClosedEnum<typeof RoutersTestsTestResponseType>;
-export declare const RoutersTestsTestResponseType$zodSchema: z.ZodEnum<{
+export type TestResponseType = ClosedEnum<typeof TestResponseType>;
+export declare const TestResponseType$zodSchema: z.ZodEnum<{
     response: "response";
     tool_call: "tool_call";
     conversation: "conversation";
 }>;
-export type RoutersTestsTestResponse = {
+export type TestResponse = {
     uuid: string;
     name: string;
-    type: RoutersTestsTestResponseType;
+    type: TestResponseType;
     config?: {
         [k: string]: any;
     } | null | undefined;
@@ -42,5 +42,5 @@ export type RoutersTestsTestResponse = {
         [k: string]: any;
     }> | undefined;
 };
-export declare const RoutersTestsTestResponse$zodSchema: z.ZodType<RoutersTestsTestResponse>;
-//# sourceMappingURL=routersteststestresponse.d.ts.map
+export declare const TestResponse$zodSchema: z.ZodType<TestResponse>;
+//# sourceMappingURL=testresponse.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"testresponse.d.ts","sourceRoot":"","sources":["../../src/models/testresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C;;;;;;;;GAQG;AACH,eAAO,MAAM,gBAAgB;;;;CAInB,CAAC;AACX;;;;;;;;GAQG;AACH,MAAM,MAAM,gBAAgB,GAAG,UAAU,CAAC,OAAO,gBAAgB,CAAC,CAAC;AAEnE,eAAO,MAAM,0BAA0B;;;;EAMtC,CAAC;AAEF,MAAM,MAAM,YAAY,GAAG;IACzB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,gBAAgB,CAAC;IACvB,MAAM,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IACjD,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,SAAS,CAAC;CACtD,CAAC;AAEF,eAAO,MAAM,sBAAsB,EAAE,CAAC,CAAC,OAAO,CAAC,YAAY,CAgBzD,CAAC"}

package/esm/models/{routersteststestresponse.js → testresponse.js}

@@ -11,23 +11,23 @@ import * as z from "zod";
  * - `tool_call`: diffs the generated tool calls
  * - `conversation`: judges the full conversation
  */
-export const RoutersTestsTestResponseType = {
+export const TestResponseType = {
     Response: "response",
     ToolCall: "tool_call",
     Conversation: "conversation",
 };
-export const RoutersTestsTestResponseType$zodSchema = z.enum([
+export const TestResponseType$zodSchema = z.enum([
     "response",
     "tool_call",
     "conversation",
 ]).describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n");
-export const RoutersTestsTestResponse$zodSchema = z.object({
+export const TestResponse$zodSchema = z.object({
     config: z.record(z.string(), z.any()).nullable().optional().describe("The stored config: `history`, `evaluation`, and an optional `settings`"),
     created_at: z.string().describe("When the test was created (ISO 8601 UTC)"),
     evaluators: z.array(z.record(z.string(), z.any())).optional().describe("Linked evaluators, resolved to their current live version at read time"),
     name: z.string().describe("Name of the test"),
-    type: RoutersTestsTestResponseType$zodSchema.describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n"),
+    type: TestResponseType$zodSchema.describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n"),
     updated_at: z.string().describe("When the test was last updated (ISO 8601 UTC)"),
     uuid: z.string().describe("Unique ID for the test"),
 });
-//# sourceMappingURL=routersteststestresponse.js.map
+//# sourceMappingURL=testresponse.js.map

package/esm/models/testresponse.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"testresponse.js","sourceRoot":"","sources":["../../src/models/testresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB;;;;;;;;GAQG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAG;IAC9B,QAAQ,EAAE,UAAU;IACpB,QAAQ,EAAE,WAAW;IACrB,YAAY,EAAE,cAAc;CACpB,CAAC;AAYX,MAAM,CAAC,MAAM,0BAA0B,GAAG,CAAC,CAAC,IAAI,CAAC;IAC/C,UAAU;IACV,WAAW;IACX,cAAc;CACf,CAAC,CAAC,QAAQ,CACT,oKAAoK,CACrK,CAAC;AAYF,MAAM,CAAC,MAAM,sBAAsB,GAA4B,CAAC,CAAC,MAAM,CAAC;IACtE,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,wEAAwE,CACzE;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,0CAA0C,CAAC;IAC3E,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpE,wEAAwE,CACzE;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,kBAAkB,CAAC;IAC7C,IAAI,EAAE,0BAA0B,CAAC,QAAQ,CACvC,oKAAoK,CACrK;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,+CAA+C,CAChD;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,wBAAwB,CAAC;CACpD,CAAC,CAAC"}

package/esm/models/updatetestendpointteststestuuidputop.d.ts

@@ -1,6 +1,6 @@
 import * as z from "zod";
 import { HTTPValidationError } from "./httpvalidationerror.js";
-import { RoutersTestsTestResponse } from "./routersteststestresponse.js";
+import { TestResponse } from "./testresponse.js";
 import { TestUpdate } from "./testupdate.js";
 export type UpdateTestEndpointTestsTestUuidPutRequest = {
     test_uuid: string;
@@ -8,6 +8,6 @@ export type UpdateTestEndpointTestsTestUuidPutRequest = {
     body: TestUpdate;
 };
 export declare const UpdateTestEndpointTestsTestUuidPutRequest$zodSchema: z.ZodType<UpdateTestEndpointTestsTestUuidPutRequest>;
-export type UpdateTestEndpointTestsTestUuidPutResponse = RoutersTestsTestResponse | HTTPValidationError;
+export type UpdateTestEndpointTestsTestUuidPutResponse = TestResponse | HTTPValidationError;
 export declare const UpdateTestEndpointTestsTestUuidPutResponse$zodSchema: z.ZodType<UpdateTestEndpointTestsTestUuidPutResponse>;
 //# sourceMappingURL=updatetestendpointteststestuuidputop.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"updatetestendpointteststestuuidputop.d.ts","sourceRoot":"","sources":["../../src/models/updatetestendpointteststestuuidputop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,wBAAwB,EAEzB,MAAM,+BAA+B,CAAC;AACvC,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AAEnE,MAAM,MAAM,yCAAyC,GAAG;IACtD,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,IAAI,EAAE,UAAU,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,mDAAmD,EAAE,CAAC,CAAC,OAAO,CACzE,yCAAyC,CAKzC,CAAC;AAEH,MAAM,MAAM,0CAA0C,GAClD,wBAAwB,GACxB,mBAAmB,CAAC;AAExB,eAAO,MAAM,oDAAoD,EAAE,CAAC,CAAC,OAAO,CAC1E,0CAA0C,CAI1C,CAAC"}
+{"version":3,"file":"updatetestendpointteststestuuidputop.d.ts","sourceRoot":"","sources":["../../src/models/updatetestendpointteststestuuidputop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAAE,YAAY,EAA0B,MAAM,mBAAmB,CAAC;AACzE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AAEnE,MAAM,MAAM,yCAAyC,GAAG;IACtD,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,IAAI,EAAE,UAAU,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,mDAAmD,EAAE,CAAC,CAAC,OAAO,CACzE,yCAAyC,CAKzC,CAAC;AAEH,MAAM,MAAM,0CAA0C,GAClD,YAAY,GACZ,mBAAmB,CAAC;AAExB,eAAO,MAAM,oDAAoD,EAAE,CAAC,CAAC,OAAO,CAC1E,0CAA0C,CAI1C,CAAC"}

package/esm/models/updatetestendpointteststestuuidputop.js

@@ -3,7 +3,7 @@
  */
 import * as z from "zod";
 import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
-import { RoutersTestsTestResponse$zodSchema, } from "./routersteststestresponse.js";
+import { TestResponse$zodSchema } from "./testresponse.js";
 import { TestUpdate$zodSchema } from "./testupdate.js";
 export const UpdateTestEndpointTestsTestUuidPutRequest$zodSchema = z.object({
     body: TestUpdate$zodSchema,
@@ -11,7 +11,7 @@ export const UpdateTestEndpointTestsTestUuidPutRequest$zodSchema = z.object({
     xAPIKey: z.string().nullable().optional(),
 });
 export const UpdateTestEndpointTestsTestUuidPutResponse$zodSchema = z.union([
-    RoutersTestsTestResponse$zodSchema,
+    TestResponse$zodSchema,
     HTTPValidationError$zodSchema,
 ]);
 //# sourceMappingURL=updatetestendpointteststestuuidputop.js.map

package/esm/models/updatetestendpointteststestuuidputop.js.map

@@ -1 +1 @@
-{"version":3,"file":"updatetestendpointteststestuuidputop.js","sourceRoot":"","sources":["../../src/models/updatetestendpointteststestuuidputop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,kCAAkC,GACnC,MAAM,+BAA+B,CAAC;AACvC,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAQnE,MAAM,CAAC,MAAM,mDAAmD,GAE5D,CAAC,CAAC,MAAM,CAAC;IACX,IAAI,EAAE,oBAAoB;IAC1B,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IAChD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAMH,MAAM,CAAC,MAAM,oDAAoD,GAE7D,CAAC,CAAC,KAAK,CAAC;IACV,kCAAkC;IAClC,6BAA6B;CAC9B,CAAC,CAAC"}
+{"version":3,"file":"updatetestendpointteststestuuidputop.js","sourceRoot":"","sources":["../../src/models/updatetestendpointteststestuuidputop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAAgB,sBAAsB,EAAE,MAAM,mBAAmB,CAAC;AACzE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAQnE,MAAM,CAAC,MAAM,mDAAmD,GAE5D,CAAC,CAAC,MAAM,CAAC;IACX,IAAI,EAAE,oBAAoB;IAC1B,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IAChD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAMH,MAAM,CAAC,MAAM,oDAAoD,GAE7D,CAAC,CAAC,KAAK,CAAC;IACV,sBAAsB;IACtB,6BAA6B;CAC9B,CAAC,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dalmia/calibrate-mcp",
-  "version": "0.0.21",
+  "version": "0.0.23",
   "author": "Speakeasy",
   "type": "module",
   "sideEffects": false,

package/src/funcs/agentTestsListForAgent.ts

@@ -3,7 +3,7 @@
  */
 
 import { CalibrateMcpCore } from "../core.js";
-import { encodeSimple } from "../lib/encodings.js";
+import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
 import { compactMap } from "../lib/primitives.js";
 import { safeParse } from "../lib/schemas.js";
 import { RequestOptions } from "../lib/sdks.js";
@@ -95,6 +95,11 @@ async function $do(
   const path$ = pathToFunc("/agent-tests/agent/{agent_uuid}/tests")(
     pathParams$,
   );
+  const query$ = encodeFormQuery({
+    "limit": payload$.limit,
+    "offset": payload$.offset,
+    "q": payload$.q,
+  });
 
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
@@ -132,6 +137,7 @@ async function $do(
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs

package/src/funcs/agentTestsListRunsForAgent.ts

@@ -3,7 +3,7 @@
  */
 
 import { CalibrateMcpCore } from "../core.js";
-import { encodeSimple } from "../lib/encodings.js";
+import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
 import { compactMap } from "../lib/primitives.js";
 import { safeParse } from "../lib/schemas.js";
 import { RequestOptions } from "../lib/sdks.js";
@@ -96,6 +96,13 @@ async function $do(
   const path$ = pathToFunc("/agent-tests/agent/{agent_uuid}/runs")(
     pathParams$,
   );
+  const query$ = encodeFormQuery({
+    "has_failures": payload$.has_failures,
+    "limit": payload$.limit,
+    "offset": payload$.offset,
+    "status": payload$.status,
+    "type": payload$.type,
+  });
 
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
@@ -132,6 +139,7 @@ async function $do(
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs

package/src/funcs/agentsList.ts

@@ -3,7 +3,7 @@
  */
 
 import { CalibrateMcpCore } from "../core.js";
-import { encodeSimple } from "../lib/encodings.js";
+import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
 import { compactMap } from "../lib/primitives.js";
 import { safeParse } from "../lib/schemas.js";
 import { RequestOptions } from "../lib/sdks.js";
@@ -84,6 +84,11 @@ async function $do(
   const payload$ = parsed$.value;
   const body$ = null;
   const path$ = pathToFunc("/agents")();
+  const query$ = encodeFormQuery({
+    "limit": payload$?.limit,
+    "offset": payload$?.offset,
+    "q": payload$?.q,
+  });
 
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
@@ -120,6 +125,7 @@ async function $do(
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs

package/src/funcs/annotationTasksList.ts

@@ -3,7 +3,7 @@
  */
 
 import { CalibrateMcpCore } from "../core.js";
-import { encodeSimple } from "../lib/encodings.js";
+import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
 import { compactMap } from "../lib/primitives.js";
 import { safeParse } from "../lib/schemas.js";
 import { RequestOptions } from "../lib/sdks.js";
@@ -87,6 +87,11 @@ async function $do(
   const payload$ = parsed$.value;
   const body$ = null;
   const path$ = pathToFunc("/annotation-tasks")();
+  const query$ = encodeFormQuery({
+    "limit": payload$?.limit,
+    "offset": payload$?.offset,
+    "q": payload$?.q,
+  });
 
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
@@ -123,6 +128,7 @@ async function $do(
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs

package/src/funcs/evaluatorsList.ts

@@ -89,6 +89,9 @@ async function $do(
     "data_type": payload$?.data_type,
     "evaluator_type": payload$?.evaluator_type,
     "include_defaults": payload$?.include_defaults,
+    "limit": payload$?.limit,
+    "offset": payload$?.offset,
+    "q": payload$?.q,
   });
 
   const headers$ = new Headers(compactMap({

package/src/funcs/testsList.ts

@@ -3,7 +3,7 @@
  */
 
 import { CalibrateMcpCore } from "../core.js";
-import { encodeSimple } from "../lib/encodings.js";
+import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
 import { compactMap } from "../lib/primitives.js";
 import { safeParse } from "../lib/schemas.js";
 import { RequestOptions } from "../lib/sdks.js";
@@ -84,6 +84,11 @@ async function $do(
   const payload$ = parsed$.value;
   const body$ = null;
   const path$ = pathToFunc("/tests")();
+  const query$ = encodeFormQuery({
+    "limit": payload$?.limit,
+    "offset": payload$?.offset,
+    "q": payload$?.q,
+  });
 
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
@@ -120,6 +125,7 @@ async function $do(
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs

package/src/landing-page.ts

@@ -930,7 +930,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
         <h1>Instructions</h1>
         <p>One-click installation for Claude Desktop users</p>
         <div class="instruction-item">
-          <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.21/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
+          <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.23/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
             📥 Download MCP Bundle
           </a>
         </div>

package/src/lib/config.ts

@@ -65,8 +65,8 @@ export function serverURLFromOptions(options: SDKOptions): URL | null {
 export const SDK_METADATA = {
   language: "typescript",
   openapiDocVersion: "0.1.0",
-  sdkVersion: "0.0.21",
+  sdkVersion: "0.0.23",
   genVersion: "2.915.1",
   userAgent:
-    "speakeasy-sdk/mcp-typescript 0.0.21 2.915.1 0.1.0 @dalmia/calibrate-mcp",
+    "speakeasy-sdk/mcp-typescript 0.0.23 2.915.1 0.1.0 @dalmia/calibrate-mcp",
 } as const;

package/src/mcp-server/mcp-server.ts

@@ -21,7 +21,7 @@ const routes = buildRouteMap({
 export const app = buildApplication(routes, {
   name: "mcp",
   versionInfo: {
-    currentVersion: "0.0.21",
+    currentVersion: "0.0.23",
   },
 });
 

package/src/mcp-server/server.ts

@@ -64,7 +64,7 @@ export function createMCPServer(deps: {
 }) {
   const server = new McpServer({
     name: "CalibrateMcp",
-    version: "0.0.21",
+    version: "0.0.23",
   });
 
   const getClient = deps.getSDK || (() =>

package/src/models/agentsummary.ts

@@ -0,0 +1,53 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import * as z from "zod";
+import { ClosedEnum } from "../types/enums.js";
+
+/**
+ * - `agent`: built inside Calibrate
+ *
+ * @remarks
+ * - `connection`: your existing agent connected to Calibrate
+ */
+export const AgentSummaryType = {
+  Agent: "agent",
+  Connection: "connection",
+} as const;
+/**
+ * - `agent`: built inside Calibrate
+ *
+ * @remarks
+ * - `connection`: your existing agent connected to Calibrate
+ */
+export type AgentSummaryType = ClosedEnum<typeof AgentSummaryType>;
+
+export const AgentSummaryType$zodSchema = z.enum([
+  "agent",
+  "connection",
+]).describe(
+  "- `agent`: built inside Calibrate\n- `connection`: your existing agent connected to Calibrate",
+);
+
+export type AgentSummary = {
+  uuid: string;
+  name: string;
+  type: AgentSummaryType;
+  updated_at: string;
+  connection_verified?: boolean | null | undefined;
+};
+
+export const AgentSummary$zodSchema: z.ZodType<AgentSummary> = z.object({
+  connection_verified: z.boolean().nullable().optional().describe(
+    "Whether the agent's connection has been verified, for a `type=connection` agent",
+  ),
+  name: z.string().describe("Name of the agent"),
+  type: AgentSummaryType$zodSchema.describe(
+    "- `agent`: built inside Calibrate\n- `connection`: your existing agent connected to Calibrate",
+  ),
+  updated_at: z.string().describe(
+    "When the agent was last updated (ISO 8601 UTC)",
+  ),
+  uuid: z.string().describe("ID of the agent"),
+});

package/src/models/annotationtaskresponse.ts

@@ -63,7 +63,7 @@ export const AnnotationTaskResponse$zodSchema: z.ZodType<
     "The task's description",
   ),
   evaluators: z.array(z.record(z.string(), z.any())).optional().describe(
-    "The evaluators linked to this task, in display order, each with its live version's rubric",
+    "The evaluators linked to this task, in display order. Fetching one task by ID enriches each with its live version's rubric. Listing tasks returns lightweight metadata (uuid, name, type, position) only",
   ),
   item_count: z.int().nullable().optional().describe(
     "Number of items in the task",

package/src/models/evaluatorliveversionsummary.ts

@@ -0,0 +1,33 @@
+/*
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
+ */
+
+import * as z from "zod";
+import { VariableSpec, VariableSpec$zodSchema } from "./variablespec.js";
+
+/**
+ * Slim view of the live version for list results, carrying only what the list needs.
+ */
+export type EvaluatorLiveVersionSummary = {
+  uuid: string;
+  version_number: number;
+  judge_model: string;
+  variables?: Array<VariableSpec> | null | undefined;
+};
+
+export const EvaluatorLiveVersionSummary$zodSchema: z.ZodType<
+  EvaluatorLiveVersionSummary
+> = z.object({
+  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`",
+  ),
+  uuid: z.string().describe("Version ID"),
+  variables: z.array(VariableSpec$zodSchema).nullable().optional().describe(
+    "Declared prompt variables",
+  ),
+  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",
+  ),
+}).describe(
+  "Slim view of the live version for list results, carrying only what the list needs.",
+);

package/src/models/getagenttestrunsagenttestsagentagentuuidrunsgetop.ts

@@ -3,32 +3,85 @@
  */
 
 import * as z from "zod";
-import {
-  AgentTestRunsResponse,
-  AgentTestRunsResponse$zodSchema,
-} from "./agenttestrunsresponse.js";
+import { ClosedEnum } from "../types/enums.js";
 import {
   HTTPValidationError,
   HTTPValidationError$zodSchema,
 } from "./httpvalidationerror.js";
+import {
+  PaginatedResponseAgentTestRunListItem,
+  PaginatedResponseAgentTestRunListItem$zodSchema,
+} from "./paginatedresponseagenttestrunlistitem.js";
+import { TaskStatus, TaskStatus$zodSchema } from "./taskstatus.js";
+
+/**
+ * Filter by run type. Omit to return both:
+ *
+ * @remarks
+ * - `llm-unit-test`: single runs of an agent's tests
+ * - `llm-benchmark`: multi-model comparisons
+ */
+export const GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetType = {
+  LlmUnitTest: "llm-unit-test",
+  LlmBenchmark: "llm-benchmark",
+} as const;
+/**
+ * Filter by run type. Omit to return both:
+ *
+ * @remarks
+ * - `llm-unit-test`: single runs of an agent's tests
+ * - `llm-benchmark`: multi-model comparisons
+ */
+export type GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetType = ClosedEnum<
+  typeof GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetType
+>;
+
+export const GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetType$zodSchema = z
+  .enum([
+    "llm-unit-test",
+    "llm-benchmark",
+  ]).describe(
+    "Filter by run type. Omit to return both:\n- `llm-unit-test`: single runs of an agent's tests\n- `llm-benchmark`: multi-model comparisons",
+  );
 
 export type GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetRequest = {
   agent_uuid: string;
+  type?: GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetType | null | undefined;
+  status?: TaskStatus | null | undefined;
+  has_failures?: boolean | null | undefined;
+  limit?: number | null | undefined;
+  offset?: number | undefined;
   xAPIKey?: string | null | undefined;
 };
 
 export const GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetRequest$zodSchema:
   z.ZodType<GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetRequest> = z.object({
     agent_uuid: z.string().describe("Agent whose test runs to list"),
+    has_failures: z.boolean().describe(
+      "Filter by whether the run has any failing test case or model. `true` returns only runs with failures (or errors), `false` only clean runs. Omit for both",
+    ).nullable().optional(),
+    limit: z.int().describe(
+      "Maximum number of items to return. Omit for no limit (all items)",
+    ).nullable().optional(),
+    offset: z.int().default(0).describe(
+      "Number of items to skip before returning results",
+    ),
+    status: TaskStatus$zodSchema.nullable().optional().describe(
+      "Filter by run status. Omit for all statuses",
+    ),
+    type: GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetType$zodSchema
+      .nullable().optional().describe(
+        "Filter by run type. Omit to return both:\n- `llm-unit-test`: single runs of an agent's tests\n- `llm-benchmark`: multi-model comparisons",
+      ),
     xAPIKey: z.string().nullable().optional(),
   });
 
 export type GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetResponse =
-  | AgentTestRunsResponse
+  | PaginatedResponseAgentTestRunListItem
   | HTTPValidationError;
 
 export const GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetResponse$zodSchema:
   z.ZodType<GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetResponse> = z.union([
-    AgentTestRunsResponse$zodSchema,
+    PaginatedResponseAgentTestRunListItem$zodSchema,
     HTTPValidationError$zodSchema,
   ]);

package/src/models/getagenttestsendpointagenttestsagentagentuuidtestsgetop.ts

@@ -8,12 +8,15 @@ import {
   HTTPValidationError$zodSchema,
 } from "./httpvalidationerror.js";
 import {
-  RoutersAgentTestsTestResponse,
-  RoutersAgentTestsTestResponse$zodSchema,
-} from "./routersagentteststestresponse.js";
+  PaginatedResponseTestListResponse,
+  PaginatedResponseTestListResponse$zodSchema,
+} from "./paginatedresponsetestlistresponse.js";
 
 export type GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetRequest = {
   agent_uuid: string;
+  q?: string | null | undefined;
+  limit?: number | null | undefined;
+  offset?: number | undefined;
   xAPIKey?: string | null | undefined;
 };
 
@@ -21,18 +24,25 @@ export const GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetRequest$zodSch
   z.ZodType<GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetRequest> = z
     .object({
       agent_uuid: z.string().describe("Agent whose linked tests to list"),
+      limit: z.int().describe(
+        "Maximum number of items to return. Omit for no limit (all items)",
+      ).nullable().optional(),
+      offset: z.int().default(0).describe(
+        "Number of items to skip before returning results",
+      ),
+      q: z.string().describe(
+        "Case-insensitive substring search on `name`. Blank is a no-op",
+      ).nullable().optional(),
       xAPIKey: z.string().nullable().optional(),
     });
 
 export type GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetResponse =
-  | Array<RoutersAgentTestsTestResponse>
+  | PaginatedResponseTestListResponse
   | HTTPValidationError;
 
 export const GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetResponse$zodSchema:
   z.ZodType<GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetResponse> = z
     .union([
-      z.array(RoutersAgentTestsTestResponse$zodSchema).describe(
-        "Successful Response",
-      ),
+      PaginatedResponseTestListResponse$zodSchema,
       HTTPValidationError$zodSchema,
     ]);

package/src/models/gettestendpointteststestuuidgetop.ts

@@ -7,10 +7,7 @@ import {
   HTTPValidationError,
   HTTPValidationError$zodSchema,
 } from "./httpvalidationerror.js";
-import {
-  RoutersTestsTestResponse,
-  RoutersTestsTestResponse$zodSchema,
-} from "./routersteststestresponse.js";
+import { TestResponse, TestResponse$zodSchema } from "./testresponse.js";
 
 export type GetTestEndpointTestsTestUuidGetRequest = {
   test_uuid: string;
@@ -25,12 +22,12 @@ export const GetTestEndpointTestsTestUuidGetRequest$zodSchema: z.ZodType<
 });
 
 export type GetTestEndpointTestsTestUuidGetResponse =
-  | RoutersTestsTestResponse
+  | TestResponse
   | HTTPValidationError;
 
 export const GetTestEndpointTestsTestUuidGetResponse$zodSchema: z.ZodType<
   GetTestEndpointTestsTestUuidGetResponse
 > = z.union([
-  RoutersTestsTestResponse$zodSchema,
+  TestResponse$zodSchema,
   HTTPValidationError$zodSchema,
 ]);