@dalmia/calibrate-mcp 0.0.48 → 0.0.50

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.
Files changed (37) hide show
  1. package/bin/mcp-server.js +14 -9
  2. package/bin/mcp-server.js.map +11 -11
  3. package/esm/landing-page.js +1 -1
  4. package/esm/lib/config.d.ts +2 -2
  5. package/esm/lib/config.js +2 -2
  6. package/esm/mcp-server/mcp-server.js +1 -1
  7. package/esm/mcp-server/server.js +1 -1
  8. package/esm/models/agenttestrunlistitem.d.ts +1 -0
  9. package/esm/models/agenttestrunlistitem.d.ts.map +1 -1
  10. package/esm/models/agenttestrunlistitem.js +1 -0
  11. package/esm/models/agenttestrunlistitem.js.map +1 -1
  12. package/esm/models/batchtestskip.d.ts +4 -0
  13. package/esm/models/batchtestskip.d.ts.map +1 -1
  14. package/esm/models/batchtestskip.js +5 -2
  15. package/esm/models/batchtestskip.js.map +1 -1
  16. package/esm/models/benchmarkstatusresponse.d.ts +1 -0
  17. package/esm/models/benchmarkstatusresponse.d.ts.map +1 -1
  18. package/esm/models/benchmarkstatusresponse.js +1 -0
  19. package/esm/models/benchmarkstatusresponse.js.map +1 -1
  20. package/esm/models/testcaseresult.d.ts +1 -0
  21. package/esm/models/testcaseresult.d.ts.map +1 -1
  22. package/esm/models/testcaseresult.js +1 -0
  23. package/esm/models/testcaseresult.js.map +1 -1
  24. package/esm/models/testrunstatusresponse.d.ts +1 -0
  25. package/esm/models/testrunstatusresponse.d.ts.map +1 -1
  26. package/esm/models/testrunstatusresponse.js +1 -0
  27. package/esm/models/testrunstatusresponse.js.map +1 -1
  28. package/package.json +1 -1
  29. package/src/landing-page.ts +1 -1
  30. package/src/lib/config.ts +2 -2
  31. package/src/mcp-server/mcp-server.ts +1 -1
  32. package/src/mcp-server/server.ts +1 -1
  33. package/src/models/agenttestrunlistitem.ts +4 -0
  34. package/src/models/batchtestskip.ts +6 -2
  35. package/src/models/benchmarkstatusresponse.ts +4 -0
  36. package/src/models/testcaseresult.ts +4 -0
  37. package/src/models/testrunstatusresponse.ts +4 -0
@@ -915,7 +915,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
915
915
  <h1>Instructions</h1>
916
916
  <p>One-click installation for Claude Desktop users</p>
917
917
  <div class="instruction-item">
918
- <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.48/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
918
+ <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.50/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
919
919
  📥 Download MCP Bundle
920
920
  </a>
921
921
  </div>
@@ -35,8 +35,8 @@ export declare function serverURLFromOptions(options: SDKOptions): URL | null;
35
35
  export declare const SDK_METADATA: {
36
36
  readonly language: "typescript";
37
37
  readonly openapiDocVersion: "0.1.0";
38
- readonly sdkVersion: "0.0.48";
38
+ readonly sdkVersion: "0.0.50";
39
39
  readonly genVersion: "2.915.1";
40
- readonly userAgent: "speakeasy-sdk/mcp-typescript 0.0.48 2.915.1 0.1.0 @dalmia/calibrate-mcp";
40
+ readonly userAgent: "speakeasy-sdk/mcp-typescript 0.0.50 2.915.1 0.1.0 @dalmia/calibrate-mcp";
41
41
  };
42
42
  //# sourceMappingURL=config.d.ts.map
package/esm/lib/config.js CHANGED
@@ -27,8 +27,8 @@ export function serverURLFromOptions(options) {
27
27
  export const SDK_METADATA = {
28
28
  language: "typescript",
29
29
  openapiDocVersion: "0.1.0",
30
- sdkVersion: "0.0.48",
30
+ sdkVersion: "0.0.50",
31
31
  genVersion: "2.915.1",
32
- userAgent: "speakeasy-sdk/mcp-typescript 0.0.48 2.915.1 0.1.0 @dalmia/calibrate-mcp",
32
+ userAgent: "speakeasy-sdk/mcp-typescript 0.0.50 2.915.1 0.1.0 @dalmia/calibrate-mcp",
33
33
  };
34
34
  //# sourceMappingURL=config.js.map
@@ -18,7 +18,7 @@ const routes = buildRouteMap({
18
18
  export const app = buildApplication(routes, {
19
19
  name: "mcp",
20
20
  versionInfo: {
21
- currentVersion: "0.0.48",
21
+ currentVersion: "0.0.50",
22
22
  },
23
23
  });
24
24
  run(app, process.argv.slice(2), buildContext(process));
@@ -50,7 +50,7 @@ import { tool$tracesCreate } from "./tools/tracesCreate.js";
50
50
  export function createMCPServer(deps) {
51
51
  const server = new McpServer({
52
52
  name: "CalibrateMcp",
53
- version: "0.0.48",
53
+ version: "0.0.50",
54
54
  });
55
55
  const getClient = deps.getSDK || (() => new CalibrateMcpCore({
56
56
  security: deps.security,
@@ -50,6 +50,7 @@ export type AgentTestRunListItem = {
50
50
  } | null | undefined;
51
51
  model_results?: Array<ModelRunSummary> | null | undefined;
52
52
  unanswered_tests?: number | null | undefined;
53
+ aborted?: boolean | undefined;
53
54
  error?: boolean | undefined;
54
55
  is_public?: boolean | undefined;
55
56
  share_token?: string | null | undefined;
@@ -1 +1 @@
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,EACL,eAAe,EAEhB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AACnE,OAAO,EACL,kBAAkB,EAEnB,MAAM,yBAAyB,CAAC;AAEjC;;;;;;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,EAAE,MAAM,CAAC;IACnB,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,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,SAAS,CAAC;IACjD,OAAO,CAAC,EAAE,KAAK,CAAC,kBAAkB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACvD,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,eAAe,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC1D,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC7C,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,CAqDvE,CAAC"}
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,EACL,eAAe,EAEhB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AACnE,OAAO,EACL,kBAAkB,EAEnB,MAAM,yBAAyB,CAAC;AAEjC;;;;;;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,EAAE,MAAM,CAAC;IACnB,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,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,SAAS,CAAC;IACjD,OAAO,CAAC,EAAE,KAAK,CAAC,kBAAkB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACvD,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,eAAe,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC1D,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC7C,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,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,CAwDvE,CAAC"}
@@ -23,6 +23,7 @@ export const AgentTestRunListItemType$zodSchema = z.enum([
23
23
  ]).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");
24
24
  export const AgentTestRunListItem$zodSchema = z
25
25
  .object({
26
+ aborted: z.boolean().default(false).describe("Whether a user stopped this run before it finished. The results collected up to that point are kept, and test cases that never ran are counted neither as passed nor as failed"),
26
27
  cost: z.record(z.string(), z.any()).nullable().optional().describe("Aggregated cost as `{mean, min, max, count}` (USD)"),
27
28
  created_at: z.string().describe("When the run was created (ISO 8601 UTC)"),
28
29
  error: z.boolean().default(false).describe("True if the run failed"),
@@ -1 +1 @@
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,EAEL,yBAAyB,GAC1B,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,4BAA4B,GAC7B,MAAM,yBAAyB,CAAC;AAEjC;;;;;;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;AAwBF,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,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yCAAyC,CAAC;IAC1E,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,CACjE,sQAAsQ,CACvQ;IACD,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,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,yBAAyB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SACpE,QAAQ,CACP,8FAA8F,CAC/F;IACH,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,4BAA4B,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SACjE,QAAQ,CACP,mFAAmF,CACpF;IACH,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,gBAAgB,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtD,4JAA4J,CAC7J;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"}
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,EAEL,yBAAyB,GAC1B,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,4BAA4B,GAC7B,MAAM,yBAAyB,CAAC;AAEjC;;;;;;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;AAyBF,MAAM,CAAC,MAAM,8BAA8B,GAAoC,CAAC;KAC7E,MAAM,CAAC;IACN,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gLAAgL,CACjL;IACD,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,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yCAAyC,CAAC;IAC1E,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,CACjE,sQAAsQ,CACvQ;IACD,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,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,yBAAyB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SACpE,QAAQ,CACP,8FAA8F,CAC/F;IACH,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,4BAA4B,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SACjE,QAAQ,CACP,mFAAmF,CACpF;IACH,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,gBAAgB,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtD,4JAA4J,CAC7J;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"}
@@ -6,10 +6,12 @@ import { ClosedEnum } from "../types/enums.js";
6
6
  * @remarks
7
7
  * - `no_linked_tests`: the agent has no tests linked
8
8
  * - `connection_not_verified`: the agent's connection is not verified
9
+ * - `over_row_limit`: the agent has more linked tests than this workspace allows in one run
9
10
  */
10
11
  export declare const Reason: {
11
12
  readonly NoLinkedTests: "no_linked_tests";
12
13
  readonly ConnectionNotVerified: "connection_not_verified";
14
+ readonly OverRowLimit: "over_row_limit";
13
15
  };
14
16
  /**
15
17
  * Why this agent was not run:
@@ -17,11 +19,13 @@ export declare const Reason: {
17
19
  * @remarks
18
20
  * - `no_linked_tests`: the agent has no tests linked
19
21
  * - `connection_not_verified`: the agent's connection is not verified
22
+ * - `over_row_limit`: the agent has more linked tests than this workspace allows in one run
20
23
  */
21
24
  export type Reason = ClosedEnum<typeof Reason>;
22
25
  export declare const Reason$zodSchema: z.ZodEnum<{
23
26
  no_linked_tests: "no_linked_tests";
24
27
  connection_not_verified: "connection_not_verified";
28
+ over_row_limit: "over_row_limit";
25
29
  }>;
26
30
  export type BatchTestSkip = {
27
31
  agent_name: string;
@@ -1 +1 @@
1
- {"version":3,"file":"batchtestskip.d.ts","sourceRoot":"","sources":["../../src/models/batchtestskip.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C;;;;;;GAMG;AACH,eAAO,MAAM,MAAM;;;CAGT,CAAC;AACX;;;;;;GAMG;AACH,MAAM,MAAM,MAAM,GAAG,UAAU,CAAC,OAAO,MAAM,CAAC,CAAC;AAE/C,eAAO,MAAM,gBAAgB;;;EAK5B,CAAC;AAEF,MAAM,MAAM,aAAa,GAAG;IAC1B,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF,eAAO,MAAM,uBAAuB,EAAE,CAAC,CAAC,OAAO,CAAC,aAAa,CAM3D,CAAC"}
1
+ {"version":3,"file":"batchtestskip.d.ts","sourceRoot":"","sources":["../../src/models/batchtestskip.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C;;;;;;;GAOG;AACH,eAAO,MAAM,MAAM;;;;CAIT,CAAC;AACX;;;;;;;GAOG;AACH,MAAM,MAAM,MAAM,GAAG,UAAU,CAAC,OAAO,MAAM,CAAC,CAAC;AAE/C,eAAO,MAAM,gBAAgB;;;;EAM5B,CAAC;AAEF,MAAM,MAAM,aAAa,GAAG;IAC1B,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF,eAAO,MAAM,uBAAuB,EAAE,CAAC,CAAC,OAAO,CAAC,aAAa,CAM3D,CAAC"}
@@ -8,18 +8,21 @@ import * as z from "zod";
8
8
  * @remarks
9
9
  * - `no_linked_tests`: the agent has no tests linked
10
10
  * - `connection_not_verified`: the agent's connection is not verified
11
+ * - `over_row_limit`: the agent has more linked tests than this workspace allows in one run
11
12
  */
12
13
  export const Reason = {
13
14
  NoLinkedTests: "no_linked_tests",
14
15
  ConnectionNotVerified: "connection_not_verified",
16
+ OverRowLimit: "over_row_limit",
15
17
  };
16
18
  export const Reason$zodSchema = z.enum([
17
19
  "no_linked_tests",
18
20
  "connection_not_verified",
19
- ]).describe("Why this agent was not run:\n- `no_linked_tests`: the agent has no tests linked\n- `connection_not_verified`: the agent's connection is not verified");
21
+ "over_row_limit",
22
+ ]).describe("Why this agent was not run:\n- `no_linked_tests`: the agent has no tests linked\n- `connection_not_verified`: the agent's connection is not verified\n- `over_row_limit`: the agent has more linked tests than this workspace allows in one run");
20
23
  export const BatchTestSkip$zodSchema = z.object({
21
24
  agent_name: z.string().describe("Name of the skipped agent"),
22
25
  agent_uuid: z.string().describe("ID of the skipped agent"),
23
- reason: Reason$zodSchema.describe("Why this agent was not run:\n- `no_linked_tests`: the agent has no tests linked\n- `connection_not_verified`: the agent's connection is not verified"),
26
+ reason: Reason$zodSchema.describe("Why this agent was not run:\n- `no_linked_tests`: the agent has no tests linked\n- `connection_not_verified`: the agent's connection is not verified\n- `over_row_limit`: the agent has more linked tests than this workspace allows in one run"),
24
27
  });
25
28
  //# sourceMappingURL=batchtestskip.js.map
@@ -1 +1 @@
1
- {"version":3,"file":"batchtestskip.js","sourceRoot":"","sources":["../../src/models/batchtestskip.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,MAAM,GAAG;IACpB,aAAa,EAAE,iBAAiB;IAChC,qBAAqB,EAAE,yBAAyB;CACxC,CAAC;AAUX,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC,IAAI,CAAC;IACrC,iBAAiB;IACjB,yBAAyB;CAC1B,CAAC,CAAC,QAAQ,CACT,sJAAsJ,CACvJ,CAAC;AAQF,MAAM,CAAC,MAAM,uBAAuB,GAA6B,CAAC,CAAC,MAAM,CAAC;IACxE,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2BAA2B,CAAC;IAC5D,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yBAAyB,CAAC;IAC1D,MAAM,EAAE,gBAAgB,CAAC,QAAQ,CAC/B,sJAAsJ,CACvJ;CACF,CAAC,CAAC"}
1
+ {"version":3,"file":"batchtestskip.js","sourceRoot":"","sources":["../../src/models/batchtestskip.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,MAAM,GAAG;IACpB,aAAa,EAAE,iBAAiB;IAChC,qBAAqB,EAAE,yBAAyB;IAChD,YAAY,EAAE,gBAAgB;CACtB,CAAC;AAWX,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC,IAAI,CAAC;IACrC,iBAAiB;IACjB,yBAAyB;IACzB,gBAAgB;CACjB,CAAC,CAAC,QAAQ,CACT,iPAAiP,CAClP,CAAC;AAQF,MAAM,CAAC,MAAM,uBAAuB,GAA6B,CAAC,CAAC,MAAM,CAAC;IACxE,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2BAA2B,CAAC;IAC5D,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yBAAyB,CAAC;IAC1D,MAAM,EAAE,gBAAgB,CAAC,QAAQ,CAC/B,iPAAiP,CAClP;CACF,CAAC,CAAC"}
@@ -11,6 +11,7 @@ export type BenchmarkStatusResponse = {
11
11
  leaderboard_summary?: Array<{
12
12
  [k: string]: any;
13
13
  }> | null | undefined;
14
+ aborted?: boolean | undefined;
14
15
  error?: boolean | undefined;
15
16
  is_public?: boolean | undefined;
16
17
  share_token?: string | null | undefined;
@@ -1 +1 @@
1
- {"version":3,"file":"benchmarkstatusresponse.d.ts","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AACnE,OAAO,EACL,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAE/B,MAAM,MAAM,uBAAuB,GAAG;IACpC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9C,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,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,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,uBAAuB,CAyBvB,CAAC"}
1
+ {"version":3,"file":"benchmarkstatusresponse.d.ts","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AACnE,OAAO,EACL,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAE/B,MAAM,MAAM,uBAAuB,GAAG;IACpC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9C,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,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,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,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,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,uBAAuB,CA4BvB,CAAC"}
@@ -6,6 +6,7 @@ import { ModelResult$zodSchema } from "./modelresult.js";
6
6
  import { TaskStatus$zodSchema } from "./taskstatus.js";
7
7
  import { TestRunEvaluator$zodSchema, } from "./testrunevaluator.js";
8
8
  export const BenchmarkStatusResponse$zodSchema = z.object({
9
+ aborted: z.boolean().default(false).describe("Whether a user stopped this run before it finished. The results collected up to that point are kept, and test cases that never ran are counted neither as passed nor as failed"),
9
10
  error: z.boolean().default(false).describe("True if the run failed"),
10
11
  evaluators: z.array(TestRunEvaluator$zodSchema).nullable().optional()
11
12
  .describe("The evaluators used in this run. Each verdict in `judge_results` links to one of these by `evaluator_uuid`"),
@@ -1 +1 @@
1
- {"version":3,"file":"benchmarkstatusresponse.js","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAc/B,MAAM,CAAC,MAAM,iCAAiC,GAE1C,CAAC,CAAC,MAAM,CAAC;IACX,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,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,oCAAoC,CACrC;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,CAAC,QAAQ,CAC1E,wBAAwB,CACzB;IACD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,yCAAyC,CAC1C;IACD,MAAM,EAAE,oBAAoB;IAC5B,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,sBAAsB,CAAC;IACpD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5D,wDAAwD,CACzD;CACF,CAAC,CAAC"}
1
+ {"version":3,"file":"benchmarkstatusresponse.js","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAe/B,MAAM,CAAC,MAAM,iCAAiC,GAE1C,CAAC,CAAC,MAAM,CAAC;IACX,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gLAAgL,CACjL;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,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,oCAAoC,CACrC;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,CAAC,QAAQ,CAC1E,wBAAwB,CACzB;IACD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,yCAAyC,CAC1C;IACD,MAAM,EAAE,oBAAoB;IAC5B,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,sBAAsB,CAAC;IACpD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5D,wDAAwD,CACzD;CACF,CAAC,CAAC"}
@@ -17,6 +17,7 @@ export type TestCaseResult = {
17
17
  latency_ms?: number | null | undefined;
18
18
  cost?: number | null | undefined;
19
19
  unanswered?: boolean | undefined;
20
+ not_run?: boolean | undefined;
20
21
  };
21
22
  export declare const TestCaseResult$zodSchema: z.ZodType<TestCaseResult>;
22
23
  //# sourceMappingURL=testcaseresult.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"testcaseresult.d.ts","sourceRoot":"","sources":["../../src/models/testcaseresult.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AAEnE,MAAM,MAAM,cAAc,GAAG;IAC3B,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACzC,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACjC,MAAM,CAAC,EAAE,OAAO,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACtC,MAAM,CAAC,EAAE,UAAU,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,SAAS,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IACpD,MAAM,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IACjD,aAAa,CAAC,EAAE,KAAK,CAAC,WAAW,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACtD,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACjC,UAAU,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;CAClC,CAAC;AAEF,eAAO,MAAM,wBAAwB,EAAE,CAAC,CAAC,OAAO,CAAC,cAAc,CA4B7D,CAAC"}
1
+ {"version":3,"file":"testcaseresult.d.ts","sourceRoot":"","sources":["../../src/models/testcaseresult.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AAEnE,MAAM,MAAM,cAAc,GAAG;IAC3B,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACzC,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACjC,MAAM,CAAC,EAAE,OAAO,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACtC,MAAM,CAAC,EAAE,UAAU,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,SAAS,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IACpD,MAAM,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IACjD,aAAa,CAAC,EAAE,KAAK,CAAC,WAAW,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACtD,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACjC,UAAU,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IACjC,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;CAC/B,CAAC;AAEF,eAAO,MAAM,wBAAwB,EAAE,CAAC,CAAC,OAAO,CAAC,cAAc,CA+B7D,CAAC"}
@@ -10,6 +10,7 @@ export const TestCaseResult$zodSchema = z.object({
10
10
  judge_results: z.array(JudgeResult$zodSchema).nullable().optional().describe("One verdict for each evaluator"),
11
11
  latency_ms: z.number().nullable().optional().describe("How long the agent took to respond, in milliseconds"),
12
12
  name: z.string().nullable().optional().describe("Name of the test"),
13
+ not_run: z.boolean().default(false).describe("Whether this case never started, because a user stopped the run first. It is counted neither as passed nor as failed"),
13
14
  output: TestOutput$zodSchema.nullable().optional().describe("The agent's output for this case"),
14
15
  passed: z.boolean().nullable().optional().describe("Whether the case passed"),
15
16
  reasoning: z.string().nullable().optional().describe("The judge's reasoning, or the tool-call diff for a tool-call test"),
@@ -1 +1 @@
1
- {"version":3,"file":"testcaseresult.js","sourceRoot":"","sources":["../../src/models/testcaseresult.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAgBnE,MAAM,CAAC,MAAM,wBAAwB,GAA8B,CAAC,CAAC,MAAM,CAAC;IAC1E,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,yBAAyB,CAAC;IAC1E,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,uHAAuH,CACxH;IACD,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC1E,gCAAgC,CACjC;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACnD,qDAAqD,CACtD;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,kBAAkB,CAAC;IACnE,MAAM,EAAE,oBAAoB,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACzD,kCAAkC,CACnC;IACD,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,yBAAyB,CAAC;IAC7E,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClD,mEAAmE,CACpE;IACD,SAAS,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACrE,uCAAuC,CACxC;IACD,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACrD,oCAAoC,CACrC;IACD,UAAU,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC7C,kLAAkL,CACnL;CACF,CAAC,CAAC"}
1
+ {"version":3,"file":"testcaseresult.js","sourceRoot":"","sources":["../../src/models/testcaseresult.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAiBnE,MAAM,CAAC,MAAM,wBAAwB,GAA8B,CAAC,CAAC,MAAM,CAAC;IAC1E,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,yBAAyB,CAAC;IAC1E,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,uHAAuH,CACxH;IACD,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC1E,gCAAgC,CACjC;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACnD,qDAAqD,CACtD;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,kBAAkB,CAAC;IACnE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,sHAAsH,CACvH;IACD,MAAM,EAAE,oBAAoB,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACzD,kCAAkC,CACnC;IACD,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,yBAAyB,CAAC;IAC7E,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClD,mEAAmE,CACpE;IACD,SAAS,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACrE,uCAAuC,CACxC;IACD,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACrD,oCAAoC,CACrC;IACD,UAAU,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC7C,kLAAkL,CACnL;CACF,CAAC,CAAC"}
@@ -22,6 +22,7 @@ export type TestRunStatusResponse = {
22
22
  results?: Array<TestCaseResult> | null | undefined;
23
23
  unanswered_tests?: number | null | undefined;
24
24
  stopped_early?: boolean | undefined;
25
+ aborted?: boolean | undefined;
25
26
  error?: boolean | undefined;
26
27
  is_public?: boolean | undefined;
27
28
  share_token?: string | null | undefined;
@@ -1 +1 @@
1
- {"version":3,"file":"testrunstatusresponse.d.ts","sourceRoot":"","sources":["../../src/models/testrunstatusresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,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,MAAM,MAAM,qBAAqB,GAAG;IAClC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9C,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,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,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,OAAO,CAAC,EAAE,KAAK,CAAC,cAAc,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACnD,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC7C,aAAa,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IACpC,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,+BAA+B,EAAE,CAAC,CAAC,OAAO,CAAC,qBAAqB,CA6CzE,CAAC"}
1
+ {"version":3,"file":"testrunstatusresponse.d.ts","sourceRoot":"","sources":["../../src/models/testrunstatusresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,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,MAAM,MAAM,qBAAqB,GAAG;IAClC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9C,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,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,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,OAAO,CAAC,EAAE,KAAK,CAAC,cAAc,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACnD,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC7C,aAAa,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IACpC,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,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,+BAA+B,EAAE,CAAC,CAAC,OAAO,CAAC,qBAAqB,CAgDzE,CAAC"}
@@ -6,6 +6,7 @@ import { TaskStatus$zodSchema } from "./taskstatus.js";
6
6
  import { TestCaseResult$zodSchema } from "./testcaseresult.js";
7
7
  import { TestRunEvaluator$zodSchema, } from "./testrunevaluator.js";
8
8
  export const TestRunStatusResponse$zodSchema = z.object({
9
+ aborted: z.boolean().default(false).describe("Whether a user stopped this run before it finished. The results collected up to that point are kept, and test cases that never ran are counted neither as passed nor as failed"),
9
10
  cost: z.record(z.string(), z.any()).nullable().optional().describe("Aggregated cost as `{mean, min, max, count}` (USD)"),
10
11
  error: z.boolean().default(false).describe("True if the run failed"),
11
12
  evaluators: z.array(TestRunEvaluator$zodSchema).nullable().optional()
@@ -1 +1 @@
1
- {"version":3,"file":"testrunstatusresponse.js","sourceRoot":"","sources":["../../src/models/testrunstatusresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAC/E,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAqB/B,MAAM,CAAC,MAAM,+BAA+B,GAC1C,CAAC,CAAC,MAAM,CAAC;IACP,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,0EAA0E,CAC3E;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,aAAa,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAChD,yFAAyF,CAC1F;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC;IAC/C,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5D,kDAAkD,CACnD;IACD,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,gBAAgB,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtD,4JAA4J,CAC7J;CACF,CAAC,CAAC"}
1
+ {"version":3,"file":"testrunstatusresponse.js","sourceRoot":"","sources":["../../src/models/testrunstatusresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAC/E,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAsB/B,MAAM,CAAC,MAAM,+BAA+B,GAC1C,CAAC,CAAC,MAAM,CAAC;IACP,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gLAAgL,CACjL;IACD,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,0EAA0E,CAC3E;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,aAAa,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAChD,yFAAyF,CAC1F;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC;IAC/C,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5D,kDAAkD,CACnD;IACD,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,gBAAgB,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtD,4JAA4J,CAC7J;CACF,CAAC,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@dalmia/calibrate-mcp",
3
- "version": "0.0.48",
3
+ "version": "0.0.50",
4
4
  "author": "Speakeasy",
5
5
  "type": "module",
6
6
  "sideEffects": false,
@@ -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/v0.0.48/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
933
+ <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.50/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: "0.0.48",
68
+ sdkVersion: "0.0.50",
69
69
  genVersion: "2.915.1",
70
70
  userAgent:
71
- "speakeasy-sdk/mcp-typescript 0.0.48 2.915.1 0.1.0 @dalmia/calibrate-mcp",
71
+ "speakeasy-sdk/mcp-typescript 0.0.50 2.915.1 0.1.0 @dalmia/calibrate-mcp",
72
72
  } as const;
@@ -21,7 +21,7 @@ const routes = buildRouteMap({
21
21
  export const app = buildApplication(routes, {
22
22
  name: "mcp",
23
23
  versionInfo: {
24
- currentVersion: "0.0.48",
24
+ currentVersion: "0.0.50",
25
25
  },
26
26
  });
27
27
 
@@ -72,7 +72,7 @@ export function createMCPServer(deps: {
72
72
  }) {
73
73
  const server = new McpServer({
74
74
  name: "CalibrateMcp",
75
- version: "0.0.48",
75
+ version: "0.0.50",
76
76
  });
77
77
 
78
78
  const getClient = deps.getSDK || (() =>
@@ -64,6 +64,7 @@ export type AgentTestRunListItem = {
64
64
  total_tokens?: { [k: string]: any } | null | undefined;
65
65
  model_results?: Array<ModelRunSummary> | null | undefined;
66
66
  unanswered_tests?: number | null | undefined;
67
+ aborted?: boolean | undefined;
67
68
  error?: boolean | undefined;
68
69
  is_public?: boolean | undefined;
69
70
  share_token?: string | null | undefined;
@@ -71,6 +72,9 @@ export type AgentTestRunListItem = {
71
72
 
72
73
  export const AgentTestRunListItem$zodSchema: z.ZodType<AgentTestRunListItem> = z
73
74
  .object({
75
+ aborted: z.boolean().default(false).describe(
76
+ "Whether a user stopped this run before it finished. The results collected up to that point are kept, and test cases that never ran are counted neither as passed nor as failed",
77
+ ),
74
78
  cost: z.record(z.string(), z.any()).nullable().optional().describe(
75
79
  "Aggregated cost as `{mean, min, max, count}` (USD)",
76
80
  ),
@@ -11,10 +11,12 @@ import { ClosedEnum } from "../types/enums.js";
11
11
  * @remarks
12
12
  * - `no_linked_tests`: the agent has no tests linked
13
13
  * - `connection_not_verified`: the agent's connection is not verified
14
+ * - `over_row_limit`: the agent has more linked tests than this workspace allows in one run
14
15
  */
15
16
  export const Reason = {
16
17
  NoLinkedTests: "no_linked_tests",
17
18
  ConnectionNotVerified: "connection_not_verified",
19
+ OverRowLimit: "over_row_limit",
18
20
  } as const;
19
21
  /**
20
22
  * Why this agent was not run:
@@ -22,14 +24,16 @@ export const Reason = {
22
24
  * @remarks
23
25
  * - `no_linked_tests`: the agent has no tests linked
24
26
  * - `connection_not_verified`: the agent's connection is not verified
27
+ * - `over_row_limit`: the agent has more linked tests than this workspace allows in one run
25
28
  */
26
29
  export type Reason = ClosedEnum<typeof Reason>;
27
30
 
28
31
  export const Reason$zodSchema = z.enum([
29
32
  "no_linked_tests",
30
33
  "connection_not_verified",
34
+ "over_row_limit",
31
35
  ]).describe(
32
- "Why this agent was not run:\n- `no_linked_tests`: the agent has no tests linked\n- `connection_not_verified`: the agent's connection is not verified",
36
+ "Why this agent was not run:\n- `no_linked_tests`: the agent has no tests linked\n- `connection_not_verified`: the agent's connection is not verified\n- `over_row_limit`: the agent has more linked tests than this workspace allows in one run",
33
37
  );
34
38
 
35
39
  export type BatchTestSkip = {
@@ -42,6 +46,6 @@ export const BatchTestSkip$zodSchema: z.ZodType<BatchTestSkip> = z.object({
42
46
  agent_name: z.string().describe("Name of the skipped agent"),
43
47
  agent_uuid: z.string().describe("ID of the skipped agent"),
44
48
  reason: Reason$zodSchema.describe(
45
- "Why this agent was not run:\n- `no_linked_tests`: the agent has no tests linked\n- `connection_not_verified`: the agent's connection is not verified",
49
+ "Why this agent was not run:\n- `no_linked_tests`: the agent has no tests linked\n- `connection_not_verified`: the agent's connection is not verified\n- `over_row_limit`: the agent has more linked tests than this workspace allows in one run",
46
50
  ),
47
51
  });
@@ -17,6 +17,7 @@ export type BenchmarkStatusResponse = {
17
17
  evaluators?: Array<TestRunEvaluator> | null | undefined;
18
18
  model_results?: Array<ModelResult> | null | undefined;
19
19
  leaderboard_summary?: Array<{ [k: string]: any }> | null | undefined;
20
+ aborted?: boolean | undefined;
20
21
  error?: boolean | undefined;
21
22
  is_public?: boolean | undefined;
22
23
  share_token?: string | null | undefined;
@@ -25,6 +26,9 @@ export type BenchmarkStatusResponse = {
25
26
  export const BenchmarkStatusResponse$zodSchema: z.ZodType<
26
27
  BenchmarkStatusResponse
27
28
  > = z.object({
29
+ aborted: z.boolean().default(false).describe(
30
+ "Whether a user stopped this run before it finished. The results collected up to that point are kept, and test cases that never ran are counted neither as passed nor as failed",
31
+ ),
28
32
  error: z.boolean().default(false).describe("True if the run failed"),
29
33
  evaluators: z.array(TestRunEvaluator$zodSchema).nullable().optional()
30
34
  .describe(
@@ -18,6 +18,7 @@ export type TestCaseResult = {
18
18
  latency_ms?: number | null | undefined;
19
19
  cost?: number | null | undefined;
20
20
  unanswered?: boolean | undefined;
21
+ not_run?: boolean | undefined;
21
22
  };
22
23
 
23
24
  export const TestCaseResult$zodSchema: z.ZodType<TestCaseResult> = z.object({
@@ -32,6 +33,9 @@ export const TestCaseResult$zodSchema: z.ZodType<TestCaseResult> = z.object({
32
33
  "How long the agent took to respond, in milliseconds",
33
34
  ),
34
35
  name: z.string().nullable().optional().describe("Name of the test"),
36
+ not_run: z.boolean().default(false).describe(
37
+ "Whether this case never started, because a user stopped the run first. It is counted neither as passed nor as failed",
38
+ ),
35
39
  output: TestOutput$zodSchema.nullable().optional().describe(
36
40
  "The agent's output for this case",
37
41
  ),
@@ -24,6 +24,7 @@ export type TestRunStatusResponse = {
24
24
  results?: Array<TestCaseResult> | null | undefined;
25
25
  unanswered_tests?: number | null | undefined;
26
26
  stopped_early?: boolean | undefined;
27
+ aborted?: boolean | undefined;
27
28
  error?: boolean | undefined;
28
29
  is_public?: boolean | undefined;
29
30
  share_token?: string | null | undefined;
@@ -31,6 +32,9 @@ export type TestRunStatusResponse = {
31
32
 
32
33
  export const TestRunStatusResponse$zodSchema: z.ZodType<TestRunStatusResponse> =
33
34
  z.object({
35
+ aborted: z.boolean().default(false).describe(
36
+ "Whether a user stopped this run before it finished. The results collected up to that point are kept, and test cases that never ran are counted neither as passed nor as failed",
37
+ ),
34
38
  cost: z.record(z.string(), z.any()).nullable().optional().describe(
35
39
  "Aggregated cost as `{mean, min, max, count}` (USD)",
36
40
  ),