@lightsage/mcp 0.1.11 → 0.2.0

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 (78) hide show
  1. package/README.md +2 -2
  2. package/bin/mcp-server.js +539 -464
  3. package/bin/mcp-server.js.map +33 -30
  4. package/esm/funcs/evalsMonitor.d.ts +20 -0
  5. package/esm/funcs/evalsMonitor.d.ts.map +1 -0
  6. package/esm/funcs/evalsMonitor.js +95 -0
  7. package/esm/funcs/evalsMonitor.js.map +1 -0
  8. package/esm/funcs/evalsRun.d.ts +16 -0
  9. package/esm/funcs/evalsRun.d.ts.map +1 -0
  10. package/esm/funcs/{evalRunsV2EvalRunsCreate.js → evalsRun.js} +4 -4
  11. package/esm/funcs/evalsRun.js.map +1 -0
  12. package/esm/landing-page.js +1 -1
  13. package/esm/lib/config.d.ts +2 -2
  14. package/esm/lib/config.js +2 -2
  15. package/esm/lib/config.js.map +1 -1
  16. package/esm/mcp-server/mcp-server.js +1 -1
  17. package/esm/mcp-server/mcp-server.js.map +1 -1
  18. package/esm/mcp-server/server.d.ts.map +1 -1
  19. package/esm/mcp-server/server.js +5 -3
  20. package/esm/mcp-server/server.js.map +1 -1
  21. package/esm/mcp-server/tools/evalsMonitor.d.ts +7 -0
  22. package/esm/mcp-server/tools/evalsMonitor.d.ts.map +1 -0
  23. package/esm/mcp-server/tools/evalsMonitor.js +35 -0
  24. package/esm/mcp-server/tools/evalsMonitor.js.map +1 -0
  25. package/esm/mcp-server/tools/{evalRunsV2EvalRunsCreate.d.ts → evalsRun.d.ts} +2 -2
  26. package/esm/mcp-server/tools/evalsRun.d.ts.map +1 -0
  27. package/esm/mcp-server/tools/{evalRunsV2EvalRunsCreate.js → evalsRun.js} +7 -7
  28. package/esm/mcp-server/tools/evalsRun.js.map +1 -0
  29. package/esm/models/createv2evalconfigurationrequest.d.ts +1 -0
  30. package/esm/models/createv2evalconfigurationrequest.d.ts.map +1 -1
  31. package/esm/models/createv2evalconfigurationrequest.js +1 -0
  32. package/esm/models/createv2evalconfigurationrequest.js.map +1 -1
  33. package/esm/models/createv2evalrunrequest.d.ts +9 -2
  34. package/esm/models/createv2evalrunrequest.d.ts.map +1 -1
  35. package/esm/models/createv2evalrunrequest.js +7 -2
  36. package/esm/models/createv2evalrunrequest.js.map +1 -1
  37. package/esm/models/updatev2evalconfigurationrequest.d.ts +1 -0
  38. package/esm/models/updatev2evalconfigurationrequest.d.ts.map +1 -1
  39. package/esm/models/updatev2evalconfigurationrequest.js +1 -0
  40. package/esm/models/updatev2evalconfigurationrequest.js.map +1 -1
  41. package/esm/models/v2evalconfiguration.d.ts +1 -0
  42. package/esm/models/v2evalconfiguration.d.ts.map +1 -1
  43. package/esm/models/v2evalconfiguration.js +1 -0
  44. package/esm/models/v2evalconfiguration.js.map +1 -1
  45. package/esm/models/v2evalrun.d.ts +3 -0
  46. package/esm/models/v2evalrun.d.ts.map +1 -1
  47. package/esm/models/v2evalrun.js +6 -3
  48. package/esm/models/v2evalrun.js.map +1 -1
  49. package/esm/models/v2evalrunsmonitorop.d.ts +28 -0
  50. package/esm/models/v2evalrunsmonitorop.d.ts.map +1 -0
  51. package/esm/models/v2evalrunsmonitorop.js +29 -0
  52. package/esm/models/v2evalrunsmonitorop.js.map +1 -0
  53. package/esm/models/v2listtasksresponse.d.ts +1 -1
  54. package/esm/models/v2taskslistop.d.ts +1 -1
  55. package/esm/tool-names.d.ts.map +1 -1
  56. package/esm/tool-names.js +8 -4
  57. package/esm/tool-names.js.map +1 -1
  58. package/package.json +1 -1
  59. package/src/funcs/evalsMonitor.ts +157 -0
  60. package/src/funcs/{evalRunsV2EvalRunsCreate.ts → evalsRun.ts} +3 -3
  61. package/src/landing-page.ts +1 -1
  62. package/src/lib/config.ts +2 -2
  63. package/src/mcp-server/mcp-server.ts +1 -1
  64. package/src/mcp-server/server.ts +5 -3
  65. package/src/mcp-server/tools/evalsMonitor.ts +43 -0
  66. package/src/mcp-server/tools/{evalRunsV2EvalRunsCreate.ts → evalsRun.ts} +6 -6
  67. package/src/models/createv2evalconfigurationrequest.ts +4 -0
  68. package/src/models/createv2evalrunrequest.ts +28 -4
  69. package/src/models/updatev2evalconfigurationrequest.ts +4 -0
  70. package/src/models/v2evalconfiguration.ts +4 -0
  71. package/src/models/v2evalrun.ts +17 -3
  72. package/src/models/v2evalrunsmonitorop.ts +62 -0
  73. package/src/tool-names.ts +8 -4
  74. package/esm/funcs/evalRunsV2EvalRunsCreate.d.ts +0 -16
  75. package/esm/funcs/evalRunsV2EvalRunsCreate.d.ts.map +0 -1
  76. package/esm/funcs/evalRunsV2EvalRunsCreate.js.map +0 -1
  77. package/esm/mcp-server/tools/evalRunsV2EvalRunsCreate.d.ts.map +0 -1
  78. package/esm/mcp-server/tools/evalRunsV2EvalRunsCreate.js.map +0 -1
@@ -5,15 +5,18 @@
5
5
  import * as z from "zod";
6
6
  import { PublicEvalRunProgress$zodSchema, } from "./publicevalrunprogress.js";
7
7
  export const V2EvalRun$zodSchema = z.object({
8
+ agent_id: z.string().nullable().optional().describe("Runnable agent ID selected for this run."),
9
+ agent_name: z.string().nullable().optional().describe("Display name captured for the selected agent."),
8
10
  completed_at: z.string().nullable().optional(),
9
11
  configuration_id: z.string().nullable().optional(),
10
12
  created_at: z.string().nullable().optional(),
11
13
  dispatch_id: z.string().nullable().optional(),
12
14
  error: z.string().nullable().optional(),
13
15
  estimated_cost: z.number().nullable().optional(),
14
- eval_id: z.string().nullable().optional(),
15
- id: z.string(),
16
- job_id: z.string(),
16
+ eval_id: z.string().nullable().optional().describe("Saved eval definition ID, including one created by a direct run."),
17
+ eval_name: z.string().nullable().optional().describe("Display name captured for the eval when the run was created."),
18
+ id: z.string().describe("Run ID to use when polling eval-run status."),
19
+ job_id: z.string().describe("Alias of `id` for the durable parent job."),
17
20
  progress: PublicEvalRunProgress$zodSchema.optional(),
18
21
  stage: z.string().nullable().optional(),
19
22
  started_at: z.string().nullable().optional(),
@@ -1 +1 @@
1
- {"version":3,"file":"v2evalrun.js","sourceRoot":"","sources":["../../src/models/v2evalrun.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,+BAA+B,GAChC,MAAM,4BAA4B,CAAC;AAkBpC,MAAM,CAAC,MAAM,mBAAmB,GAAyB,CAAC,CAAC,MAAM,CAAC;IAChE,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC9C,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAClD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC5C,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC7C,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvC,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAChD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACzC,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE;IACd,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;IAClB,QAAQ,EAAE,+BAA+B,CAAC,QAAQ,EAAE;IACpD,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvC,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC5C,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC"}
1
+ {"version":3,"file":"v2evalrun.js","sourceRoot":"","sources":["../../src/models/v2evalrun.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,+BAA+B,GAChC,MAAM,4BAA4B,CAAC;AAqBpC,MAAM,CAAC,MAAM,mBAAmB,GAAyB,CAAC,CAAC,MAAM,CAAC;IAChE,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACjD,0CAA0C,CAC3C;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACnD,+CAA+C,CAChD;IACD,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC9C,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAClD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC5C,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC7C,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvC,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAChD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAChD,kEAAkE,CACnE;IACD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClD,8DAA8D,CAC/D;IACD,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,6CAA6C,CAAC;IACtE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2CAA2C,CAAC;IACxE,QAAQ,EAAE,+BAA+B,CAAC,QAAQ,EAAE;IACpD,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvC,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC5C,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC"}
@@ -0,0 +1,28 @@
1
+ import * as z from "zod";
2
+ import { ClosedEnum } from "../types/enums.js";
3
+ import { HTTPValidationError } from "./httpvalidationerror.js";
4
+ /**
5
+ * Reduced activity or the full sanitized durable trace.
6
+ */
7
+ export declare const Detail: {
8
+ readonly Summary: "summary";
9
+ readonly Verbose: "verbose";
10
+ };
11
+ /**
12
+ * Reduced activity or the full sanitized durable trace.
13
+ */
14
+ export type Detail = ClosedEnum<typeof Detail>;
15
+ export declare const Detail$zodSchema: z.ZodEnum<{
16
+ summary: "summary";
17
+ verbose: "verbose";
18
+ }>;
19
+ export type V2EvalRunsMonitorRequest = {
20
+ run_id?: string | undefined;
21
+ all?: boolean | undefined;
22
+ detail?: Detail | undefined;
23
+ after_id?: number | undefined;
24
+ };
25
+ export declare const V2EvalRunsMonitorRequest$zodSchema: z.ZodType<V2EvalRunsMonitorRequest>;
26
+ export type V2EvalRunsMonitorResponse = string | HTTPValidationError;
27
+ export declare const V2EvalRunsMonitorResponse$zodSchema: z.ZodType<V2EvalRunsMonitorResponse>;
28
+ //# sourceMappingURL=v2evalrunsmonitorop.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunsmonitorop.d.ts","sourceRoot":"","sources":["../../src/models/v2evalrunsmonitorop.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,eAAO,MAAM,MAAM;;;CAGT,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,MAAM,GAAG,UAAU,CAAC,OAAO,MAAM,CAAC,CAAC;AAE/C,eAAO,MAAM,gBAAgB;;;EAGuC,CAAC;AAErE,MAAM,MAAM,wBAAwB,GAAG;IACrC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,GAAG,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC1B,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC/B,CAAC;AAEF,eAAO,MAAM,kCAAkC,EAAE,CAAC,CAAC,OAAO,CACxD,wBAAwB,CAaxB,CAAC;AAEH,MAAM,MAAM,yBAAyB,GAAG,MAAM,GAAG,mBAAmB,CAAC;AAErE,eAAO,MAAM,mCAAmC,EAAE,CAAC,CAAC,OAAO,CACzD,yBAAyB,CAMzB,CAAC"}
@@ -0,0 +1,29 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: f4df01489d97
4
+ */
5
+ import * as z from "zod";
6
+ import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
7
+ /**
8
+ * Reduced activity or the full sanitized durable trace.
9
+ */
10
+ export const Detail = {
11
+ Summary: "summary",
12
+ Verbose: "verbose",
13
+ };
14
+ export const Detail$zodSchema = z.enum([
15
+ "summary",
16
+ "verbose",
17
+ ]).describe("Reduced activity or the full sanitized durable trace.");
18
+ export const V2EvalRunsMonitorRequest$zodSchema = z.object({
19
+ after_id: z.int().default(0).describe("Durable child-event cursor used when reconnecting."),
20
+ all: z.boolean().default(false).describe("Capture all currently active public v2 eval runs."),
21
+ detail: Detail$zodSchema.default("summary").describe("Reduced activity or the full sanitized durable trace."),
22
+ run_id: z.string().describe("Parent run ID returned by `POST /v2/eval-runs`.")
23
+ .optional(),
24
+ });
25
+ export const V2EvalRunsMonitorResponse$zodSchema = z.union([
26
+ z.string().describe("A server-sent event stream of eval-run status and activity."),
27
+ HTTPValidationError$zodSchema,
28
+ ]);
29
+ //# sourceMappingURL=v2evalrunsmonitorop.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunsmonitorop.js","sourceRoot":"","sources":["../../src/models/v2evalrunsmonitorop.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,MAAM,CAAC,MAAM,MAAM,GAAG;IACpB,OAAO,EAAE,SAAS;IAClB,OAAO,EAAE,SAAS;CACV,CAAC;AAMX,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC,IAAI,CAAC;IACrC,SAAS;IACT,SAAS;CACV,CAAC,CAAC,QAAQ,CAAC,uDAAuD,CAAC,CAAC;AASrE,MAAM,CAAC,MAAM,kCAAkC,GAE3C,CAAC,CAAC,MAAM,CAAC;IACX,QAAQ,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,QAAQ,CACnC,oDAAoD,CACrD;IACD,GAAG,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CACtC,mDAAmD,CACpD;IACD,MAAM,EAAE,gBAAgB,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,QAAQ,CAClD,uDAAuD,CACxD;IACD,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,iDAAiD,CAAC;SAC3E,QAAQ,EAAE;CACd,CAAC,CAAC;AAIH,MAAM,CAAC,MAAM,mCAAmC,GAE5C,CAAC,CAAC,KAAK,CAAC;IACV,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACjB,6DAA6D,CAC9D;IACD,6BAA6B;CAC9B,CAAC,CAAC"}
@@ -20,13 +20,13 @@ export declare const V2ListTasksResponseStatus: {
20
20
  export type V2ListTasksResponseStatus = ClosedEnum<typeof V2ListTasksResponseStatus>;
21
21
  export declare const V2ListTasksResponseStatus$zodSchema: z.ZodEnum<{
22
22
  completed: "completed";
23
+ all: "all";
23
24
  active: "active";
24
25
  todo: "todo";
25
26
  new: "new";
26
27
  in_progress: "in_progress";
27
28
  done: "done";
28
29
  dismissed: "dismissed";
29
- all: "all";
30
30
  }>;
31
31
  export type V2ListTasksResponse = {
32
32
  action_date: string;
@@ -15,13 +15,13 @@ export declare const V2TasksListStatus: {
15
15
  export type V2TasksListStatus = ClosedEnum<typeof V2TasksListStatus>;
16
16
  export declare const V2TasksListStatus$zodSchema: z.ZodEnum<{
17
17
  completed: "completed";
18
+ all: "all";
18
19
  active: "active";
19
20
  todo: "todo";
20
21
  new: "new";
21
22
  in_progress: "in_progress";
22
23
  done: "done";
23
24
  dismissed: "dismissed";
24
- all: "all";
25
25
  }>;
26
26
  export type V2TasksListRequest = {
27
27
  action_date?: string | undefined;
@@ -1 +1 @@
1
- {"version":3,"file":"tool-names.d.ts","sourceRoot":"","sources":["../src/tool-names.ts"],"names":[],"mappings":"AACA,eAAO,MAAM,SAAS,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAiNlE,CAAC"}
1
+ {"version":3,"file":"tool-names.d.ts","sourceRoot":"","sources":["../src/tool-names.ts"],"names":[],"mappings":"AACA,eAAO,MAAM,SAAS,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAqNlE,CAAC"}
package/esm/tool-names.js CHANGED
@@ -149,12 +149,16 @@ export const toolNames = [
149
149
  "description": "Delete Configuration\n\nPermanently deletes an eval configuration and returns its last saved state."
150
150
  },
151
151
  {
152
- "name": "eval-runs-v2-eval-runs-list",
153
- "description": "List Runs\n\nReturns jobs created through `POST /v2/eval-runs`, ordered newest first, with lifecycle state and progress."
152
+ "name": "evals-run",
153
+ "description": "Create Run\n\nQueues a durable eval job and returns immediately with a polling ID. Poll `GET /v2/eval-runs/{run_id}` until its status reaches a terminal value. Choose exactly one request mode. For saved mode, provide `eval_id`, `configuration_id`, and `agent_id`. For direct mode, omit `eval_id` and provide `prompt`, `judges`, and `agent_id`. Direct runs may also provide a repository, write-only environment variables, and public skill packages."
154
+ },
155
+ {
156
+ "name": "evals-monitor",
157
+ "description": "Monitor Runs\n\nStreams an immediate snapshot followed by live status and sanitized activity for one public v2 eval run or the active runs captured when the connection opens. Exactly one of `run_id` or `all=true` is required."
154
158
  },
155
159
  {
156
- "name": "eval-runs-v2-eval-runs-create",
157
- "description": "Create Run\n\nQueues a new eval job for one configuration and agent, and returns immediately with a polling ID."
160
+ "name": "eval-runs-v2-eval-runs-list",
161
+ "description": "List Runs\n\nReturns jobs created through `POST /v2/eval-runs`, ordered newest first, with lifecycle state and progress."
158
162
  },
159
163
  {
160
164
  "name": "eval-runs-v2-eval-runs-retrieve",
@@ -1 +1 @@
1
- {"version":3,"file":"tool-names.js","sourceRoot":"","sources":["../src/tool-names.ts"],"names":[],"mappings":"AAAA,+BAA+B;AAC/B,MAAM,CAAC,MAAM,SAAS,GAAgD;IACpE;QACE,MAAM,EAAE,sBAAsB;QAC9B,aAAa,EAAE,6GAA6G;KAC7H;IACD;QACE,MAAM,EAAE,2BAA2B;QACnC,aAAa,EAAE,0HAA0H;KAC1I;IACD;QACE,MAAM,EAAE,2CAA2C;QACnD,aAAa,EAAE,oOAAoO;KACpP;IACD;QACE,MAAM,EAAE,4CAA4C;QACpD,aAAa,EAAE,oKAAoK;KACpL;IACD;QACE,MAAM,EAAE,yCAAyC;QACjD,aAAa,EAAE,iKAAiK;KACjL;IACD;QACE,MAAM,EAAE,2CAA2C;QACnD,aAAa,EAAE,wKAAwK;KACxL;IACD;QACE,MAAM,EAAE,yBAAyB;QACjC,aAAa,EAAE,qHAAqH;KACrI;IACD;QACE,MAAM,EAAE,2BAA2B;QACnC,aAAa,EAAE,6JAA6J;KAC7K;IACD;QACE,MAAM,EAAE,+BAA+B;QACvC,aAAa,EAAE,yEAAyE;KACzF;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,qEAAqE;KACrF;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,uEAAuE;KACvF;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,wJAAwJ;KACxK;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,wJAAwJ;KACxK;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,6FAA6F;KAC7G;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,4FAA4F;KAC5G;IACD;QACE,MAAM,EAAE,2BAA2B;QACnC,aAAa,EAAE,mHAAmH;KACnI;IACD;QACE,MAAM,EAAE,2BAA2B;QACnC,aAAa,EAAE,qHAAqH;KACrI;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,iMAAiM;KACjN;IACD;QACE,MAAM,EAAE,yCAAyC;QACjD,aAAa,EAAE,kHAAkH;KAClI;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,0IAA0I;KAC1J;IACD;QACE,MAAM,EAAE,yBAAyB;QACjC,aAAa,EAAE,uEAAuE;KACvF;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,4IAA4I;KAC5J;IACD;QACE,MAAM,EAAE,kCAAkC;QAC1C,aAAa,EAAE,6KAA6K;KAC7L;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,4HAA4H;KAC5I;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,uGAAuG;KACvH;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,6NAA6N;KAC7O;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,uHAAuH;KACvI;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,iHAAiH;KACjI;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,oHAAoH;KACpI;IACD;QACE,MAAM,EAAE,yBAAyB;QACjC,aAAa,EAAE,gEAAgE;KAChF;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,yGAAyG;KACzH;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,iFAAiF;KACjG;IACD;QACE,MAAM,EAAE,mCAAmC;QAC3C,aAAa,EAAE,sIAAsI;KACtJ;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,sIAAsI;KACtJ;IACD;QACE,MAAM,EAAE,uCAAuC;QAC/C,aAAa,EAAE,wEAAwE;KACxF;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,yFAAyF;KACzG;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,qGAAqG;KACrH;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,0HAA0H;KAC1I;IACD;QACE,MAAM,EAAE,+BAA+B;QACvC,aAAa,EAAE,iHAAiH;KACjI;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,iHAAiH;KACjI;IACD;QACE,MAAM,EAAE,mCAAmC;QAC3C,aAAa,EAAE,gKAAgK;KAChL;IACD;QACE,MAAM,EAAE,uCAAuC;QAC/C,aAAa,EAAE,qGAAqG;KACrH;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,8GAA8G;KAC9H;IACD;QACE,MAAM,EAAE,mCAAmC;QAC3C,aAAa,EAAE,yGAAyG;KACzH;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,wFAAwF;KACxG;IACD;QACE,MAAM,EAAE,uCAAuC;QAC/C,aAAa,EAAE,mEAAmE;KACnF;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,uFAAuF;KACvG;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,mGAAmG;KACnH;IACD;QACE,MAAM,EAAE,mCAAmC;QAC3C,aAAa,EAAE,oFAAoF;KACpG;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,iGAAiG;KACjH;IACD;QACE,MAAM,EAAE,oBAAoB;QAC5B,aAAa,EAAE,qFAAqF;KACrG;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,wFAAwF;KACxG;CACF,CAAC"}
1
+ {"version":3,"file":"tool-names.js","sourceRoot":"","sources":["../src/tool-names.ts"],"names":[],"mappings":"AAAA,+BAA+B;AAC/B,MAAM,CAAC,MAAM,SAAS,GAAgD;IACpE;QACE,MAAM,EAAE,sBAAsB;QAC9B,aAAa,EAAE,6GAA6G;KAC7H;IACD;QACE,MAAM,EAAE,2BAA2B;QACnC,aAAa,EAAE,0HAA0H;KAC1I;IACD;QACE,MAAM,EAAE,2CAA2C;QACnD,aAAa,EAAE,oOAAoO;KACpP;IACD;QACE,MAAM,EAAE,4CAA4C;QACpD,aAAa,EAAE,oKAAoK;KACpL;IACD;QACE,MAAM,EAAE,yCAAyC;QACjD,aAAa,EAAE,iKAAiK;KACjL;IACD;QACE,MAAM,EAAE,2CAA2C;QACnD,aAAa,EAAE,wKAAwK;KACxL;IACD;QACE,MAAM,EAAE,yBAAyB;QACjC,aAAa,EAAE,qHAAqH;KACrI;IACD;QACE,MAAM,EAAE,2BAA2B;QACnC,aAAa,EAAE,6JAA6J;KAC7K;IACD;QACE,MAAM,EAAE,+BAA+B;QACvC,aAAa,EAAE,yEAAyE;KACzF;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,qEAAqE;KACrF;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,uEAAuE;KACvF;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,wJAAwJ;KACxK;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,wJAAwJ;KACxK;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,6FAA6F;KAC7G;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,4FAA4F;KAC5G;IACD;QACE,MAAM,EAAE,2BAA2B;QACnC,aAAa,EAAE,mHAAmH;KACnI;IACD;QACE,MAAM,EAAE,2BAA2B;QACnC,aAAa,EAAE,qHAAqH;KACrI;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,iMAAiM;KACjN;IACD;QACE,MAAM,EAAE,yCAAyC;QACjD,aAAa,EAAE,kHAAkH;KAClI;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,0IAA0I;KAC1J;IACD;QACE,MAAM,EAAE,yBAAyB;QACjC,aAAa,EAAE,uEAAuE;KACvF;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,4IAA4I;KAC5J;IACD;QACE,MAAM,EAAE,kCAAkC;QAC1C,aAAa,EAAE,6KAA6K;KAC7L;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,4HAA4H;KAC5I;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,uGAAuG;KACvH;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,6NAA6N;KAC7O;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,uHAAuH;KACvI;IACD;QACE,MAAM,EAAE,qBAAqB;QAC7B,aAAa,EAAE,iHAAiH;KACjI;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,oHAAoH;KACpI;IACD;QACE,MAAM,EAAE,yBAAyB;QACjC,aAAa,EAAE,gEAAgE;KAChF;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,yGAAyG;KACzH;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,iFAAiF;KACjG;IACD;QACE,MAAM,EAAE,mCAAmC;QAC3C,aAAa,EAAE,sIAAsI;KACtJ;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,sIAAsI;KACtJ;IACD;QACE,MAAM,EAAE,uCAAuC;QAC/C,aAAa,EAAE,wEAAwE;KACxF;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,yFAAyF;KACzG;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,qGAAqG;KACrH;IACD;QACE,MAAM,EAAE,WAAW;QACnB,aAAa,EAAE,icAAic;KACjd;IACD;QACE,MAAM,EAAE,eAAe;QACvB,aAAa,EAAE,mOAAmO;KACnP;IACD;QACE,MAAM,EAAE,6BAA6B;QACrC,aAAa,EAAE,0HAA0H;KAC1I;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,iHAAiH;KACjI;IACD;QACE,MAAM,EAAE,mCAAmC;QAC3C,aAAa,EAAE,gKAAgK;KAChL;IACD;QACE,MAAM,EAAE,uCAAuC;QAC/C,aAAa,EAAE,qGAAqG;KACrH;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,8GAA8G;KAC9H;IACD;QACE,MAAM,EAAE,mCAAmC;QAC3C,aAAa,EAAE,yGAAyG;KACzH;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,wFAAwF;KACxG;IACD;QACE,MAAM,EAAE,uCAAuC;QAC/C,aAAa,EAAE,mEAAmE;KACnF;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,uFAAuF;KACvG;IACD;QACE,MAAM,EAAE,qCAAqC;QAC7C,aAAa,EAAE,mGAAmG;KACnH;IACD;QACE,MAAM,EAAE,mCAAmC;QAC3C,aAAa,EAAE,oFAAoF;KACpG;IACD;QACE,MAAM,EAAE,iCAAiC;QACzC,aAAa,EAAE,iGAAiG;KACjH;IACD;QACE,MAAM,EAAE,oBAAoB;QAC5B,aAAa,EAAE,qFAAqF;KACrG;IACD;QACE,MAAM,EAAE,uBAAuB;QAC/B,aAAa,EAAE,wFAAwF;KACxG;CACF,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@lightsage/mcp",
3
- "version": "0.1.11",
3
+ "version": "0.2.0",
4
4
  "author": "Speakeasy",
5
5
  "type": "module",
6
6
  "sideEffects": false,
@@ -0,0 +1,157 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: da8fcf6d681f
4
+ */
5
+
6
+ import { LightsageMcpCore } from "../core.js";
7
+ import { encodeFormQuery } from "../lib/encodings.js";
8
+ import { compactMap } from "../lib/primitives.js";
9
+ import { safeParse } from "../lib/schemas.js";
10
+ import { RequestOptions } from "../lib/sdks.js";
11
+ import { extractSecurity, resolveGlobalSecurity } from "../lib/security.js";
12
+ import { pathToFunc } from "../lib/url.js";
13
+ import { APIError } from "../models/errors/apierror.js";
14
+ import {
15
+ ConnectionError,
16
+ InvalidRequestError,
17
+ RequestAbortedError,
18
+ RequestTimeoutError,
19
+ UnexpectedClientError,
20
+ } from "../models/errors/httpclienterrors.js";
21
+ import { SDKValidationError } from "../models/errors/sdkvalidationerror.js";
22
+ import {
23
+ V2EvalRunsMonitorRequest,
24
+ V2EvalRunsMonitorRequest$zodSchema,
25
+ } from "../models/v2evalrunsmonitorop.js";
26
+ import { APICall, APIPromise } from "../types/async.js";
27
+ import { Result } from "../types/fp.js";
28
+
29
+ export enum MonitorAcceptEnum {
30
+ applicationJsonAccept = "application/json",
31
+ textEventStreamAccept = "text/event-stream",
32
+ }
33
+
34
+ /**
35
+ * Monitor Runs
36
+ *
37
+ * @remarks
38
+ * Streams an immediate snapshot followed by live status and sanitized activity for one public v2 eval run or the active runs captured when the connection opens. Exactly one of `run_id` or `all=true` is required.
39
+ */
40
+ export function evalsMonitor(
41
+ client$: LightsageMcpCore,
42
+ request?: V2EvalRunsMonitorRequest | undefined,
43
+ options?: RequestOptions,
44
+ ): APIPromise<
45
+ Result<
46
+ Response,
47
+ | APIError
48
+ | SDKValidationError
49
+ | UnexpectedClientError
50
+ | InvalidRequestError
51
+ | RequestAbortedError
52
+ | RequestTimeoutError
53
+ | ConnectionError
54
+ >
55
+ > {
56
+ return new APIPromise($do(
57
+ client$,
58
+ request,
59
+ options,
60
+ ));
61
+ }
62
+
63
+ async function $do(
64
+ client$: LightsageMcpCore,
65
+ request?: V2EvalRunsMonitorRequest | undefined,
66
+ options?: RequestOptions & { acceptHeaderOverride?: MonitorAcceptEnum },
67
+ ): Promise<
68
+ [
69
+ Result<
70
+ Response,
71
+ | APIError
72
+ | SDKValidationError
73
+ | UnexpectedClientError
74
+ | InvalidRequestError
75
+ | RequestAbortedError
76
+ | RequestTimeoutError
77
+ | ConnectionError
78
+ >,
79
+ APICall,
80
+ ]
81
+ > {
82
+ const parsed$ = safeParse(
83
+ request,
84
+ (value$) => V2EvalRunsMonitorRequest$zodSchema.optional().parse(value$),
85
+ "Input validation failed",
86
+ );
87
+ if (!parsed$.ok) {
88
+ return [parsed$, { status: "invalid" }];
89
+ }
90
+ const payload$ = parsed$.value;
91
+ const body$ = null;
92
+ const path$ = pathToFunc("/v2/eval-runs/monitor")();
93
+ const query$ = encodeFormQuery({
94
+ "after_id": payload$?.after_id,
95
+ "all": payload$?.all,
96
+ "detail": payload$?.detail,
97
+ "run_id": payload$?.run_id,
98
+ });
99
+
100
+ const headers$ = new Headers(compactMap({
101
+ Accept: options?.acceptHeaderOverride
102
+ || "application/json;q=1, text/event-stream;q=0",
103
+ }));
104
+ const securityInput = await extractSecurity(client$._options.security);
105
+ const requestSecurity = resolveGlobalSecurity(securityInput);
106
+
107
+ const context = {
108
+ options: client$._options,
109
+ baseURL: options?.serverURL ?? client$._baseURL ?? "",
110
+ operationID: "v2_eval_runs_monitor",
111
+ oAuth2Scopes: null,
112
+ resolvedSecurity: requestSecurity,
113
+ securitySource: client$._options.security,
114
+ retryConfig: options?.retries
115
+ || client$._options.retryConfig
116
+ || { strategy: "none" },
117
+ retryCodes: options?.retryCodes || [
118
+ "429",
119
+ "500",
120
+ "502",
121
+ "503",
122
+ "504",
123
+ ],
124
+ };
125
+
126
+ const requestRes = client$._createRequest(context, {
127
+ security: requestSecurity,
128
+ method: "GET",
129
+ baseURL: options?.serverURL,
130
+ path: path$,
131
+ headers: headers$,
132
+ query: query$,
133
+ body: body$,
134
+ userAgent: client$._options.userAgent,
135
+ timeoutMs: options?.timeoutMs || client$._options.timeoutMs
136
+ || -1,
137
+ }, options);
138
+ if (!requestRes.ok) {
139
+ return [requestRes, { status: "invalid" }];
140
+ }
141
+ const req$ = requestRes.value;
142
+
143
+ const doResult = await client$._do(req$, {
144
+ context,
145
+ errorCodes: [],
146
+ retryConfig: context.retryConfig,
147
+ retryCodes: context.retryCodes,
148
+ });
149
+ if (!doResult.ok) {
150
+ return [doResult, { status: "request-error", request: req$ }];
151
+ }
152
+ return [doResult, {
153
+ status: "complete",
154
+ "request": req$,
155
+ response: doResult.value,
156
+ }];
157
+ }
@@ -1,6 +1,6 @@
1
1
  /*
2
2
  * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
- * @generated-id: 2f3e997d0ec6
3
+ * @generated-id: 777689f28553
4
4
  */
5
5
 
6
6
  import { LightsageMcpCore } from "../core.js";
@@ -30,9 +30,9 @@ import { Result } from "../types/fp.js";
30
30
  * Create Run
31
31
  *
32
32
  * @remarks
33
- * Queues a new eval job for one configuration and agent, and returns immediately with a polling ID.
33
+ * Queues a durable eval job and returns immediately with a polling ID. Poll `GET /v2/eval-runs/{run_id}` until its status reaches a terminal value. Choose exactly one request mode. For saved mode, provide `eval_id`, `configuration_id`, and `agent_id`. For direct mode, omit `eval_id` and provide `prompt`, `judges`, and `agent_id`. Direct runs may also provide a repository, write-only environment variables, and public skill packages.
34
34
  */
35
- export function evalRunsV2EvalRunsCreate(
35
+ export function evalsRun(
36
36
  client$: LightsageMcpCore,
37
37
  request: CreateV2EvalRunRequest,
38
38
  options?: RequestOptions,
@@ -931,7 +931,7 @@ http_headers = { "lightsage-api-key-auth" = "YOUR_LIGHTSAGE_API_KEY_AUTH" }`;
931
931
  <h1>Instructions</h1>
932
932
  <p>One-click installation for Claude Desktop users</p>
933
933
  <div class="instruction-item">
934
- <a href="https://github.com/lightsagehq/mcp/releases/download/v0.1.11/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
934
+ <a href="https://github.com/lightsagehq/mcp/releases/download/v0.2.0/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
935
935
  📥 Download MCP Bundle
936
936
  </a>
937
937
  </div>
package/src/lib/config.ts CHANGED
@@ -66,7 +66,7 @@ export function serverURLFromOptions(options: SDKOptions): URL | null {
66
66
  export const SDK_METADATA = {
67
67
  language: "typescript",
68
68
  openapiDocVersion: "1.0.0",
69
- sdkVersion: "0.1.11",
69
+ sdkVersion: "0.2.0",
70
70
  genVersion: "2.935.1",
71
- userAgent: "speakeasy-sdk/mcp-typescript 0.1.11 2.935.1 1.0.0 @lightsage/mcp",
71
+ userAgent: "speakeasy-sdk/mcp-typescript 0.2.0 2.935.1 1.0.0 @lightsage/mcp",
72
72
  } as const;
@@ -22,7 +22,7 @@ const routes = buildRouteMap({
22
22
  export const app = buildApplication(routes, {
23
23
  name: "mcp",
24
24
  versionInfo: {
25
- currentVersion: "0.1.11",
25
+ currentVersion: "0.2.0",
26
26
  },
27
27
  });
28
28
 
@@ -26,9 +26,10 @@ import { tool$contentV2ContentRetrieve } from "./tools/contentV2ContentRetrieve.
26
26
  import { tool$environmentsV2EnvironmentsList } from "./tools/environmentsV2EnvironmentsList.js";
27
27
  import { tool$evalResultsV2EvalResultsList } from "./tools/evalResultsV2EvalResultsList.js";
28
28
  import { tool$evalResultsV2EvalResultsRetrieve } from "./tools/evalResultsV2EvalResultsRetrieve.js";
29
- import { tool$evalRunsV2EvalRunsCreate } from "./tools/evalRunsV2EvalRunsCreate.js";
30
29
  import { tool$evalRunsV2EvalRunsList } from "./tools/evalRunsV2EvalRunsList.js";
31
30
  import { tool$evalRunsV2EvalRunsRetrieve } from "./tools/evalRunsV2EvalRunsRetrieve.js";
31
+ import { tool$evalsMonitor } from "./tools/evalsMonitor.js";
32
+ import { tool$evalsRun } from "./tools/evalsRun.js";
32
33
  import { tool$evalsV2EvalConfigurationsCreate } from "./tools/evalsV2EvalConfigurationsCreate.js";
33
34
  import { tool$evalsV2EvalConfigurationsDelete } from "./tools/evalsV2EvalConfigurationsDelete.js";
34
35
  import { tool$evalsV2EvalConfigurationsList } from "./tools/evalsV2EvalConfigurationsList.js";
@@ -90,7 +91,7 @@ export function createMCPServer(deps: {
90
91
  sizes: ["any"],
91
92
  }],
92
93
  name: "LightsageMcp",
93
- version: "0.1.11",
94
+ version: "0.2.0",
94
95
  });
95
96
 
96
97
  const getClient = deps.getSDK || (() =>
@@ -172,8 +173,9 @@ export function createMCPServer(deps: {
172
173
  tool(tool$evalsV2EvalConfigurationsRetrieve);
173
174
  tool(tool$evalsV2EvalConfigurationsUpdate);
174
175
  tool(tool$evalsV2EvalConfigurationsDelete);
176
+ tool(tool$evalsRun);
177
+ tool(tool$evalsMonitor);
175
178
  tool(tool$evalRunsV2EvalRunsList);
176
- tool(tool$evalRunsV2EvalRunsCreate);
177
179
  tool(tool$evalRunsV2EvalRunsRetrieve);
178
180
  tool(tool$evalResultsV2EvalResultsList);
179
181
  tool(tool$evalResultsV2EvalResultsRetrieve);
@@ -0,0 +1,43 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: f7c3344a66fd
4
+ */
5
+
6
+ import { evalsMonitor } from "../../funcs/evalsMonitor.js";
7
+ import { V2EvalRunsMonitorRequest$zodSchema } from "../../models/v2evalrunsmonitorop.js";
8
+ import { formatResult, ToolDefinition } from "../tools.js";
9
+
10
+ const args = {
11
+ request: V2EvalRunsMonitorRequest$zodSchema.optional(),
12
+ };
13
+
14
+ export const tool$evalsMonitor: ToolDefinition<typeof args> = {
15
+ name: "evals-monitor",
16
+ description: `Monitor Runs
17
+
18
+ Streams an immediate snapshot followed by live status and sanitized activity for one public v2 eval run or the active runs captured when the connection opens. Exactly one of \`run_id\` or \`all=true\` is required.`,
19
+ annotations: {
20
+ "title": "",
21
+ "destructiveHint": false,
22
+ "idempotentHint": false,
23
+ "openWorldHint": false,
24
+ "readOnlyHint": true,
25
+ },
26
+ args,
27
+ tool: async (client, args, ctx) => {
28
+ const [result] = await evalsMonitor(
29
+ client,
30
+ args.request,
31
+ { fetchOptions: { signal: ctx.signal } },
32
+ ).$inspect();
33
+
34
+ if (!result.ok) {
35
+ return {
36
+ content: [{ type: "text", text: result.error.message }],
37
+ isError: true,
38
+ };
39
+ }
40
+
41
+ return formatResult(result.value);
42
+ },
43
+ };
@@ -1,9 +1,9 @@
1
1
  /*
2
2
  * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
- * @generated-id: 4c15182fbf87
3
+ * @generated-id: a4afa8b6d32d
4
4
  */
5
5
 
6
- import { evalRunsV2EvalRunsCreate } from "../../funcs/evalRunsV2EvalRunsCreate.js";
6
+ import { evalsRun } from "../../funcs/evalsRun.js";
7
7
  import { CreateV2EvalRunRequest$zodSchema } from "../../models/createv2evalrunrequest.js";
8
8
  import { formatResult, ToolDefinition } from "../tools.js";
9
9
 
@@ -11,11 +11,11 @@ const args = {
11
11
  request: CreateV2EvalRunRequest$zodSchema,
12
12
  };
13
13
 
14
- export const tool$evalRunsV2EvalRunsCreate: ToolDefinition<typeof args> = {
15
- name: "eval-runs-v2-eval-runs-create",
14
+ export const tool$evalsRun: ToolDefinition<typeof args> = {
15
+ name: "evals-run",
16
16
  description: `Create Run
17
17
 
18
- Queues a new eval job for one configuration and agent, and returns immediately with a polling ID.`,
18
+ Queues a durable eval job and returns immediately with a polling ID. Poll \`GET /v2/eval-runs/{run_id}\` until its status reaches a terminal value. Choose exactly one request mode. For saved mode, provide \`eval_id\`, \`configuration_id\`, and \`agent_id\`. For direct mode, omit \`eval_id\` and provide \`prompt\`, \`judges\`, and \`agent_id\`. Direct runs may also provide a repository, write-only environment variables, and public skill packages.`,
19
19
  annotations: {
20
20
  "title": "",
21
21
  "destructiveHint": false,
@@ -25,7 +25,7 @@ Queues a new eval job for one configuration and agent, and returns immediately w
25
25
  },
26
26
  args,
27
27
  tool: async (client, args, ctx) => {
28
- const [result] = await evalRunsV2EvalRunsCreate(
28
+ const [result] = await evalsRun(
29
29
  client,
30
30
  args.request,
31
31
  { fetchOptions: { signal: ctx.signal } },
@@ -12,6 +12,7 @@ export type CreateV2EvalConfigurationRequest = {
12
12
  mcp_server_ids?: Array<string> | undefined;
13
13
  cli_ids?: Array<string> | undefined;
14
14
  skill_ids?: Array<string> | undefined;
15
+ skills?: Array<string> | undefined;
15
16
  include_docs?: boolean | undefined;
16
17
  };
17
18
 
@@ -25,4 +26,7 @@ export const CreateV2EvalConfigurationRequest$zodSchema: z.ZodType<
25
26
  name: z.string(),
26
27
  repository_id: z.string().optional(),
27
28
  skill_ids: z.array(z.string()).optional(),
29
+ skills: z.array(z.string()).optional().describe(
30
+ "Public skill packages saved with this configuration, using GitHub `owner/repository` references such as `resend/resend-skills`.",
31
+ ),
28
32
  });
@@ -6,15 +6,39 @@
6
6
  import * as z from "zod";
7
7
 
8
8
  export type CreateV2EvalRunRequest = {
9
- eval_id: string;
10
- configuration_id: string;
9
+ eval_id?: string | undefined;
10
+ configuration_id?: string | undefined;
11
11
  agent_id: string;
12
+ prompt?: string | undefined;
13
+ judges?: Array<string> | undefined;
14
+ repository?: string | undefined;
15
+ env?: { [k: string]: string } | undefined;
16
+ skills?: Array<string> | undefined;
12
17
  };
13
18
 
14
19
  export const CreateV2EvalRunRequest$zodSchema: z.ZodType<
15
20
  CreateV2EvalRunRequest
16
21
  > = z.object({
17
22
  agent_id: z.string().describe("Runnable agent ID from `GET /v2/agents`."),
18
- configuration_id: z.string(),
19
- eval_id: z.string(),
23
+ configuration_id: z.string().optional().describe(
24
+ "Saved runtime configuration. Required when `eval_id` is provided.",
25
+ ),
26
+ env: z.record(z.string(), z.string()).optional().describe(
27
+ "Write-only environment variables for a direct run. Values are stored only in the queued job's private configuration.",
28
+ ),
29
+ eval_id: z.string().optional().describe(
30
+ "Existing eval definition to run. Omit to create and run an eval directly.",
31
+ ),
32
+ judges: z.array(z.string()).optional().describe(
33
+ "Success criteria for a direct create-and-run request. CLI clients may expose this string array as a repeatable `--judge` flag.",
34
+ ),
35
+ prompt: z.string().optional().describe(
36
+ "Task prompt for a direct create-and-run request.",
37
+ ),
38
+ repository: z.string().optional().describe(
39
+ "Repository for a direct run: a saved repository ID, built-in alias such as `nextjs-starter`, or a public GitHub HTTPS URL.",
40
+ ),
41
+ skills: z.array(z.string()).optional().describe(
42
+ "Public skill packages to install before the agent starts, using GitHub `owner/repository` references such as `resend/resend-skills`.",
43
+ ),
20
44
  });
@@ -12,6 +12,7 @@ export type UpdateV2EvalConfigurationRequest = {
12
12
  mcp_server_ids?: Array<string> | null | undefined;
13
13
  cli_ids?: Array<string> | null | undefined;
14
14
  skill_ids?: Array<string> | null | undefined;
15
+ skills?: Array<string> | null | undefined;
15
16
  include_docs?: boolean | undefined;
16
17
  };
17
18
 
@@ -25,4 +26,7 @@ export const UpdateV2EvalConfigurationRequest$zodSchema: z.ZodType<
25
26
  name: z.string().optional(),
26
27
  repository_id: z.string().optional(),
27
28
  skill_ids: z.array(z.string()).nullable().optional(),
29
+ skills: z.array(z.string()).nullable().optional().describe(
30
+ "Public skill packages saved with this configuration, using GitHub `owner/repository` references.",
31
+ ),
28
32
  });
@@ -14,6 +14,7 @@ export type V2EvalConfiguration = {
14
14
  mcp_server_ids?: Array<string> | undefined;
15
15
  cli_ids?: Array<string> | undefined;
16
16
  skill_ids?: Array<string> | undefined;
17
+ skills?: Array<string> | undefined;
17
18
  include_docs?: boolean | undefined;
18
19
  created_at?: string | null | undefined;
19
20
  updated_at?: string | null | undefined;
@@ -31,5 +32,8 @@ export const V2EvalConfiguration$zodSchema: z.ZodType<V2EvalConfiguration> = z
31
32
  name: z.string(),
32
33
  repository_id: z.string().nullable().optional(),
33
34
  skill_ids: z.array(z.string()).optional(),
35
+ skills: z.array(z.string()).optional().describe(
36
+ "Public GitHub `owner/repository` skill packages installed for this configuration.",
37
+ ),
34
38
  updated_at: z.string().nullable().optional(),
35
39
  });
@@ -13,7 +13,10 @@ export type V2EvalRun = {
13
13
  id: string;
14
14
  job_id: string;
15
15
  eval_id?: string | null | undefined;
16
+ eval_name?: string | null | undefined;
16
17
  configuration_id?: string | null | undefined;
18
+ agent_id?: string | null | undefined;
19
+ agent_name?: string | null | undefined;
17
20
  status: string;
18
21
  stage?: string | null | undefined;
19
22
  dispatch_id?: string | null | undefined;
@@ -26,15 +29,26 @@ export type V2EvalRun = {
26
29
  };
27
30
 
28
31
  export const V2EvalRun$zodSchema: z.ZodType<V2EvalRun> = z.object({
32
+ agent_id: z.string().nullable().optional().describe(
33
+ "Runnable agent ID selected for this run.",
34
+ ),
35
+ agent_name: z.string().nullable().optional().describe(
36
+ "Display name captured for the selected agent.",
37
+ ),
29
38
  completed_at: z.string().nullable().optional(),
30
39
  configuration_id: z.string().nullable().optional(),
31
40
  created_at: z.string().nullable().optional(),
32
41
  dispatch_id: z.string().nullable().optional(),
33
42
  error: z.string().nullable().optional(),
34
43
  estimated_cost: z.number().nullable().optional(),
35
- eval_id: z.string().nullable().optional(),
36
- id: z.string(),
37
- job_id: z.string(),
44
+ eval_id: z.string().nullable().optional().describe(
45
+ "Saved eval definition ID, including one created by a direct run.",
46
+ ),
47
+ eval_name: z.string().nullable().optional().describe(
48
+ "Display name captured for the eval when the run was created.",
49
+ ),
50
+ id: z.string().describe("Run ID to use when polling eval-run status."),
51
+ job_id: z.string().describe("Alias of `id` for the durable parent job."),
38
52
  progress: PublicEvalRunProgress$zodSchema.optional(),
39
53
  stage: z.string().nullable().optional(),
40
54
  started_at: z.string().nullable().optional(),