scorecard-ai-mcp 1.0.0-alpha.9 → 1.1.0
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.
- package/README.md +19 -14
- package/compat.d.mts +2 -0
- package/compat.d.mts.map +1 -1
- package/compat.d.ts +2 -0
- package/compat.d.ts.map +1 -1
- package/compat.js +38 -1
- package/compat.js.map +1 -1
- package/compat.mjs +37 -0
- package/compat.mjs.map +1 -1
- package/index.js.map +1 -1
- package/index.mjs.map +1 -1
- package/options.d.mts +9 -6
- package/options.d.mts.map +1 -1
- package/options.d.ts +9 -6
- package/options.d.ts.map +1 -1
- package/options.js +5 -46
- package/options.js.map +1 -1
- package/options.mjs +5 -46
- package/options.mjs.map +1 -1
- package/package.json +4 -4
- package/server.d.mts +16 -2
- package/server.d.mts.map +1 -1
- package/server.d.ts +16 -2
- package/server.d.ts.map +1 -1
- package/server.js +28 -10
- package/server.js.map +1 -1
- package/server.mjs +28 -11
- package/server.mjs.map +1 -1
- package/src/compat.ts +40 -0
- package/src/index.ts +2 -2
- package/src/options.ts +12 -54
- package/src/server.ts +40 -15
- package/src/tools/index.ts +13 -8
- package/src/tools/metrics/create-metrics.ts +252 -0
- package/src/tools/metrics/update-metrics.ts +252 -0
- package/src/tools/runs/create-runs.ts +2 -2
- package/src/tools/systems/delete-systems.ts +1 -1
- package/src/tools/systems/versions/create-systems-versions.ts +45 -0
- package/src/tools/{system-configs/get-system-configs.ts → systems/versions/get-systems-versions.ts} +9 -9
- package/src/tools/{system-configs/list-system-configs.ts → systems/versions/list-systems-versions.ts} +6 -6
- package/src/tools/testsets/get-testsets.ts +1 -1
- package/src/tools/types.ts +0 -1
- package/tools/index.d.mts.map +1 -1
- package/tools/index.d.ts.map +1 -1
- package/tools/index.js +13 -8
- package/tools/index.js.map +1 -1
- package/tools/index.mjs +13 -8
- package/tools/index.mjs.map +1 -1
- package/tools/{system-configs/list-system-configs.d.mts → metrics/create-metrics.d.mts} +10 -1
- package/tools/metrics/create-metrics.d.mts.map +1 -0
- package/tools/{system-configs/create-system-configs.d.ts → metrics/create-metrics.d.ts} +10 -1
- package/tools/metrics/create-metrics.d.ts.map +1 -0
- package/tools/metrics/create-metrics.js +245 -0
- package/tools/metrics/create-metrics.js.map +1 -0
- package/tools/metrics/create-metrics.mjs +241 -0
- package/tools/metrics/create-metrics.mjs.map +1 -0
- package/tools/{system-configs/get-system-configs.d.mts → metrics/update-metrics.d.mts} +10 -1
- package/tools/metrics/update-metrics.d.mts.map +1 -0
- package/tools/{system-configs/get-system-configs.d.ts → metrics/update-metrics.d.ts} +10 -1
- package/tools/metrics/update-metrics.d.ts.map +1 -0
- package/tools/metrics/update-metrics.js +245 -0
- package/tools/metrics/update-metrics.js.map +1 -0
- package/tools/metrics/update-metrics.mjs +241 -0
- package/tools/metrics/update-metrics.mjs.map +1 -0
- package/tools/projects/create-projects.d.mts +9 -0
- package/tools/projects/create-projects.d.mts.map +1 -1
- package/tools/projects/create-projects.d.ts +9 -0
- package/tools/projects/create-projects.d.ts.map +1 -1
- package/tools/projects/list-projects.d.mts +9 -0
- package/tools/projects/list-projects.d.mts.map +1 -1
- package/tools/projects/list-projects.d.ts +9 -0
- package/tools/projects/list-projects.d.ts.map +1 -1
- package/tools/records/create-records.d.mts +9 -0
- package/tools/records/create-records.d.mts.map +1 -1
- package/tools/records/create-records.d.ts +9 -0
- package/tools/records/create-records.d.ts.map +1 -1
- package/tools/runs/create-runs.d.mts +9 -0
- package/tools/runs/create-runs.d.mts.map +1 -1
- package/tools/runs/create-runs.d.ts +9 -0
- package/tools/runs/create-runs.d.ts.map +1 -1
- package/tools/runs/create-runs.js +2 -2
- package/tools/runs/create-runs.js.map +1 -1
- package/tools/runs/create-runs.mjs +2 -2
- package/tools/runs/create-runs.mjs.map +1 -1
- package/tools/scores/upsert-scores.d.mts +9 -0
- package/tools/scores/upsert-scores.d.mts.map +1 -1
- package/tools/scores/upsert-scores.d.ts +9 -0
- package/tools/scores/upsert-scores.d.ts.map +1 -1
- package/tools/systems/create-systems.d.mts +9 -0
- package/tools/systems/create-systems.d.mts.map +1 -1
- package/tools/systems/create-systems.d.ts +9 -0
- package/tools/systems/create-systems.d.ts.map +1 -1
- package/tools/systems/delete-systems.d.mts +9 -0
- package/tools/systems/delete-systems.d.mts.map +1 -1
- package/tools/systems/delete-systems.d.ts +9 -0
- package/tools/systems/delete-systems.d.ts.map +1 -1
- package/tools/systems/delete-systems.js +1 -1
- package/tools/systems/delete-systems.js.map +1 -1
- package/tools/systems/delete-systems.mjs +1 -1
- package/tools/systems/delete-systems.mjs.map +1 -1
- package/tools/systems/get-systems.d.mts +9 -0
- package/tools/systems/get-systems.d.mts.map +1 -1
- package/tools/systems/get-systems.d.ts +9 -0
- package/tools/systems/get-systems.d.ts.map +1 -1
- package/tools/systems/list-systems.d.mts +9 -0
- package/tools/systems/list-systems.d.mts.map +1 -1
- package/tools/systems/list-systems.d.ts +9 -0
- package/tools/systems/list-systems.d.ts.map +1 -1
- package/tools/systems/update-systems.d.mts +9 -0
- package/tools/systems/update-systems.d.mts.map +1 -1
- package/tools/systems/update-systems.d.ts +9 -0
- package/tools/systems/update-systems.d.ts.map +1 -1
- package/tools/systems/versions/create-systems-versions.d.mts +41 -0
- package/tools/systems/versions/create-systems-versions.d.mts.map +1 -0
- package/tools/systems/versions/create-systems-versions.d.ts +41 -0
- package/tools/systems/versions/create-systems-versions.d.ts.map +1 -0
- package/tools/systems/versions/create-systems-versions.js +40 -0
- package/tools/systems/versions/create-systems-versions.js.map +1 -0
- package/tools/systems/versions/create-systems-versions.mjs +36 -0
- package/tools/systems/versions/create-systems-versions.mjs.map +1 -0
- package/tools/systems/versions/get-systems-versions.d.mts +41 -0
- package/tools/systems/versions/get-systems-versions.d.mts.map +1 -0
- package/tools/{system-configs/create-system-configs.d.mts → systems/versions/get-systems-versions.d.ts} +11 -2
- package/tools/systems/versions/get-systems-versions.d.ts.map +1 -0
- package/tools/{system-configs/get-system-configs.js → systems/versions/get-systems-versions.js} +9 -9
- package/tools/systems/versions/get-systems-versions.js.map +1 -0
- package/tools/{system-configs/get-system-configs.mjs → systems/versions/get-systems-versions.mjs} +9 -9
- package/tools/systems/versions/get-systems-versions.mjs.map +1 -0
- package/tools/systems/versions/list-systems-versions.d.mts +41 -0
- package/tools/systems/versions/list-systems-versions.d.mts.map +1 -0
- package/tools/systems/versions/list-systems-versions.d.ts +41 -0
- package/tools/systems/versions/list-systems-versions.d.ts.map +1 -0
- package/tools/{system-configs/list-system-configs.js → systems/versions/list-systems-versions.js} +6 -6
- package/tools/systems/versions/list-systems-versions.js.map +1 -0
- package/tools/{system-configs/list-system-configs.mjs → systems/versions/list-systems-versions.mjs} +6 -6
- package/tools/systems/versions/list-systems-versions.mjs.map +1 -0
- package/tools/testcases/create-testcases.d.mts +9 -0
- package/tools/testcases/create-testcases.d.mts.map +1 -1
- package/tools/testcases/create-testcases.d.ts +9 -0
- package/tools/testcases/create-testcases.d.ts.map +1 -1
- package/tools/testcases/delete-testcases.d.mts +9 -0
- package/tools/testcases/delete-testcases.d.mts.map +1 -1
- package/tools/testcases/delete-testcases.d.ts +9 -0
- package/tools/testcases/delete-testcases.d.ts.map +1 -1
- package/tools/testcases/get-testcases.d.mts +9 -0
- package/tools/testcases/get-testcases.d.mts.map +1 -1
- package/tools/testcases/get-testcases.d.ts +9 -0
- package/tools/testcases/get-testcases.d.ts.map +1 -1
- package/tools/testcases/list-testcases.d.mts +9 -0
- package/tools/testcases/list-testcases.d.mts.map +1 -1
- package/tools/testcases/list-testcases.d.ts +9 -0
- package/tools/testcases/list-testcases.d.ts.map +1 -1
- package/tools/testcases/update-testcases.d.mts +9 -0
- package/tools/testcases/update-testcases.d.mts.map +1 -1
- package/tools/testcases/update-testcases.d.ts +9 -0
- package/tools/testcases/update-testcases.d.ts.map +1 -1
- package/tools/testsets/create-testsets.d.mts +9 -0
- package/tools/testsets/create-testsets.d.mts.map +1 -1
- package/tools/testsets/create-testsets.d.ts +9 -0
- package/tools/testsets/create-testsets.d.ts.map +1 -1
- package/tools/testsets/delete-testsets.d.mts +9 -0
- package/tools/testsets/delete-testsets.d.mts.map +1 -1
- package/tools/testsets/delete-testsets.d.ts +9 -0
- package/tools/testsets/delete-testsets.d.ts.map +1 -1
- package/tools/testsets/get-testsets.d.mts +9 -0
- package/tools/testsets/get-testsets.d.mts.map +1 -1
- package/tools/testsets/get-testsets.d.ts +9 -0
- package/tools/testsets/get-testsets.d.ts.map +1 -1
- package/tools/testsets/get-testsets.js +1 -1
- package/tools/testsets/get-testsets.js.map +1 -1
- package/tools/testsets/get-testsets.mjs +1 -1
- package/tools/testsets/get-testsets.mjs.map +1 -1
- package/tools/testsets/list-testsets.d.mts +9 -0
- package/tools/testsets/list-testsets.d.mts.map +1 -1
- package/tools/testsets/list-testsets.d.ts +9 -0
- package/tools/testsets/list-testsets.d.ts.map +1 -1
- package/tools/testsets/update-testsets.d.mts +9 -0
- package/tools/testsets/update-testsets.d.mts.map +1 -1
- package/tools/testsets/update-testsets.d.ts +9 -0
- package/tools/testsets/update-testsets.d.ts.map +1 -1
- package/tools/types.d.mts.map +1 -1
- package/tools/types.d.ts.map +1 -1
- package/src/tools/system-configs/create-system-configs.ts +0 -64
- package/tools/system-configs/create-system-configs.d.mts.map +0 -1
- package/tools/system-configs/create-system-configs.d.ts.map +0 -1
- package/tools/system-configs/create-system-configs.js +0 -58
- package/tools/system-configs/create-system-configs.js.map +0 -1
- package/tools/system-configs/create-system-configs.mjs +0 -54
- package/tools/system-configs/create-system-configs.mjs.map +0 -1
- package/tools/system-configs/get-system-configs.d.mts.map +0 -1
- package/tools/system-configs/get-system-configs.d.ts.map +0 -1
- package/tools/system-configs/get-system-configs.js.map +0 -1
- package/tools/system-configs/get-system-configs.mjs.map +0 -1
- package/tools/system-configs/list-system-configs.d.mts.map +0 -1
- package/tools/system-configs/list-system-configs.d.ts +0 -32
- package/tools/system-configs/list-system-configs.d.ts.map +0 -1
- package/tools/system-configs/list-system-configs.js.map +0 -1
- package/tools/system-configs/list-system-configs.mjs.map +0 -1
|
@@ -3,22 +3,22 @@
|
|
|
3
3
|
import { asTextContentResult } from 'scorecard-ai-mcp/tools/types';
|
|
4
4
|
|
|
5
5
|
import { Tool } from '@modelcontextprotocol/sdk/types.js';
|
|
6
|
-
import type { Metadata } from '
|
|
6
|
+
import type { Metadata } from '../../';
|
|
7
7
|
import Scorecard from 'scorecard-ai';
|
|
8
8
|
|
|
9
9
|
export const metadata: Metadata = {
|
|
10
|
-
resource: '
|
|
10
|
+
resource: 'systems.versions',
|
|
11
11
|
operation: 'read',
|
|
12
12
|
tags: [],
|
|
13
13
|
httpMethod: 'get',
|
|
14
14
|
httpPath: '/systems/{systemId}/configs',
|
|
15
|
-
operationId: '
|
|
15
|
+
operationId: 'listSystemVersions',
|
|
16
16
|
};
|
|
17
17
|
|
|
18
18
|
export const tool: Tool = {
|
|
19
|
-
name: '
|
|
19
|
+
name: 'list_systems_versions',
|
|
20
20
|
description:
|
|
21
|
-
'Retrieve a paginated list of
|
|
21
|
+
'Retrieve a paginated list of system versions for a specific system.\n\nSystem versions provide concrete parameter values for a System Under Test, defining exactly how the system should be configured during an evaluation run.',
|
|
22
22
|
inputSchema: {
|
|
23
23
|
type: 'object',
|
|
24
24
|
properties: {
|
|
@@ -41,7 +41,7 @@ export const tool: Tool = {
|
|
|
41
41
|
|
|
42
42
|
export const handler = async (client: Scorecard, args: Record<string, unknown> | undefined) => {
|
|
43
43
|
const { systemId, ...body } = args as any;
|
|
44
|
-
return asTextContentResult(await client.
|
|
44
|
+
return asTextContentResult(await client.systems.versions.list(systemId, body));
|
|
45
45
|
};
|
|
46
46
|
|
|
47
47
|
export default { metadata, tool, handler };
|
package/src/tools/types.ts
CHANGED
package/tools/index.d.mts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.mts","sourceRoot":"","sources":["../src/tools/index.ts"],"names":[],"mappings":"OAEO,EAAE,QAAQ,EAAE,QAAQ,EAAE,eAAe,EAAE;AAE9C,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,eAAe,EAAE,CAAC;
|
|
1
|
+
{"version":3,"file":"index.d.mts","sourceRoot":"","sources":["../src/tools/index.ts"],"names":[],"mappings":"OAEO,EAAE,QAAQ,EAAE,QAAQ,EAAE,eAAe,EAAE;AAE9C,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,eAAe,EAAE,CAAC;AA4B/C,eAAO,MAAM,SAAS,EAAE,QAAQ,EAAO,CAAC;AAgCxC,MAAM,MAAM,MAAM,GAAG;IACnB,IAAI,EAAE,UAAU,GAAG,WAAW,GAAG,KAAK,GAAG,MAAM,CAAC;IAChD,EAAE,EAAE,SAAS,GAAG,SAAS,CAAC;IAC1B,KAAK,EAAE,MAAM,CAAC;CACf,CAAC;AAEF,wBAAgB,KAAK,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,QAAQ,EAAE,GAAG,QAAQ,EAAE,CA4B1E"}
|
package/tools/index.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/tools/index.ts"],"names":[],"mappings":"OAEO,EAAE,QAAQ,EAAE,QAAQ,EAAE,eAAe,EAAE;AAE9C,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,eAAe,EAAE,CAAC;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/tools/index.ts"],"names":[],"mappings":"OAEO,EAAE,QAAQ,EAAE,QAAQ,EAAE,eAAe,EAAE;AAE9C,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,eAAe,EAAE,CAAC;AA4B/C,eAAO,MAAM,SAAS,EAAE,QAAQ,EAAO,CAAC;AAgCxC,MAAM,MAAM,MAAM,GAAG;IACnB,IAAI,EAAE,UAAU,GAAG,WAAW,GAAG,KAAK,GAAG,MAAM,CAAC;IAChD,EAAE,EAAE,SAAS,GAAG,SAAS,CAAC;IAC1B,KAAK,EAAE,MAAM,CAAC;CACf,CAAC;AAEF,wBAAgB,KAAK,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,QAAQ,EAAE,GAAG,QAAQ,EAAE,CA4B1E"}
|
package/tools/index.js
CHANGED
|
@@ -19,6 +19,8 @@ const list_testcases_1 = __importDefault(require("./testcases/list-testcases.js"
|
|
|
19
19
|
const delete_testcases_1 = __importDefault(require("./testcases/delete-testcases.js"));
|
|
20
20
|
const get_testcases_1 = __importDefault(require("./testcases/get-testcases.js"));
|
|
21
21
|
const create_runs_1 = __importDefault(require("./runs/create-runs.js"));
|
|
22
|
+
const create_metrics_1 = __importDefault(require("./metrics/create-metrics.js"));
|
|
23
|
+
const update_metrics_1 = __importDefault(require("./metrics/update-metrics.js"));
|
|
22
24
|
const create_records_1 = __importDefault(require("./records/create-records.js"));
|
|
23
25
|
const upsert_scores_1 = __importDefault(require("./scores/upsert-scores.js"));
|
|
24
26
|
const create_systems_1 = __importDefault(require("./systems/create-systems.js"));
|
|
@@ -26,9 +28,9 @@ const update_systems_1 = __importDefault(require("./systems/update-systems.js"))
|
|
|
26
28
|
const list_systems_1 = __importDefault(require("./systems/list-systems.js"));
|
|
27
29
|
const delete_systems_1 = __importDefault(require("./systems/delete-systems.js"));
|
|
28
30
|
const get_systems_1 = __importDefault(require("./systems/get-systems.js"));
|
|
29
|
-
const
|
|
30
|
-
const
|
|
31
|
-
const
|
|
31
|
+
const create_systems_versions_1 = __importDefault(require("./systems/versions/create-systems-versions.js"));
|
|
32
|
+
const list_systems_versions_1 = __importDefault(require("./systems/versions/list-systems-versions.js"));
|
|
33
|
+
const get_systems_versions_1 = __importDefault(require("./systems/versions/get-systems-versions.js"));
|
|
32
34
|
exports.endpoints = [];
|
|
33
35
|
function addEndpoint(endpoint) {
|
|
34
36
|
exports.endpoints.push(endpoint);
|
|
@@ -46,6 +48,8 @@ addEndpoint(list_testcases_1.default);
|
|
|
46
48
|
addEndpoint(delete_testcases_1.default);
|
|
47
49
|
addEndpoint(get_testcases_1.default);
|
|
48
50
|
addEndpoint(create_runs_1.default);
|
|
51
|
+
addEndpoint(create_metrics_1.default);
|
|
52
|
+
addEndpoint(update_metrics_1.default);
|
|
49
53
|
addEndpoint(create_records_1.default);
|
|
50
54
|
addEndpoint(upsert_scores_1.default);
|
|
51
55
|
addEndpoint(create_systems_1.default);
|
|
@@ -53,9 +57,9 @@ addEndpoint(update_systems_1.default);
|
|
|
53
57
|
addEndpoint(list_systems_1.default);
|
|
54
58
|
addEndpoint(delete_systems_1.default);
|
|
55
59
|
addEndpoint(get_systems_1.default);
|
|
56
|
-
addEndpoint(
|
|
57
|
-
addEndpoint(
|
|
58
|
-
addEndpoint(
|
|
60
|
+
addEndpoint(create_systems_versions_1.default);
|
|
61
|
+
addEndpoint(list_systems_versions_1.default);
|
|
62
|
+
addEndpoint(get_systems_versions_1.default);
|
|
59
63
|
function query(filters, endpoints) {
|
|
60
64
|
const allExcludes = filters.length > 0 && filters.every((filter) => filter.op === 'exclude');
|
|
61
65
|
const unmatchedFilters = new Set(filters);
|
|
@@ -70,8 +74,9 @@ function query(filters, endpoints) {
|
|
|
70
74
|
return included;
|
|
71
75
|
});
|
|
72
76
|
// Check if any filters didn't match
|
|
73
|
-
|
|
74
|
-
|
|
77
|
+
const unmatched = Array.from(unmatchedFilters).filter((f) => f.type === 'tool' || f.type === 'resource');
|
|
78
|
+
if (unmatched.length > 0) {
|
|
79
|
+
throw new Error(`The following filters did not match any endpoints: ${unmatched
|
|
75
80
|
.map((f) => `${f.type}=${f.value}`)
|
|
76
81
|
.join(', ')}`);
|
|
77
82
|
}
|
package/tools/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/tools/index.ts"],"names":[],"mappings":";AAAA,sFAAsF;;;;;;
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/tools/index.ts"],"names":[],"mappings":";AAAA,sFAAsF;;;;;;AAsEtF,sBA4BC;AA5FD,oFAAyD;AACzD,gFAAqD;AACrD,oFAAyD;AACzD,oFAAyD;AACzD,gFAAqD;AACrD,oFAAyD;AACzD,8EAAmD;AACnD,uFAA4D;AAC5D,uFAA4D;AAC5D,mFAAwD;AACxD,uFAA4D;AAC5D,iFAAsD;AACtD,wEAA6C;AAC7C,iFAAsD;AACtD,iFAAsD;AACtD,iFAAsD;AACtD,8EAAmD;AACnD,iFAAsD;AACtD,iFAAsD;AACtD,6EAAkD;AAClD,iFAAsD;AACtD,2EAAgD;AAChD,4GAAiF;AACjF,wGAA6E;AAC7E,sGAA2E;AAE9D,QAAA,SAAS,GAAe,EAAE,CAAC;AAExC,SAAS,WAAW,CAAC,QAAkB;IACrC,iBAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;AAC3B,CAAC;AAED,WAAW,CAAC,yBAAe,CAAC,CAAC;AAC7B,WAAW,CAAC,uBAAa,CAAC,CAAC;AAC3B,WAAW,CAAC,yBAAe,CAAC,CAAC;AAC7B,WAAW,CAAC,yBAAe,CAAC,CAAC;AAC7B,WAAW,CAAC,uBAAa,CAAC,CAAC;AAC3B,WAAW,CAAC,yBAAe,CAAC,CAAC;AAC7B,WAAW,CAAC,sBAAY,CAAC,CAAC;AAC1B,WAAW,CAAC,0BAAgB,CAAC,CAAC;AAC9B,WAAW,CAAC,0BAAgB,CAAC,CAAC;AAC9B,WAAW,CAAC,wBAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,0BAAgB,CAAC,CAAC;AAC9B,WAAW,CAAC,uBAAa,CAAC,CAAC;AAC3B,WAAW,CAAC,qBAAW,CAAC,CAAC;AACzB,WAAW,CAAC,wBAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,wBAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,wBAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,uBAAa,CAAC,CAAC;AAC3B,WAAW,CAAC,wBAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,wBAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,sBAAY,CAAC,CAAC;AAC1B,WAAW,CAAC,wBAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,qBAAW,CAAC,CAAC;AACzB,WAAW,CAAC,iCAAuB,CAAC,CAAC;AACrC,WAAW,CAAC,+BAAqB,CAAC,CAAC;AACnC,WAAW,CAAC,8BAAoB,CAAC,CAAC;AAQlC,SAAgB,KAAK,CAAC,OAAiB,EAAE,SAAqB;IAC5D,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,CAAC,KAAK,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,EAAE,KAAK,SAAS,CAAC,CAAC;IAC7F,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC;IAE1C,MAAM,QAAQ,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,QAAkB,EAAE,EAAE;QACvD,IAAI,QAAQ,GAAG,KAAK,IAAI,WAAW,CAAC;QAEpC,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;YAC7B,IAAI,KAAK,CAAC,MAAM,EAAE,QAAQ,CAAC,EAAE,CAAC;gBAC5B,gBAAgB,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;gBAChC,QAAQ,GAAG,MAAM,CAAC,EAAE,KAAK,SAAS,CAAC;YACrC,CAAC;QACH,CAAC;QAED,OAAO,QAAQ,CAAC;IAClB,CAAC,CAAC,CAAC;IAEH,oCAAoC;IACpC,MAAM,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,IAAI,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;IACzG,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACzB,MAAM,IAAI,KAAK,CACb,sDAAsD,SAAS;aAC5D,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,KAAK,EAAE,CAAC;aAClC,IAAI,CAAC,IAAI,CAAC,EAAE,CAChB,CAAC;IACJ,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,KAAK,CAAC,EAAE,IAAI,EAAE,KAAK,EAAU,EAAE,QAAkB;IACxD,QAAQ,IAAI,EAAE,CAAC;QACb,KAAK,UAAU,CAAC,CAAC,CAAC;YAChB,MAAM,QAAQ,GAAG,GAAG,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,GAAG,CAAC;YAC3E,MAAM,KAAK,GAAG,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC;YACnC,OAAO,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,QAAQ,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC;QACnE,CAAC;QACD,KAAK,WAAW;YACd,OAAO,QAAQ,CAAC,QAAQ,CAAC,SAAS,KAAK,KAAK,CAAC;QAC/C,KAAK,KAAK;YACR,OAAO,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAChD,KAAK,MAAM;YACT,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,KAAK,KAAK,CAAC;IACxC,CAAC;AACH,CAAC;AAED,SAAS,iBAAiB,CAAC,QAAgB;IACzC,OAAO,QAAQ,CAAC,WAAW,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,EAAE,CAAC,CAAC;AAC7D,CAAC"}
|
package/tools/index.mjs
CHANGED
|
@@ -12,6 +12,8 @@ import list_testcases from "./testcases/list-testcases.mjs";
|
|
|
12
12
|
import delete_testcases from "./testcases/delete-testcases.mjs";
|
|
13
13
|
import get_testcases from "./testcases/get-testcases.mjs";
|
|
14
14
|
import create_runs from "./runs/create-runs.mjs";
|
|
15
|
+
import create_metrics from "./metrics/create-metrics.mjs";
|
|
16
|
+
import update_metrics from "./metrics/update-metrics.mjs";
|
|
15
17
|
import create_records from "./records/create-records.mjs";
|
|
16
18
|
import upsert_scores from "./scores/upsert-scores.mjs";
|
|
17
19
|
import create_systems from "./systems/create-systems.mjs";
|
|
@@ -19,9 +21,9 @@ import update_systems from "./systems/update-systems.mjs";
|
|
|
19
21
|
import list_systems from "./systems/list-systems.mjs";
|
|
20
22
|
import delete_systems from "./systems/delete-systems.mjs";
|
|
21
23
|
import get_systems from "./systems/get-systems.mjs";
|
|
22
|
-
import
|
|
23
|
-
import
|
|
24
|
-
import
|
|
24
|
+
import create_systems_versions from "./systems/versions/create-systems-versions.mjs";
|
|
25
|
+
import list_systems_versions from "./systems/versions/list-systems-versions.mjs";
|
|
26
|
+
import get_systems_versions from "./systems/versions/get-systems-versions.mjs";
|
|
25
27
|
export const endpoints = [];
|
|
26
28
|
function addEndpoint(endpoint) {
|
|
27
29
|
endpoints.push(endpoint);
|
|
@@ -39,6 +41,8 @@ addEndpoint(list_testcases);
|
|
|
39
41
|
addEndpoint(delete_testcases);
|
|
40
42
|
addEndpoint(get_testcases);
|
|
41
43
|
addEndpoint(create_runs);
|
|
44
|
+
addEndpoint(create_metrics);
|
|
45
|
+
addEndpoint(update_metrics);
|
|
42
46
|
addEndpoint(create_records);
|
|
43
47
|
addEndpoint(upsert_scores);
|
|
44
48
|
addEndpoint(create_systems);
|
|
@@ -46,9 +50,9 @@ addEndpoint(update_systems);
|
|
|
46
50
|
addEndpoint(list_systems);
|
|
47
51
|
addEndpoint(delete_systems);
|
|
48
52
|
addEndpoint(get_systems);
|
|
49
|
-
addEndpoint(
|
|
50
|
-
addEndpoint(
|
|
51
|
-
addEndpoint(
|
|
53
|
+
addEndpoint(create_systems_versions);
|
|
54
|
+
addEndpoint(list_systems_versions);
|
|
55
|
+
addEndpoint(get_systems_versions);
|
|
52
56
|
export function query(filters, endpoints) {
|
|
53
57
|
const allExcludes = filters.length > 0 && filters.every((filter) => filter.op === 'exclude');
|
|
54
58
|
const unmatchedFilters = new Set(filters);
|
|
@@ -63,8 +67,9 @@ export function query(filters, endpoints) {
|
|
|
63
67
|
return included;
|
|
64
68
|
});
|
|
65
69
|
// Check if any filters didn't match
|
|
66
|
-
|
|
67
|
-
|
|
70
|
+
const unmatched = Array.from(unmatchedFilters).filter((f) => f.type === 'tool' || f.type === 'resource');
|
|
71
|
+
if (unmatched.length > 0) {
|
|
72
|
+
throw new Error(`The following filters did not match any endpoints: ${unmatched
|
|
68
73
|
.map((f) => `${f.type}=${f.value}`)
|
|
69
74
|
.join(', ')}`);
|
|
70
75
|
}
|
package/tools/index.mjs.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.mjs","sourceRoot":"","sources":["../src/tools/index.ts"],"names":[],"mappings":"AAAA,sFAAsF;OAM/E,eAAe;OACf,aAAa;OACb,eAAe;OACf,eAAe;OACf,aAAa;OACb,eAAe;OACf,YAAY;OACZ,gBAAgB;OAChB,gBAAgB;OAChB,cAAc;OACd,gBAAgB;OAChB,aAAa;OACb,WAAW;OACX,cAAc;OACd,aAAa;OACb,cAAc;OACd,cAAc;OACd,YAAY;OACZ,cAAc;OACd,WAAW;OACX,qBAAqB;OACrB,
|
|
1
|
+
{"version":3,"file":"index.mjs","sourceRoot":"","sources":["../src/tools/index.ts"],"names":[],"mappings":"AAAA,sFAAsF;OAM/E,eAAe;OACf,aAAa;OACb,eAAe;OACf,eAAe;OACf,aAAa;OACb,eAAe;OACf,YAAY;OACZ,gBAAgB;OAChB,gBAAgB;OAChB,cAAc;OACd,gBAAgB;OAChB,aAAa;OACb,WAAW;OACX,cAAc;OACd,cAAc;OACd,cAAc;OACd,aAAa;OACb,cAAc;OACd,cAAc;OACd,YAAY;OACZ,cAAc;OACd,WAAW;OACX,uBAAuB;OACvB,qBAAqB;OACrB,oBAAoB;AAE3B,MAAM,CAAC,MAAM,SAAS,GAAe,EAAE,CAAC;AAExC,SAAS,WAAW,CAAC,QAAkB;IACrC,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;AAC3B,CAAC;AAED,WAAW,CAAC,eAAe,CAAC,CAAC;AAC7B,WAAW,CAAC,aAAa,CAAC,CAAC;AAC3B,WAAW,CAAC,eAAe,CAAC,CAAC;AAC7B,WAAW,CAAC,eAAe,CAAC,CAAC;AAC7B,WAAW,CAAC,aAAa,CAAC,CAAC;AAC3B,WAAW,CAAC,eAAe,CAAC,CAAC;AAC7B,WAAW,CAAC,YAAY,CAAC,CAAC;AAC1B,WAAW,CAAC,gBAAgB,CAAC,CAAC;AAC9B,WAAW,CAAC,gBAAgB,CAAC,CAAC;AAC9B,WAAW,CAAC,cAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,gBAAgB,CAAC,CAAC;AAC9B,WAAW,CAAC,aAAa,CAAC,CAAC;AAC3B,WAAW,CAAC,WAAW,CAAC,CAAC;AACzB,WAAW,CAAC,cAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,cAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,cAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,aAAa,CAAC,CAAC;AAC3B,WAAW,CAAC,cAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,cAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,YAAY,CAAC,CAAC;AAC1B,WAAW,CAAC,cAAc,CAAC,CAAC;AAC5B,WAAW,CAAC,WAAW,CAAC,CAAC;AACzB,WAAW,CAAC,uBAAuB,CAAC,CAAC;AACrC,WAAW,CAAC,qBAAqB,CAAC,CAAC;AACnC,WAAW,CAAC,oBAAoB,CAAC,CAAC;AAQlC,MAAM,UAAU,KAAK,CAAC,OAAiB,EAAE,SAAqB;IAC5D,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,CAAC,KAAK,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,EAAE,KAAK,SAAS,CAAC,CAAC;IAC7F,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC;IAE1C,MAAM,QAAQ,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,QAAkB,EAAE,EAAE;QACvD,IAAI,QAAQ,GAAG,KAAK,IAAI,WAAW,CAAC;QAEpC,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;YAC7B,IAAI,KAAK,CAAC,MAAM,EAAE,QAAQ,CAAC,EAAE,CAAC;gBAC5B,gBAAgB,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;gBAChC,QAAQ,GAAG,MAAM,CAAC,EAAE,KAAK,SAAS,CAAC;YACrC,CAAC;QACH,CAAC;QAED,OAAO,QAAQ,CAAC;IAClB,CAAC,CAAC,CAAC;IAEH,oCAAoC;IACpC,MAAM,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,IAAI,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;IACzG,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACzB,MAAM,IAAI,KAAK,CACb,sDAAsD,SAAS;aAC5D,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,KAAK,EAAE,CAAC;aAClC,IAAI,CAAC,IAAI,CAAC,EAAE,CAChB,CAAC;IACJ,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,KAAK,CAAC,EAAE,IAAI,EAAE,KAAK,EAAU,EAAE,QAAkB;IACxD,QAAQ,IAAI,EAAE,CAAC;QACb,KAAK,UAAU,CAAC,CAAC,CAAC;YAChB,MAAM,QAAQ,GAAG,GAAG,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,GAAG,CAAC;YAC3E,MAAM,KAAK,GAAG,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC;YACnC,OAAO,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC,QAAQ,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC;QACnE,CAAC;QACD,KAAK,WAAW;YACd,OAAO,QAAQ,CAAC,QAAQ,CAAC,SAAS,KAAK,KAAK,CAAC;QAC/C,KAAK,KAAK;YACR,OAAO,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAChD,KAAK,MAAM;YACT,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,KAAK,KAAK,CAAC;IACxC,CAAC;AACH,CAAC;AAED,SAAS,iBAAiB,CAAC,QAAgB;IACzC,OAAO,QAAQ,CAAC,WAAW,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,EAAE,CAAC,CAAC;AAC7D,CAAC"}
|
|
@@ -15,8 +15,17 @@ declare const _default: {
|
|
|
15
15
|
properties?: {
|
|
16
16
|
[x: string]: unknown;
|
|
17
17
|
} | undefined;
|
|
18
|
+
required?: string[] | undefined;
|
|
18
19
|
};
|
|
19
20
|
description?: string | undefined;
|
|
21
|
+
outputSchema?: {
|
|
22
|
+
[x: string]: unknown;
|
|
23
|
+
type: "object";
|
|
24
|
+
properties?: {
|
|
25
|
+
[x: string]: unknown;
|
|
26
|
+
} | undefined;
|
|
27
|
+
required?: string[] | undefined;
|
|
28
|
+
} | undefined;
|
|
20
29
|
annotations?: {
|
|
21
30
|
[x: string]: unknown;
|
|
22
31
|
title?: string | undefined;
|
|
@@ -29,4 +38,4 @@ declare const _default: {
|
|
|
29
38
|
handler: (client: Scorecard, args: Record<string, unknown> | undefined) => Promise<import("scorecard-ai-mcp/tools/types").ToolCallResult>;
|
|
30
39
|
};
|
|
31
40
|
export default _default;
|
|
32
|
-
//# sourceMappingURL=
|
|
41
|
+
//# sourceMappingURL=create-metrics.d.mts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"create-metrics.d.mts","sourceRoot":"","sources":["../../src/tools/metrics/create-metrics.ts"],"names":[],"mappings":"OAIO,EAAE,IAAI,EAAE,MAAM,oCAAoC;OAClD,KAAK,EAAE,QAAQ,EAAE;OACjB,SAAS,MAAM,cAAc;AAEpC,eAAO,MAAM,QAAQ,EAAE,QAOtB,CAAC;AAEF,eAAO,MAAM,IAAI,EAAE,IAmOlB,CAAC;AAEF,eAAO,MAAM,OAAO,GAAU,QAAQ,SAAS,EAAE,MAAM,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,mEAGzF,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;sBAHoC,SAAS,QAAQ,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS;;AAK1F,wBAA2C"}
|
|
@@ -15,8 +15,17 @@ declare const _default: {
|
|
|
15
15
|
properties?: {
|
|
16
16
|
[x: string]: unknown;
|
|
17
17
|
} | undefined;
|
|
18
|
+
required?: string[] | undefined;
|
|
18
19
|
};
|
|
19
20
|
description?: string | undefined;
|
|
21
|
+
outputSchema?: {
|
|
22
|
+
[x: string]: unknown;
|
|
23
|
+
type: "object";
|
|
24
|
+
properties?: {
|
|
25
|
+
[x: string]: unknown;
|
|
26
|
+
} | undefined;
|
|
27
|
+
required?: string[] | undefined;
|
|
28
|
+
} | undefined;
|
|
20
29
|
annotations?: {
|
|
21
30
|
[x: string]: unknown;
|
|
22
31
|
title?: string | undefined;
|
|
@@ -29,4 +38,4 @@ declare const _default: {
|
|
|
29
38
|
handler: (client: Scorecard, args: Record<string, unknown> | undefined) => Promise<import("scorecard-ai-mcp/tools/types").ToolCallResult>;
|
|
30
39
|
};
|
|
31
40
|
export default _default;
|
|
32
|
-
//# sourceMappingURL=create-
|
|
41
|
+
//# sourceMappingURL=create-metrics.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"create-metrics.d.ts","sourceRoot":"","sources":["../../src/tools/metrics/create-metrics.ts"],"names":[],"mappings":"OAIO,EAAE,IAAI,EAAE,MAAM,oCAAoC;OAClD,KAAK,EAAE,QAAQ,EAAE;OACjB,SAAS,MAAM,cAAc;AAEpC,eAAO,MAAM,QAAQ,EAAE,QAOtB,CAAC;AAEF,eAAO,MAAM,IAAI,EAAE,IAmOlB,CAAC;AAEF,eAAO,MAAM,OAAO,GAAU,QAAQ,SAAS,EAAE,MAAM,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,mEAGzF,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;sBAHoC,SAAS,QAAQ,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS;;AAK1F,wBAA2C"}
|
|
@@ -0,0 +1,245 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
// File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
|
|
3
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
4
|
+
exports.handler = exports.tool = exports.metadata = void 0;
|
|
5
|
+
const types_1 = require("scorecard-ai-mcp/tools/types");
|
|
6
|
+
exports.metadata = {
|
|
7
|
+
resource: 'metrics',
|
|
8
|
+
operation: 'write',
|
|
9
|
+
tags: [],
|
|
10
|
+
httpMethod: 'post',
|
|
11
|
+
httpPath: '/projects/{projectId}/metrics',
|
|
12
|
+
operationId: 'createMetric',
|
|
13
|
+
};
|
|
14
|
+
exports.tool = {
|
|
15
|
+
name: 'create_metrics',
|
|
16
|
+
description: 'Create a new Metric for evaluating system outputs. The structure of a metric depends on the evalType and outputType of the metric.',
|
|
17
|
+
inputSchema: {
|
|
18
|
+
type: 'object',
|
|
19
|
+
anyOf: [
|
|
20
|
+
{
|
|
21
|
+
type: 'object',
|
|
22
|
+
properties: {
|
|
23
|
+
projectId: {
|
|
24
|
+
type: 'string',
|
|
25
|
+
},
|
|
26
|
+
evalType: {
|
|
27
|
+
type: 'string',
|
|
28
|
+
description: 'AI-based evaluation type.',
|
|
29
|
+
enum: ['ai'],
|
|
30
|
+
},
|
|
31
|
+
name: {
|
|
32
|
+
type: 'string',
|
|
33
|
+
description: 'The name of the Metric.',
|
|
34
|
+
},
|
|
35
|
+
outputType: {
|
|
36
|
+
type: 'string',
|
|
37
|
+
description: 'Integer output type.',
|
|
38
|
+
enum: ['int'],
|
|
39
|
+
},
|
|
40
|
+
promptTemplate: {
|
|
41
|
+
type: 'string',
|
|
42
|
+
description: 'The complete prompt template for AI evaluation. Should include placeholders for dynamic content.',
|
|
43
|
+
},
|
|
44
|
+
description: {
|
|
45
|
+
type: 'string',
|
|
46
|
+
description: 'The description of the Metric.',
|
|
47
|
+
},
|
|
48
|
+
evalModelName: {
|
|
49
|
+
type: 'string',
|
|
50
|
+
description: 'The AI model to use for evaluation.',
|
|
51
|
+
},
|
|
52
|
+
guidelines: {
|
|
53
|
+
type: 'string',
|
|
54
|
+
description: 'Guidelines for AI evaluation on how to score the metric.',
|
|
55
|
+
},
|
|
56
|
+
passingThreshold: {
|
|
57
|
+
type: 'integer',
|
|
58
|
+
description: 'The threshold for determining pass/fail from integer scores (1-5).',
|
|
59
|
+
},
|
|
60
|
+
temperature: {
|
|
61
|
+
type: 'number',
|
|
62
|
+
description: 'The temperature for AI evaluation (0-2).',
|
|
63
|
+
},
|
|
64
|
+
},
|
|
65
|
+
},
|
|
66
|
+
{
|
|
67
|
+
type: 'object',
|
|
68
|
+
properties: {
|
|
69
|
+
projectId: {
|
|
70
|
+
type: 'string',
|
|
71
|
+
},
|
|
72
|
+
evalType: {
|
|
73
|
+
type: 'string',
|
|
74
|
+
description: 'Human-based evaluation type.',
|
|
75
|
+
enum: ['human'],
|
|
76
|
+
},
|
|
77
|
+
name: {
|
|
78
|
+
type: 'string',
|
|
79
|
+
description: 'The name of the Metric.',
|
|
80
|
+
},
|
|
81
|
+
outputType: {
|
|
82
|
+
type: 'string',
|
|
83
|
+
description: 'Integer output type.',
|
|
84
|
+
enum: ['int'],
|
|
85
|
+
},
|
|
86
|
+
description: {
|
|
87
|
+
type: 'string',
|
|
88
|
+
description: 'The description of the Metric.',
|
|
89
|
+
},
|
|
90
|
+
guidelines: {
|
|
91
|
+
type: 'string',
|
|
92
|
+
description: 'Guidelines for human evaluators.',
|
|
93
|
+
},
|
|
94
|
+
passingThreshold: {
|
|
95
|
+
type: 'integer',
|
|
96
|
+
description: 'The threshold for determining pass/fail from integer scores (1-5).',
|
|
97
|
+
},
|
|
98
|
+
},
|
|
99
|
+
},
|
|
100
|
+
{
|
|
101
|
+
type: 'object',
|
|
102
|
+
properties: {
|
|
103
|
+
projectId: {
|
|
104
|
+
type: 'string',
|
|
105
|
+
},
|
|
106
|
+
evalType: {
|
|
107
|
+
type: 'string',
|
|
108
|
+
description: 'Heuristic-based evaluation type.',
|
|
109
|
+
enum: ['heuristic'],
|
|
110
|
+
},
|
|
111
|
+
name: {
|
|
112
|
+
type: 'string',
|
|
113
|
+
description: 'The name of the Metric.',
|
|
114
|
+
},
|
|
115
|
+
outputType: {
|
|
116
|
+
type: 'string',
|
|
117
|
+
description: 'Integer output type.',
|
|
118
|
+
enum: ['int'],
|
|
119
|
+
},
|
|
120
|
+
description: {
|
|
121
|
+
type: 'string',
|
|
122
|
+
description: 'The description of the Metric.',
|
|
123
|
+
},
|
|
124
|
+
guidelines: {
|
|
125
|
+
type: 'string',
|
|
126
|
+
description: 'Optional guidelines for heuristic evaluation logic.',
|
|
127
|
+
},
|
|
128
|
+
passingThreshold: {
|
|
129
|
+
type: 'integer',
|
|
130
|
+
description: 'The threshold for determining pass/fail from integer scores (1-5).',
|
|
131
|
+
},
|
|
132
|
+
},
|
|
133
|
+
},
|
|
134
|
+
{
|
|
135
|
+
type: 'object',
|
|
136
|
+
properties: {
|
|
137
|
+
projectId: {
|
|
138
|
+
type: 'string',
|
|
139
|
+
},
|
|
140
|
+
evalType: {
|
|
141
|
+
type: 'string',
|
|
142
|
+
description: 'AI-based evaluation type.',
|
|
143
|
+
enum: ['ai'],
|
|
144
|
+
},
|
|
145
|
+
name: {
|
|
146
|
+
type: 'string',
|
|
147
|
+
description: 'The name of the Metric.',
|
|
148
|
+
},
|
|
149
|
+
outputType: {
|
|
150
|
+
type: 'string',
|
|
151
|
+
description: 'Boolean output type.',
|
|
152
|
+
enum: ['boolean'],
|
|
153
|
+
},
|
|
154
|
+
promptTemplate: {
|
|
155
|
+
type: 'string',
|
|
156
|
+
description: 'The complete prompt template for AI evaluation. Should include placeholders for dynamic content.',
|
|
157
|
+
},
|
|
158
|
+
description: {
|
|
159
|
+
type: 'string',
|
|
160
|
+
description: 'The description of the Metric.',
|
|
161
|
+
},
|
|
162
|
+
evalModelName: {
|
|
163
|
+
type: 'string',
|
|
164
|
+
description: 'The AI model to use for evaluation.',
|
|
165
|
+
},
|
|
166
|
+
guidelines: {
|
|
167
|
+
type: 'string',
|
|
168
|
+
description: 'Guidelines for AI evaluation on how to score the metric.',
|
|
169
|
+
},
|
|
170
|
+
temperature: {
|
|
171
|
+
type: 'number',
|
|
172
|
+
description: 'The temperature for AI evaluation (0-2).',
|
|
173
|
+
},
|
|
174
|
+
},
|
|
175
|
+
},
|
|
176
|
+
{
|
|
177
|
+
type: 'object',
|
|
178
|
+
properties: {
|
|
179
|
+
projectId: {
|
|
180
|
+
type: 'string',
|
|
181
|
+
},
|
|
182
|
+
evalType: {
|
|
183
|
+
type: 'string',
|
|
184
|
+
description: 'Human-based evaluation type.',
|
|
185
|
+
enum: ['human'],
|
|
186
|
+
},
|
|
187
|
+
name: {
|
|
188
|
+
type: 'string',
|
|
189
|
+
description: 'The name of the Metric.',
|
|
190
|
+
},
|
|
191
|
+
outputType: {
|
|
192
|
+
type: 'string',
|
|
193
|
+
description: 'Boolean output type.',
|
|
194
|
+
enum: ['boolean'],
|
|
195
|
+
},
|
|
196
|
+
description: {
|
|
197
|
+
type: 'string',
|
|
198
|
+
description: 'The description of the Metric.',
|
|
199
|
+
},
|
|
200
|
+
guidelines: {
|
|
201
|
+
type: 'string',
|
|
202
|
+
description: 'Guidelines for human evaluators.',
|
|
203
|
+
},
|
|
204
|
+
},
|
|
205
|
+
},
|
|
206
|
+
{
|
|
207
|
+
type: 'object',
|
|
208
|
+
properties: {
|
|
209
|
+
projectId: {
|
|
210
|
+
type: 'string',
|
|
211
|
+
},
|
|
212
|
+
evalType: {
|
|
213
|
+
type: 'string',
|
|
214
|
+
description: 'Heuristic-based evaluation type.',
|
|
215
|
+
enum: ['heuristic'],
|
|
216
|
+
},
|
|
217
|
+
name: {
|
|
218
|
+
type: 'string',
|
|
219
|
+
description: 'The name of the Metric.',
|
|
220
|
+
},
|
|
221
|
+
outputType: {
|
|
222
|
+
type: 'string',
|
|
223
|
+
description: 'Boolean output type.',
|
|
224
|
+
enum: ['boolean'],
|
|
225
|
+
},
|
|
226
|
+
description: {
|
|
227
|
+
type: 'string',
|
|
228
|
+
description: 'The description of the Metric.',
|
|
229
|
+
},
|
|
230
|
+
guidelines: {
|
|
231
|
+
type: 'string',
|
|
232
|
+
description: 'Optional guidelines for heuristic evaluation logic.',
|
|
233
|
+
},
|
|
234
|
+
},
|
|
235
|
+
},
|
|
236
|
+
],
|
|
237
|
+
},
|
|
238
|
+
};
|
|
239
|
+
const handler = async (client, args) => {
|
|
240
|
+
const { projectId, ...body } = args;
|
|
241
|
+
return (0, types_1.asTextContentResult)(await client.metrics.create(projectId, body));
|
|
242
|
+
};
|
|
243
|
+
exports.handler = handler;
|
|
244
|
+
exports.default = { metadata: exports.metadata, tool: exports.tool, handler: exports.handler };
|
|
245
|
+
//# sourceMappingURL=create-metrics.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"create-metrics.js","sourceRoot":"","sources":["../../src/tools/metrics/create-metrics.ts"],"names":[],"mappings":";AAAA,sFAAsF;;;AAEtF,wDAAmE;AAMtD,QAAA,QAAQ,GAAa;IAChC,QAAQ,EAAE,SAAS;IACnB,SAAS,EAAE,OAAO;IAClB,IAAI,EAAE,EAAE;IACR,UAAU,EAAE,MAAM;IAClB,QAAQ,EAAE,+BAA+B;IACzC,WAAW,EAAE,cAAc;CAC5B,CAAC;AAEW,QAAA,IAAI,GAAS;IACxB,IAAI,EAAE,gBAAgB;IACtB,WAAW,EACT,oIAAoI;IACtI,WAAW,EAAE;QACX,IAAI,EAAE,QAAQ;QACd,KAAK,EAAE;YACL;gBACE,IAAI,EAAE,QAAQ;gBACd,UAAU,EAAE;oBACV,SAAS,EAAE;wBACT,IAAI,EAAE,QAAQ;qBACf;oBACD,QAAQ,EAAE;wBACR,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,2BAA2B;wBACxC,IAAI,EAAE,CAAC,IAAI,CAAC;qBACb;oBACD,IAAI,EAAE;wBACJ,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,yBAAyB;qBACvC;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,sBAAsB;wBACnC,IAAI,EAAE,CAAC,KAAK,CAAC;qBACd;oBACD,cAAc,EAAE;wBACd,IAAI,EAAE,QAAQ;wBACd,WAAW,EACT,kGAAkG;qBACrG;oBACD,WAAW,EAAE;wBACX,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,gCAAgC;qBAC9C;oBACD,aAAa,EAAE;wBACb,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,qCAAqC;qBACnD;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,0DAA0D;qBACxE;oBACD,gBAAgB,EAAE;wBAChB,IAAI,EAAE,SAAS;wBACf,WAAW,EAAE,oEAAoE;qBAClF;oBACD,WAAW,EAAE;wBACX,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,0CAA0C;qBACxD;iBACF;aACF;YACD;gBACE,IAAI,EAAE,QAAQ;gBACd,UAAU,EAAE;oBACV,SAAS,EAAE;wBACT,IAAI,EAAE,QAAQ;qBACf;oBACD,QAAQ,EAAE;wBACR,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,8BAA8B;wBAC3C,IAAI,EAAE,CAAC,OAAO,CAAC;qBAChB;oBACD,IAAI,EAAE;wBACJ,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,yBAAyB;qBACvC;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,sBAAsB;wBACnC,IAAI,EAAE,CAAC,KAAK,CAAC;qBACd;oBACD,WAAW,EAAE;wBACX,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,gCAAgC;qBAC9C;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,kCAAkC;qBAChD;oBACD,gBAAgB,EAAE;wBAChB,IAAI,EAAE,SAAS;wBACf,WAAW,EAAE,oEAAoE;qBAClF;iBACF;aACF;YACD;gBACE,IAAI,EAAE,QAAQ;gBACd,UAAU,EAAE;oBACV,SAAS,EAAE;wBACT,IAAI,EAAE,QAAQ;qBACf;oBACD,QAAQ,EAAE;wBACR,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,kCAAkC;wBAC/C,IAAI,EAAE,CAAC,WAAW,CAAC;qBACpB;oBACD,IAAI,EAAE;wBACJ,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,yBAAyB;qBACvC;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,sBAAsB;wBACnC,IAAI,EAAE,CAAC,KAAK,CAAC;qBACd;oBACD,WAAW,EAAE;wBACX,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,gCAAgC;qBAC9C;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,qDAAqD;qBACnE;oBACD,gBAAgB,EAAE;wBAChB,IAAI,EAAE,SAAS;wBACf,WAAW,EAAE,oEAAoE;qBAClF;iBACF;aACF;YACD;gBACE,IAAI,EAAE,QAAQ;gBACd,UAAU,EAAE;oBACV,SAAS,EAAE;wBACT,IAAI,EAAE,QAAQ;qBACf;oBACD,QAAQ,EAAE;wBACR,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,2BAA2B;wBACxC,IAAI,EAAE,CAAC,IAAI,CAAC;qBACb;oBACD,IAAI,EAAE;wBACJ,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,yBAAyB;qBACvC;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,sBAAsB;wBACnC,IAAI,EAAE,CAAC,SAAS,CAAC;qBAClB;oBACD,cAAc,EAAE;wBACd,IAAI,EAAE,QAAQ;wBACd,WAAW,EACT,kGAAkG;qBACrG;oBACD,WAAW,EAAE;wBACX,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,gCAAgC;qBAC9C;oBACD,aAAa,EAAE;wBACb,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,qCAAqC;qBACnD;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,0DAA0D;qBACxE;oBACD,WAAW,EAAE;wBACX,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,0CAA0C;qBACxD;iBACF;aACF;YACD;gBACE,IAAI,EAAE,QAAQ;gBACd,UAAU,EAAE;oBACV,SAAS,EAAE;wBACT,IAAI,EAAE,QAAQ;qBACf;oBACD,QAAQ,EAAE;wBACR,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,8BAA8B;wBAC3C,IAAI,EAAE,CAAC,OAAO,CAAC;qBAChB;oBACD,IAAI,EAAE;wBACJ,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,yBAAyB;qBACvC;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,sBAAsB;wBACnC,IAAI,EAAE,CAAC,SAAS,CAAC;qBAClB;oBACD,WAAW,EAAE;wBACX,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,gCAAgC;qBAC9C;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,kCAAkC;qBAChD;iBACF;aACF;YACD;gBACE,IAAI,EAAE,QAAQ;gBACd,UAAU,EAAE;oBACV,SAAS,EAAE;wBACT,IAAI,EAAE,QAAQ;qBACf;oBACD,QAAQ,EAAE;wBACR,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,kCAAkC;wBAC/C,IAAI,EAAE,CAAC,WAAW,CAAC;qBACpB;oBACD,IAAI,EAAE;wBACJ,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,yBAAyB;qBACvC;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,sBAAsB;wBACnC,IAAI,EAAE,CAAC,SAAS,CAAC;qBAClB;oBACD,WAAW,EAAE;wBACX,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,gCAAgC;qBAC9C;oBACD,UAAU,EAAE;wBACV,IAAI,EAAE,QAAQ;wBACd,WAAW,EAAE,qDAAqD;qBACnE;iBACF;aACF;SACF;KACF;CACF,CAAC;AAEK,MAAM,OAAO,GAAG,KAAK,EAAE,MAAiB,EAAE,IAAyC,EAAE,EAAE;IAC5F,MAAM,EAAE,SAAS,EAAE,GAAG,IAAI,EAAE,GAAG,IAAW,CAAC;IAC3C,OAAO,IAAA,2BAAmB,EAAC,MAAM,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAC;AAC3E,CAAC,CAAC;AAHW,QAAA,OAAO,WAGlB;AAEF,kBAAe,EAAE,QAAQ,EAAR,gBAAQ,EAAE,IAAI,EAAJ,YAAI,EAAE,OAAO,EAAP,eAAO,EAAE,CAAC"}
|