@agentuity/evals 0.1.39 → 0.1.41
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/dist/_utils.d.ts +9 -1
- package/dist/_utils.d.ts.map +1 -1
- package/dist/_utils.js +21 -2
- package/dist/_utils.js.map +1 -1
- package/dist/adversarial.d.ts +2 -0
- package/dist/adversarial.d.ts.map +1 -1
- package/dist/ambiguity.d.ts +2 -0
- package/dist/ambiguity.d.ts.map +1 -1
- package/dist/answer-completeness.d.ts +2 -0
- package/dist/answer-completeness.d.ts.map +1 -1
- package/dist/conciseness.d.ts +2 -0
- package/dist/conciseness.d.ts.map +1 -1
- package/dist/extraneous-content.d.ts +2 -0
- package/dist/extraneous-content.d.ts.map +1 -1
- package/dist/format.d.ts +2 -0
- package/dist/format.d.ts.map +1 -1
- package/dist/index.d.ts +1 -1
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +1 -1
- package/dist/index.js.map +1 -1
- package/dist/knowledge-retention.d.ts +2 -0
- package/dist/knowledge-retention.d.ts.map +1 -1
- package/dist/pii.d.ts +2 -0
- package/dist/pii.d.ts.map +1 -1
- package/dist/politeness.d.ts +2 -0
- package/dist/politeness.d.ts.map +1 -1
- package/dist/role-adherence.d.ts +2 -0
- package/dist/role-adherence.d.ts.map +1 -1
- package/dist/safety.d.ts +2 -0
- package/dist/safety.d.ts.map +1 -1
- package/dist/self-reference.d.ts +2 -0
- package/dist/self-reference.d.ts.map +1 -1
- package/dist/types.d.ts +30 -0
- package/dist/types.d.ts.map +1 -1
- package/dist/types.js.map +1 -1
- package/package.json +5 -5
- package/src/_utils.ts +33 -3
- package/src/index.ts +6 -1
- package/src/types.ts +37 -0
package/dist/_utils.d.ts
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import type { InferOutput, StandardSchemaV1 } from '@agentuity/core';
|
|
2
2
|
import { type CreateEvalConfig, type EvalContext, type EvalHandlerResult } from '@agentuity/runtime';
|
|
3
|
-
import type { BaseEvalOptions, EvalMiddleware } from './types';
|
|
3
|
+
import type { BaseEvalOptions, EvalLifecycleHooks, EvalMiddleware } from './types';
|
|
4
4
|
import { type LanguageModel } from 'ai';
|
|
5
5
|
export declare const DefaultEvalInputSchema: import("@agentuity/schema").ObjectSchema<{
|
|
6
6
|
request: import("@agentuity/schema").StringSchema;
|
|
@@ -48,6 +48,14 @@ type PresetEvalOverrides<TEvalInput extends StandardSchemaV1 | undefined, TEvalO
|
|
|
48
48
|
description?: string;
|
|
49
49
|
} & TOptions> & {
|
|
50
50
|
middleware?: EvalMiddleware<TAgentInput, TAgentOutput, InferSchemaOutput<TEvalInput>, InferSchemaOutput<TEvalOutput>>;
|
|
51
|
+
/**
|
|
52
|
+
* Called before the eval handler runs.
|
|
53
|
+
*/
|
|
54
|
+
onStart?: EvalLifecycleHooks<TAgentInput, TAgentOutput>['onStart'];
|
|
55
|
+
/**
|
|
56
|
+
* Called after the eval handler completes successfully.
|
|
57
|
+
*/
|
|
58
|
+
onComplete?: EvalLifecycleHooks<TAgentInput, TAgentOutput>['onComplete'];
|
|
51
59
|
};
|
|
52
60
|
type PresetEvalResult<TOptions extends BaseEvalOptions> = CreateEvalConfig<any, any> & {
|
|
53
61
|
name: string;
|
package/dist/_utils.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"_utils.d.ts","sourceRoot":"","sources":["../src/_utils.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACrE,OAAO,EAEN,KAAK,gBAAgB,EACrB,KAAK,WAAW,EAChB,KAAK,iBAAiB,EACtB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,KAAK,EAAE,eAAe,EAAE,cAAc,EAAE,MAAM,SAAS,CAAC;
|
|
1
|
+
{"version":3,"file":"_utils.d.ts","sourceRoot":"","sources":["../src/_utils.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACrE,OAAO,EAEN,KAAK,gBAAgB,EACrB,KAAK,WAAW,EAChB,KAAK,iBAAiB,EACtB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,KAAK,EAAE,eAAe,EAAE,kBAAkB,EAAE,cAAc,EAAE,MAAM,SAAS,CAAC;AAEnF,OAAO,EAAgB,KAAK,aAAa,EAAE,MAAM,IAAI,CAAC;AAGtD,eAAO,MAAM,sBAAsB;;;EAGjC,CAAC;AACH,eAAO,MAAM,uBAAuB;;EAElC,CAAC;AACH,MAAM,MAAM,gBAAgB,GAAG,OAAO,sBAAsB,CAAC;AAC7D,MAAM,MAAM,iBAAiB,GAAG,OAAO,uBAAuB,CAAC;AAE/D;;;;;;;;;;;GAWG;AACH,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,MAAM,CAK7F;AAED,MAAM,MAAM,yBAAyB,GAAG;IACvC,KAAK,EAAE,aAAa,CAAC;IACrB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF;;;;;;;;;;;GAWG;AACH,wBAAsB,kBAAkB,CACvC,OAAO,EAAE,yBAAyB,GAChC,OAAO,CAAC,iBAAiB,CAAC,CA0B5B;AAID,KAAK,iBAAiB,CAAC,CAAC,IAAI,CAAC,SAAS,gBAAgB,GAAG,WAAW,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC;AAE9E,KAAK,mBAAmB,CACvB,UAAU,SAAS,gBAAgB,GAAG,SAAS,EAC/C,WAAW,SAAS,gBAAgB,GAAG,SAAS,EAChD,QAAQ,SAAS,eAAe,EAChC,WAAW,EACX,YAAY,IACT,OAAO,CAAC;IAAE,IAAI,CAAC,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,QAAQ,CAAC,GAAG;IACjE,UAAU,CAAC,EAAE,cAAc,CAC1B,WAAW,EACX,YAAY,EACZ,iBAAiB,CAAC,UAAU,CAAC,EAC7B,iBAAiB,CAAC,WAAW,CAAC,CAC9B,CAAC;IACF;;OAEG;IACH,OAAO,CAAC,EAAE,kBAAkB,CAAC,WAAW,EAAE,YAAY,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE;;OAEG;IACH,UAAU,CAAC,EAAE,kBAAkB,CAAC,WAAW,EAAE,YAAY,CAAC,CAAC,YAAY,CAAC,CAAC;CACzE,CAAC;AAIF,KAAK,gBAAgB,CAAC,QAAQ,SAAS,eAAe,IAAI,gBAAgB,CAAC,GAAG,EAAE,GAAG,CAAC,GAAG;IACtF,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,QAAQ,CAAC;CAClB,CAAC;AAEF,wBAAgB,gBAAgB,CAC/B,UAAU,SAAS,gBAAgB,GAAG,SAAS,GAAG,gBAAgB,GAAG,SAAS,EAC9E,WAAW,SAAS,gBAAgB,GAAG,SAAS,GAAG,gBAAgB,GAAG,SAAS,EAC/E,QAAQ,SAAS,eAAe,GAAG,eAAe,EACjD,MAAM,EAAE;IACT,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,CACR,GAAG,EAAE,WAAW,EAChB,KAAK,EAAE,iBAAiB,CAAC,UAAU,CAAC,EACpC,MAAM,EAAE,iBAAiB,CAAC,WAAW,CAAC,EACtC,OAAO,EAAE,QAAQ,KACb,UAAU,CAAC,gBAAgB,CAAC,UAAU,EAAE,WAAW,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC;IACtE,OAAO,EAAE,QAAQ,CAAC;CAClB,GAAG,CAEH,WAAW,SAAS,gBAAgB,GAAG,SAAS,GAAG,GAAG,EAEtD,YAAY,SAAS,gBAAgB,GAAG,SAAS,GAAG,GAAG,EAEvD,SAAS,CAAC,EAAE,mBAAmB,CAC9B,UAAU,EACV,WAAW,EACX,QAAQ,EACR,iBAAiB,CAAC,WAAW,CAAC,EAC9B,iBAAiB,CAAC,YAAY,CAAC,CAC/B,KACG,gBAAgB,CAAC,QAAQ,CAAC,CA4C9B"}
|
package/dist/_utils.js
CHANGED
|
@@ -59,20 +59,39 @@ export async function generateEvalResult(options) {
|
|
|
59
59
|
}
|
|
60
60
|
export function createPresetEval(config) {
|
|
61
61
|
return (overrides) => {
|
|
62
|
-
const { name, description, middleware, ...optionOverrides } = overrides ?? {};
|
|
62
|
+
const { name, description, middleware, onStart, onComplete, ...optionOverrides } = overrides ?? {};
|
|
63
63
|
const currentOptions = { ...config.options, ...optionOverrides };
|
|
64
64
|
return {
|
|
65
65
|
name: name ?? config.name,
|
|
66
66
|
description: description ?? config.description,
|
|
67
67
|
// eslint-disable-next-line @typescript-eslint/no-explicit-any
|
|
68
68
|
handler: (async (ctx, input, output) => {
|
|
69
|
+
// Call onStart hook if provided
|
|
70
|
+
if (onStart) {
|
|
71
|
+
try {
|
|
72
|
+
await onStart(ctx, input, output);
|
|
73
|
+
}
|
|
74
|
+
catch (error) {
|
|
75
|
+
ctx.logger.warn('Eval onStart hook failed', { error: String(error) });
|
|
76
|
+
}
|
|
77
|
+
}
|
|
69
78
|
const evalInput = middleware?.transformInput
|
|
70
79
|
? middleware.transformInput(input)
|
|
71
80
|
: input;
|
|
72
81
|
const evalOutput = middleware?.transformOutput
|
|
73
82
|
? middleware.transformOutput(output)
|
|
74
83
|
: output;
|
|
75
|
-
|
|
84
|
+
const result = await config.handler(ctx, evalInput, evalOutput, currentOptions);
|
|
85
|
+
// Call onComplete hook if provided
|
|
86
|
+
if (onComplete) {
|
|
87
|
+
try {
|
|
88
|
+
await onComplete(ctx, input, output, result);
|
|
89
|
+
}
|
|
90
|
+
catch (error) {
|
|
91
|
+
ctx.logger.warn('Eval onComplete hook failed', { error: String(error) });
|
|
92
|
+
}
|
|
93
|
+
}
|
|
94
|
+
return result;
|
|
76
95
|
// eslint-disable-next-line @typescript-eslint/no-explicit-any
|
|
77
96
|
}),
|
|
78
97
|
options: currentOptions,
|
package/dist/_utils.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"_utils.js","sourceRoot":"","sources":["../src/_utils.ts"],"names":[],"mappings":"AACA,OAAO,EACN,uBAAuB,GAIvB,MAAM,oBAAoB,CAAC;AAE5B,OAAO,EAAE,CAAC,EAAE,MAAM,mBAAmB,CAAC;AACtC,OAAO,EAAE,YAAY,EAAsB,MAAM,IAAI,CAAC;AAEtD,sEAAsE;AACtE,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAC9B,CAAC,CAAC;AACH,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC/C,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE;CACpB,CAAC,CAAC;AAIH;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,iBAAiB,CAAC,QAAgB,EAAE,SAAiC;IACpF,OAAO,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,MAAM,CACtC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,UAAU,CAAC,KAAK,GAAG,IAAI,EAAE,KAAK,CAAC,EAChE,QAAQ,CACR,CAAC;AACH,CAAC;AAQD;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACvC,OAAkC;IAElC,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,UAAU,GAAG,CAAC,EAAE,GAAG,OAAO,CAAC;IAElD,IAAI,SAA4B,CAAC;IAEjC,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;QACvD,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAErD,IAAI,CAAC;YACJ,4DAA4D;YAC5D,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,8BAA8B,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,CAAC;YAC3F,MAAM,QAAQ,GAAG,SAAS,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;YAE5D,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;YACpC,OAAO,uBAAuB,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QAC9C,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,SAAS,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;YAEtE,8BAA8B;YAC9B,IAAI,OAAO,KAAK,UAAU,GAAG,CAAC;gBAAE,MAAM;QACvC,CAAC;IACF,CAAC;IAED,MAAM,IAAI,KAAK,CACd,8CAA8C,UAAU,cAAc,SAAS,EAAE,OAAO,EAAE,CAC1F,CAAC;AACH,CAAC;
|
|
1
|
+
{"version":3,"file":"_utils.js","sourceRoot":"","sources":["../src/_utils.ts"],"names":[],"mappings":"AACA,OAAO,EACN,uBAAuB,GAIvB,MAAM,oBAAoB,CAAC;AAE5B,OAAO,EAAE,CAAC,EAAE,MAAM,mBAAmB,CAAC;AACtC,OAAO,EAAE,YAAY,EAAsB,MAAM,IAAI,CAAC;AAEtD,sEAAsE;AACtE,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAC9B,CAAC,CAAC;AACH,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC/C,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE;CACpB,CAAC,CAAC;AAIH;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,iBAAiB,CAAC,QAAgB,EAAE,SAAiC;IACpF,OAAO,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,MAAM,CACtC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,UAAU,CAAC,KAAK,GAAG,IAAI,EAAE,KAAK,CAAC,EAChE,QAAQ,CACR,CAAC;AACH,CAAC;AAQD;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACvC,OAAkC;IAElC,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,UAAU,GAAG,CAAC,EAAE,GAAG,OAAO,CAAC;IAElD,IAAI,SAA4B,CAAC;IAEjC,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;QACvD,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAErD,IAAI,CAAC;YACJ,4DAA4D;YAC5D,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,8BAA8B,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,CAAC;YAC3F,MAAM,QAAQ,GAAG,SAAS,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;YAE5D,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;YACpC,OAAO,uBAAuB,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QAC9C,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,SAAS,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;YAEtE,8BAA8B;YAC9B,IAAI,OAAO,KAAK,UAAU,GAAG,CAAC;gBAAE,MAAM;QACvC,CAAC;IACF,CAAC;IAED,MAAM,IAAI,KAAK,CACd,8CAA8C,UAAU,cAAc,SAAS,EAAE,OAAO,EAAE,CAC1F,CAAC;AACH,CAAC;AAoCD,MAAM,UAAU,gBAAgB,CAI9B,MAUD;IAcA,OAAO,CAAC,SAAS,EAAE,EAAE;QACpB,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,UAAU,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,eAAe,EAAE,GAC/E,SAAS,IAAI,EAAE,CAAC;QACjB,MAAM,cAAc,GAAG,EAAE,GAAG,MAAM,CAAC,OAAO,EAAE,GAAG,eAAe,EAAc,CAAC;QAE7E,OAAO;YACN,IAAI,EAAE,IAAI,IAAI,MAAM,CAAC,IAAI;YACzB,WAAW,EAAE,WAAW,IAAI,MAAM,CAAC,WAAW;YAC9C,8DAA8D;YAC9D,OAAO,EAAE,CAAC,KAAK,EAAE,GAAgB,EAAE,KAAU,EAAE,MAAW,EAAE,EAAE;gBAC7D,gCAAgC;gBAChC,IAAI,OAAO,EAAE,CAAC;oBACb,IAAI,CAAC;wBACJ,MAAM,OAAO,CAAC,GAAG,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC;oBACnC,CAAC;oBAAC,OAAO,KAAK,EAAE,CAAC;wBAChB,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,0BAA0B,EAAE,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;oBACvE,CAAC;gBACF,CAAC;gBAED,MAAM,SAAS,GAAG,UAAU,EAAE,cAAc;oBAC3C,CAAC,CAAC,UAAU,CAAC,cAAc,CAAC,KAAK,CAAC;oBAClC,CAAC,CAAE,KAAuC,CAAC;gBAC5C,MAAM,UAAU,GAAG,UAAU,EAAE,eAAe;oBAC7C,CAAC,CAAC,UAAU,CAAC,eAAe,CAAC,MAAM,CAAC;oBACpC,CAAC,CAAE,MAAyC,CAAC;gBAE9C,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,OAAO,CAAC,GAAG,EAAE,SAAS,EAAE,UAAU,EAAE,cAAc,CAAC,CAAC;gBAEhF,mCAAmC;gBACnC,IAAI,UAAU,EAAE,CAAC;oBAChB,IAAI,CAAC;wBACJ,MAAM,UAAU,CAAC,GAAG,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;oBAC9C,CAAC;oBAAC,OAAO,KAAK,EAAE,CAAC;wBAChB,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,6BAA6B,EAAE,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;oBAC1E,CAAC;gBACF,CAAC;gBAED,OAAO,MAAM,CAAC;gBACd,8DAA8D;YAC/D,CAAC,CAAQ;YACT,OAAO,EAAE,cAAc;SACvB,CAAC;IACH,CAAC,CAAC;AACH,CAAC"}
|
package/dist/adversarial.d.ts
CHANGED
|
@@ -11,6 +11,8 @@ export declare const adversarial: <TAgentInput extends import("@agentuity/core")
|
|
|
11
11
|
}, {
|
|
12
12
|
response: string;
|
|
13
13
|
} & {}> | undefined;
|
|
14
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
15
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
14
16
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
15
17
|
name: string;
|
|
16
18
|
options: BaseEvalOptions;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"adversarial.d.ts","sourceRoot":"","sources":["../src/adversarial.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,iBAAiB,i4EA8C5B,CAAC;AAEH,eAAO,MAAM,WAAW
|
|
1
|
+
{"version":3,"file":"adversarial.d.ts","sourceRoot":"","sources":["../src/adversarial.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,iBAAiB,i4EA8C5B,CAAC;AAEH,eAAO,MAAM,WAAW;;;;;;;;;;;;;;;;CActB,CAAC"}
|
package/dist/ambiguity.d.ts
CHANGED
|
@@ -16,6 +16,8 @@ export declare const ambiguity: <TAgentInput extends import("@agentuity/core").S
|
|
|
16
16
|
}, {
|
|
17
17
|
response: string;
|
|
18
18
|
} & {}> | undefined;
|
|
19
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
20
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
19
21
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
20
22
|
name: string;
|
|
21
23
|
options: AmbiguityEvalOptions;
|
package/dist/ambiguity.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ambiguity.d.ts","sourceRoot":"","sources":["../src/ambiguity.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,eAAe,2/DA+C1B,CAAC;AAEH,KAAK,oBAAoB,GAAG,eAAe,GAAG;IAC7C,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,SAAS;;;;eAHV,MAAM
|
|
1
|
+
{"version":3,"file":"ambiguity.d.ts","sourceRoot":"","sources":["../src/ambiguity.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,eAAe,2/DA+C1B,CAAC;AAEH,KAAK,oBAAoB,GAAG,eAAe,GAAG;IAC7C,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,SAAS;;;;eAHV,MAAM;;;;;;;;;;;;;;CA4BhB,CAAC"}
|
|
@@ -16,6 +16,8 @@ export declare const answerCompleteness: <TAgentInput extends import("@agentuity
|
|
|
16
16
|
}, {
|
|
17
17
|
response: string;
|
|
18
18
|
} & {}> | undefined;
|
|
19
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
20
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
19
21
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
20
22
|
name: string;
|
|
21
23
|
options: AnswerCompletenessEvalOptions;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"answer-completeness.d.ts","sourceRoot":"","sources":["../src/answer-completeness.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,wBAAwB,ilEAiDnC,CAAC;AAEH,KAAK,6BAA6B,GAAG,eAAe,GAAG;IACtD,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,kBAAkB;;;;eAHnB,MAAM
|
|
1
|
+
{"version":3,"file":"answer-completeness.d.ts","sourceRoot":"","sources":["../src/answer-completeness.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,wBAAwB,ilEAiDnC,CAAC;AAEH,KAAK,6BAA6B,GAAG,eAAe,GAAG;IACtD,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,kBAAkB;;;;eAHnB,MAAM;;;;;;;;;;;;;;CA2BhB,CAAC"}
|
package/dist/conciseness.d.ts
CHANGED
|
@@ -16,6 +16,8 @@ export declare const conciseness: <TAgentInput extends import("@agentuity/core")
|
|
|
16
16
|
}, {
|
|
17
17
|
response: string;
|
|
18
18
|
} & {}> | undefined;
|
|
19
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
20
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
19
21
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
20
22
|
name: string;
|
|
21
23
|
options: ConcisenessEvalOptions;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"conciseness.d.ts","sourceRoot":"","sources":["../src/conciseness.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,iBAAiB,g/DAgD5B,CAAC;AAEH,KAAK,sBAAsB,GAAG,eAAe,GAAG;IAC/C,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,WAAW;;;;eAHZ,MAAM
|
|
1
|
+
{"version":3,"file":"conciseness.d.ts","sourceRoot":"","sources":["../src/conciseness.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,iBAAiB,g/DAgD5B,CAAC;AAEH,KAAK,sBAAsB,GAAG,eAAe,GAAG;IAC/C,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,WAAW;;;;eAHZ,MAAM;;;;;;;;;;;;;;CA2BhB,CAAC"}
|
|
@@ -16,6 +16,8 @@ export declare const extraneousContent: <TAgentInput extends import("@agentuity/
|
|
|
16
16
|
}, {
|
|
17
17
|
response: string;
|
|
18
18
|
} & {}> | undefined;
|
|
19
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
20
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
19
21
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
20
22
|
name: string;
|
|
21
23
|
options: ExtraneousContentEvalOptions;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"extraneous-content.d.ts","sourceRoot":"","sources":["../src/extraneous-content.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,uBAAuB,woEAkDlC,CAAC;AAEH,KAAK,4BAA4B,GAAG,eAAe,GAAG;IACrD,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,iBAAiB;;;;eAHlB,MAAM
|
|
1
|
+
{"version":3,"file":"extraneous-content.d.ts","sourceRoot":"","sources":["../src/extraneous-content.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,uBAAuB,woEAkDlC,CAAC;AAEH,KAAK,4BAA4B,GAAG,eAAe,GAAG;IACrD,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,iBAAiB;;;;eAHlB,MAAM;;;;;;;;;;;;;;CA4BhB,CAAC"}
|
package/dist/format.d.ts
CHANGED
|
@@ -11,6 +11,8 @@ export declare const format: <TAgentInput extends import("@agentuity/core").Stan
|
|
|
11
11
|
}, {
|
|
12
12
|
response: string;
|
|
13
13
|
} & {}> | undefined;
|
|
14
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
15
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
14
16
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
15
17
|
name: string;
|
|
16
18
|
options: BaseEvalOptions;
|
package/dist/format.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"format.d.ts","sourceRoot":"","sources":["../src/format.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,YAAY,gjEA4CvB,CAAC;AAEH,eAAO,MAAM,MAAM
|
|
1
|
+
{"version":3,"file":"format.d.ts","sourceRoot":"","sources":["../src/format.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,YAAY,gjEA4CvB,CAAC;AAEH,eAAO,MAAM,MAAM;;;;;;;;;;;;;;;;CAcjB,CAAC"}
|
package/dist/index.d.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
export { createPresetEval, interpolatePrompt, generateEvalResult, type DefaultEvalInput, type DefaultEvalOutput, type GenerateEvalResultOptions, } from './_utils';
|
|
2
|
-
export { DEFAULT_EVAL_MODEL, type BaseEvalOptions, type EvalMiddleware } from './types';
|
|
2
|
+
export { DEFAULT_EVAL_MODEL, type BaseEvalOptions, type EvalLifecycleHooks, type EvalMiddleware, } from './types';
|
|
3
3
|
export { politeness, politenessPrompt } from './politeness';
|
|
4
4
|
export { safety, safetyPrompt } from './safety';
|
|
5
5
|
export { pii, piiPrompt } from './pii';
|
package/dist/index.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,EAClB,KAAK,gBAAgB,EACrB,KAAK,iBAAiB,EACtB,KAAK,yBAAyB,GAC9B,MAAM,UAAU,CAAC;AAClB,OAAO,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,EAClB,KAAK,gBAAgB,EACrB,KAAK,iBAAiB,EACtB,KAAK,yBAAyB,GAC9B,MAAM,UAAU,CAAC;AAClB,OAAO,EACN,kBAAkB,EAClB,KAAK,eAAe,EACpB,KAAK,kBAAkB,EACvB,KAAK,cAAc,GACnB,MAAM,SAAS,CAAC;AAGjB,OAAO,EAAE,UAAU,EAAE,gBAAgB,EAAE,MAAM,cAAc,CAAC;AAC5D,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAChD,OAAO,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,OAAO,CAAC;AACvC,OAAO,EAAE,WAAW,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAC/D,OAAO,EAAE,WAAW,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAC/D,OAAO,EAAE,SAAS,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AACzD,OAAO,EAAE,kBAAkB,EAAE,wBAAwB,EAAE,MAAM,uBAAuB,CAAC;AACrF,OAAO,EAAE,iBAAiB,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAC;AAClF,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAChD,OAAO,EAAE,kBAAkB,EAAE,wBAAwB,EAAE,MAAM,uBAAuB,CAAC;AACrF,OAAO,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAC"}
|
package/dist/index.js
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
export { createPresetEval, interpolatePrompt, generateEvalResult, } from './_utils';
|
|
2
|
-
export { DEFAULT_EVAL_MODEL } from './types';
|
|
2
|
+
export { DEFAULT_EVAL_MODEL, } from './types';
|
|
3
3
|
// Evals (each file contains both the prompt and the eval)
|
|
4
4
|
export { politeness, politenessPrompt } from './politeness';
|
|
5
5
|
export { safety, safetyPrompt } from './safety';
|
package/dist/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAIlB,MAAM,UAAU,CAAC;AAClB,OAAO,
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAIlB,MAAM,UAAU,CAAC;AAClB,OAAO,EACN,kBAAkB,GAIlB,MAAM,SAAS,CAAC;AAEjB,0DAA0D;AAC1D,OAAO,EAAE,UAAU,EAAE,gBAAgB,EAAE,MAAM,cAAc,CAAC;AAC5D,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAChD,OAAO,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,OAAO,CAAC;AACvC,OAAO,EAAE,WAAW,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAC/D,OAAO,EAAE,WAAW,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAC/D,OAAO,EAAE,SAAS,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AACzD,OAAO,EAAE,kBAAkB,EAAE,wBAAwB,EAAE,MAAM,uBAAuB,CAAC;AACrF,OAAO,EAAE,iBAAiB,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAC;AAClF,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,UAAU,CAAC;AAChD,OAAO,EAAE,kBAAkB,EAAE,wBAAwB,EAAE,MAAM,uBAAuB,CAAC;AACrF,OAAO,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAC"}
|
|
@@ -16,6 +16,8 @@ export declare const knowledgeRetention: <TAgentInput extends import("@agentuity
|
|
|
16
16
|
}, {
|
|
17
17
|
response: string;
|
|
18
18
|
} & {}> | undefined;
|
|
19
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
20
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
19
21
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
20
22
|
name: string;
|
|
21
23
|
options: KnowledgeRetentionEvalOptions;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"knowledge-retention.d.ts","sourceRoot":"","sources":["../src/knowledge-retention.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,wBAAwB,kxEAqDnC,CAAC;AAEH,KAAK,6BAA6B,GAAG,eAAe,GAAG;IACtD,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,kBAAkB;;;;eAHnB,MAAM
|
|
1
|
+
{"version":3,"file":"knowledge-retention.d.ts","sourceRoot":"","sources":["../src/knowledge-retention.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,wBAAwB,kxEAqDnC,CAAC;AAEH,KAAK,6BAA6B,GAAG,eAAe,GAAG;IACtD,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,kBAAkB;;;;eAHnB,MAAM;;;;;;;;;;;;;;CA2BhB,CAAC"}
|
package/dist/pii.d.ts
CHANGED
|
@@ -11,6 +11,8 @@ export declare const pii: <TAgentInput extends import("@agentuity/core").Standar
|
|
|
11
11
|
}, {
|
|
12
12
|
response: string;
|
|
13
13
|
} & {}> | undefined;
|
|
14
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
15
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
14
16
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
15
17
|
name: string;
|
|
16
18
|
options: BaseEvalOptions;
|
package/dist/pii.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"pii.d.ts","sourceRoot":"","sources":["../src/pii.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,SAAS,gvDAyCpB,CAAC;AAEH,eAAO,MAAM,GAAG
|
|
1
|
+
{"version":3,"file":"pii.d.ts","sourceRoot":"","sources":["../src/pii.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,SAAS,gvDAyCpB,CAAC;AAEH,eAAO,MAAM,GAAG;;;;;;;;;;;;;;;;CAcd,CAAC"}
|
package/dist/politeness.d.ts
CHANGED
|
@@ -16,6 +16,8 @@ export declare const politeness: <TAgentInput extends import("@agentuity/core").
|
|
|
16
16
|
}, {
|
|
17
17
|
response: string;
|
|
18
18
|
} & {}> | undefined;
|
|
19
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
20
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
19
21
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
20
22
|
name: string;
|
|
21
23
|
options: PolitenessEvalOptions;
|
package/dist/politeness.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"politeness.d.ts","sourceRoot":"","sources":["../src/politeness.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,gBAAgB,ktCAuC3B,CAAC;AAEH,KAAK,qBAAqB,GAAG,eAAe,GAAG;IAC9C,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,UAAU;;;;eAHX,MAAM
|
|
1
|
+
{"version":3,"file":"politeness.d.ts","sourceRoot":"","sources":["../src/politeness.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,gBAAgB,ktCAuC3B,CAAC;AAEH,KAAK,qBAAqB,GAAG,eAAe,GAAG;IAC9C,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,UAAU;;;;eAHX,MAAM;;;;;;;;;;;;;;CA2BhB,CAAC"}
|
package/dist/role-adherence.d.ts
CHANGED
|
@@ -16,6 +16,8 @@ export declare const roleAdherence: <TAgentInput extends import("@agentuity/core
|
|
|
16
16
|
}, {
|
|
17
17
|
response: string;
|
|
18
18
|
} & {}> | undefined;
|
|
19
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
20
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
19
21
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
20
22
|
name: string;
|
|
21
23
|
options: RoleAdherenceEvalOptions;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"role-adherence.d.ts","sourceRoot":"","sources":["../src/role-adherence.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,mBAAmB,g5EAsD9B,CAAC;AAEH,KAAK,wBAAwB,GAAG,eAAe,GAAG;IACjD,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,aAAa;;;;eAHd,MAAM
|
|
1
|
+
{"version":3,"file":"role-adherence.d.ts","sourceRoot":"","sources":["../src/role-adherence.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,mBAAmB,g5EAsD9B,CAAC;AAEH,KAAK,wBAAwB,GAAG,eAAe,GAAG;IACjD,SAAS,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF,eAAO,MAAM,aAAa;;;;eAHd,MAAM;;;;;;;;;;;;;;CA2BhB,CAAC"}
|
package/dist/safety.d.ts
CHANGED
|
@@ -11,6 +11,8 @@ export declare const safety: <TAgentInput extends import("@agentuity/core").Stan
|
|
|
11
11
|
}, {
|
|
12
12
|
response: string;
|
|
13
13
|
} & {}> | undefined;
|
|
14
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
15
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
14
16
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
15
17
|
name: string;
|
|
16
18
|
options: BaseEvalOptions;
|
package/dist/safety.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"safety.d.ts","sourceRoot":"","sources":["../src/safety.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,YAAY,+uFA8CvB,CAAC;AAEH,eAAO,MAAM,MAAM
|
|
1
|
+
{"version":3,"file":"safety.d.ts","sourceRoot":"","sources":["../src/safety.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,YAAY,+uFA8CvB,CAAC;AAEH,eAAO,MAAM,MAAM;;;;;;;;;;;;;;;;CAcjB,CAAC"}
|
package/dist/self-reference.d.ts
CHANGED
|
@@ -11,6 +11,8 @@ export declare const selfReference: <TAgentInput extends import("@agentuity/core
|
|
|
11
11
|
}, {
|
|
12
12
|
response: string;
|
|
13
13
|
} & {}> | undefined;
|
|
14
|
+
onStart?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any) => void | Promise<void>) | undefined;
|
|
15
|
+
onComplete?: ((ctx: import("@agentuity/runtime").EvalContext, input: TAgentInput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentInput> : any, output: TAgentOutput extends import("@agentuity/core").StandardSchemaV1<unknown, unknown> ? import("@agentuity/core").InferOutput<TAgentOutput> : any, result: import("@agentuity/runtime").EvalHandlerResult) => void | Promise<void>) | undefined;
|
|
14
16
|
}) | undefined) => import("@agentuity/runtime").CreateEvalConfig<any, any> & {
|
|
15
17
|
name: string;
|
|
16
18
|
options: BaseEvalOptions;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"self-reference.d.ts","sourceRoot":"","sources":["../src/self-reference.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,mBAAmB,q3DAoC9B,CAAC;AAEH,eAAO,MAAM,aAAa
|
|
1
|
+
{"version":3,"file":"self-reference.d.ts","sourceRoot":"","sources":["../src/self-reference.ts"],"names":[],"mappings":"AAOA,OAAO,EAAsB,KAAK,eAAe,EAAE,MAAM,SAAS,CAAC;AAEnE,eAAO,MAAM,mBAAmB,q3DAoC9B,CAAC;AAEH,eAAO,MAAM,aAAa;;;;;;;;;;;;;;;;CAgBzB,CAAC"}
|
package/dist/types.d.ts
CHANGED
|
@@ -1,8 +1,38 @@
|
|
|
1
1
|
import type { LanguageModel } from 'ai';
|
|
2
|
+
import type { EvalContext, EvalHandlerResult } from '@agentuity/runtime';
|
|
2
3
|
export declare const DEFAULT_EVAL_MODEL: LanguageModel;
|
|
3
4
|
export type BaseEvalOptions = {
|
|
4
5
|
model: LanguageModel;
|
|
5
6
|
};
|
|
7
|
+
/**
|
|
8
|
+
* Lifecycle hooks for preset evals.
|
|
9
|
+
* Called before and after the eval handler executes.
|
|
10
|
+
* Hook errors are logged but don't prevent eval execution.
|
|
11
|
+
*
|
|
12
|
+
* @template TInput - The input type passed to hooks (agent's input, not transformed)
|
|
13
|
+
* @template TOutput - The output type passed to hooks (agent's output, not transformed)
|
|
14
|
+
*/
|
|
15
|
+
export type EvalLifecycleHooks<TInput = unknown, TOutput = unknown> = {
|
|
16
|
+
/**
|
|
17
|
+
* Called before the eval handler runs.
|
|
18
|
+
* Use for logging, setup, or metrics.
|
|
19
|
+
*
|
|
20
|
+
* @param ctx - Eval context with logger, tracer, storage access
|
|
21
|
+
* @param input - The agent's input (before middleware transform)
|
|
22
|
+
* @param output - The agent's output (before middleware transform)
|
|
23
|
+
*/
|
|
24
|
+
onStart?: (ctx: EvalContext, input: TInput, output: TOutput) => void | Promise<void>;
|
|
25
|
+
/**
|
|
26
|
+
* Called after the eval handler completes successfully.
|
|
27
|
+
* Use for logging results, cleanup, or metrics.
|
|
28
|
+
*
|
|
29
|
+
* @param ctx - Eval context with logger, tracer, storage access
|
|
30
|
+
* @param input - The agent's input (before middleware transform)
|
|
31
|
+
* @param output - The agent's output (before middleware transform)
|
|
32
|
+
* @param result - The eval handler's result
|
|
33
|
+
*/
|
|
34
|
+
onComplete?: (ctx: EvalContext, input: TInput, output: TOutput, result: EvalHandlerResult) => void | Promise<void>;
|
|
35
|
+
};
|
|
6
36
|
/**
|
|
7
37
|
* Middleware to transform agent input/output to preset eval input/output.
|
|
8
38
|
* Allows reusing preset evals across agents with different schemas.
|
package/dist/types.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAExC,OAAO,KAAK,EAAE,WAAW,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AAEzE,eAAO,MAAM,kBAAkB,EAAE,aAA0C,CAAC;AAE5E,MAAM,MAAM,eAAe,GAAG;IAC7B,KAAK,EAAE,aAAa,CAAC;CACrB,CAAC;AAEF;;;;;;;GAOG;AACH,MAAM,MAAM,kBAAkB,CAAC,MAAM,GAAG,OAAO,EAAE,OAAO,GAAG,OAAO,IAAI;IACrE;;;;;;;OAOG;IACH,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,WAAW,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAErF;;;;;;;;OAQG;IACH,UAAU,CAAC,EAAE,CACZ,GAAG,EAAE,WAAW,EAChB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,OAAO,EACf,MAAM,EAAE,iBAAiB,KACrB,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CAC1B,CAAC;AAEF;;;;;;;;;GASG;AACH,MAAM,MAAM,cAAc,CAAC,WAAW,EAAE,YAAY,EAAE,UAAU,EAAE,WAAW,IAC1E;IACA,cAAc,EAAE,CAAC,UAAU,EAAE,WAAW,KAAK,UAAU,CAAC;IACxD,eAAe,CAAC,EAAE,CAAC,WAAW,EAAE,YAAY,KAAK,WAAW,CAAC;CAC5D,GACD;IACA,cAAc,CAAC,EAAE,CAAC,UAAU,EAAE,WAAW,KAAK,UAAU,CAAC;IACzD,eAAe,EAAE,CAAC,WAAW,EAAE,YAAY,KAAK,WAAW,CAAC;CAC3D,CAAC"}
|
package/dist/types.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,MAAM,cAAc,CAAC;
|
|
1
|
+
{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,MAAM,cAAc,CAAC;AAGpC,MAAM,CAAC,MAAM,kBAAkB,GAAkB,IAAI,CAAC,oBAAoB,CAAC,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@agentuity/evals",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.41",
|
|
4
4
|
"license": "Apache-2.0",
|
|
5
5
|
"author": "Agentuity employees and contributors",
|
|
6
6
|
"type": "module",
|
|
@@ -26,14 +26,14 @@
|
|
|
26
26
|
"prepublishOnly": "bun run clean && bun run build"
|
|
27
27
|
},
|
|
28
28
|
"dependencies": {
|
|
29
|
-
"@agentuity/core": "0.1.
|
|
30
|
-
"@agentuity/runtime": "0.1.
|
|
31
|
-
"@agentuity/schema": "0.1.
|
|
29
|
+
"@agentuity/core": "0.1.41",
|
|
30
|
+
"@agentuity/runtime": "0.1.41",
|
|
31
|
+
"@agentuity/schema": "0.1.41",
|
|
32
32
|
"@ai-sdk/groq": "latest",
|
|
33
33
|
"ai": "^6.0.0"
|
|
34
34
|
},
|
|
35
35
|
"devDependencies": {
|
|
36
|
-
"@agentuity/test-utils": "0.1.
|
|
36
|
+
"@agentuity/test-utils": "0.1.41",
|
|
37
37
|
"@types/bun": "latest",
|
|
38
38
|
"bun-types": "latest",
|
|
39
39
|
"typescript": "^5.9.0"
|
package/src/_utils.ts
CHANGED
|
@@ -5,7 +5,7 @@ import {
|
|
|
5
5
|
type EvalContext,
|
|
6
6
|
type EvalHandlerResult,
|
|
7
7
|
} from '@agentuity/runtime';
|
|
8
|
-
import type { BaseEvalOptions, EvalMiddleware } from './types';
|
|
8
|
+
import type { BaseEvalOptions, EvalLifecycleHooks, EvalMiddleware } from './types';
|
|
9
9
|
import { s } from '@agentuity/schema';
|
|
10
10
|
import { generateText, type LanguageModel } from 'ai';
|
|
11
11
|
|
|
@@ -104,6 +104,14 @@ type PresetEvalOverrides<
|
|
|
104
104
|
InferSchemaOutput<TEvalInput>,
|
|
105
105
|
InferSchemaOutput<TEvalOutput>
|
|
106
106
|
>;
|
|
107
|
+
/**
|
|
108
|
+
* Called before the eval handler runs.
|
|
109
|
+
*/
|
|
110
|
+
onStart?: EvalLifecycleHooks<TAgentInput, TAgentOutput>['onStart'];
|
|
111
|
+
/**
|
|
112
|
+
* Called after the eval handler completes successfully.
|
|
113
|
+
*/
|
|
114
|
+
onComplete?: EvalLifecycleHooks<TAgentInput, TAgentOutput>['onComplete'];
|
|
107
115
|
};
|
|
108
116
|
|
|
109
117
|
// Return type is compatible with any agent's createEval method
|
|
@@ -142,7 +150,8 @@ export function createPresetEval<
|
|
|
142
150
|
>
|
|
143
151
|
) => PresetEvalResult<TOptions> {
|
|
144
152
|
return (overrides) => {
|
|
145
|
-
const { name, description, middleware, ...optionOverrides } =
|
|
153
|
+
const { name, description, middleware, onStart, onComplete, ...optionOverrides } =
|
|
154
|
+
overrides ?? {};
|
|
146
155
|
const currentOptions = { ...config.options, ...optionOverrides } as TOptions;
|
|
147
156
|
|
|
148
157
|
return {
|
|
@@ -150,13 +159,34 @@ export function createPresetEval<
|
|
|
150
159
|
description: description ?? config.description,
|
|
151
160
|
// eslint-disable-next-line @typescript-eslint/no-explicit-any
|
|
152
161
|
handler: (async (ctx: EvalContext, input: any, output: any) => {
|
|
162
|
+
// Call onStart hook if provided
|
|
163
|
+
if (onStart) {
|
|
164
|
+
try {
|
|
165
|
+
await onStart(ctx, input, output);
|
|
166
|
+
} catch (error) {
|
|
167
|
+
ctx.logger.warn('Eval onStart hook failed', { error: String(error) });
|
|
168
|
+
}
|
|
169
|
+
}
|
|
170
|
+
|
|
153
171
|
const evalInput = middleware?.transformInput
|
|
154
172
|
? middleware.transformInput(input)
|
|
155
173
|
: (input as InferSchemaOutput<TEvalInput>);
|
|
156
174
|
const evalOutput = middleware?.transformOutput
|
|
157
175
|
? middleware.transformOutput(output)
|
|
158
176
|
: (output as InferSchemaOutput<TEvalOutput>);
|
|
159
|
-
|
|
177
|
+
|
|
178
|
+
const result = await config.handler(ctx, evalInput, evalOutput, currentOptions);
|
|
179
|
+
|
|
180
|
+
// Call onComplete hook if provided
|
|
181
|
+
if (onComplete) {
|
|
182
|
+
try {
|
|
183
|
+
await onComplete(ctx, input, output, result);
|
|
184
|
+
} catch (error) {
|
|
185
|
+
ctx.logger.warn('Eval onComplete hook failed', { error: String(error) });
|
|
186
|
+
}
|
|
187
|
+
}
|
|
188
|
+
|
|
189
|
+
return result;
|
|
160
190
|
// eslint-disable-next-line @typescript-eslint/no-explicit-any
|
|
161
191
|
}) as any,
|
|
162
192
|
options: currentOptions,
|
package/src/index.ts
CHANGED
|
@@ -6,7 +6,12 @@ export {
|
|
|
6
6
|
type DefaultEvalOutput,
|
|
7
7
|
type GenerateEvalResultOptions,
|
|
8
8
|
} from './_utils';
|
|
9
|
-
export {
|
|
9
|
+
export {
|
|
10
|
+
DEFAULT_EVAL_MODEL,
|
|
11
|
+
type BaseEvalOptions,
|
|
12
|
+
type EvalLifecycleHooks,
|
|
13
|
+
type EvalMiddleware,
|
|
14
|
+
} from './types';
|
|
10
15
|
|
|
11
16
|
// Evals (each file contains both the prompt and the eval)
|
|
12
17
|
export { politeness, politenessPrompt } from './politeness';
|
package/src/types.ts
CHANGED
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import type { LanguageModel } from 'ai';
|
|
2
2
|
import { groq } from '@ai-sdk/groq';
|
|
3
|
+
import type { EvalContext, EvalHandlerResult } from '@agentuity/runtime';
|
|
3
4
|
|
|
4
5
|
export const DEFAULT_EVAL_MODEL: LanguageModel = groq('openai/gpt-oss-20b');
|
|
5
6
|
|
|
@@ -7,6 +8,42 @@ export type BaseEvalOptions = {
|
|
|
7
8
|
model: LanguageModel;
|
|
8
9
|
};
|
|
9
10
|
|
|
11
|
+
/**
|
|
12
|
+
* Lifecycle hooks for preset evals.
|
|
13
|
+
* Called before and after the eval handler executes.
|
|
14
|
+
* Hook errors are logged but don't prevent eval execution.
|
|
15
|
+
*
|
|
16
|
+
* @template TInput - The input type passed to hooks (agent's input, not transformed)
|
|
17
|
+
* @template TOutput - The output type passed to hooks (agent's output, not transformed)
|
|
18
|
+
*/
|
|
19
|
+
export type EvalLifecycleHooks<TInput = unknown, TOutput = unknown> = {
|
|
20
|
+
/**
|
|
21
|
+
* Called before the eval handler runs.
|
|
22
|
+
* Use for logging, setup, or metrics.
|
|
23
|
+
*
|
|
24
|
+
* @param ctx - Eval context with logger, tracer, storage access
|
|
25
|
+
* @param input - The agent's input (before middleware transform)
|
|
26
|
+
* @param output - The agent's output (before middleware transform)
|
|
27
|
+
*/
|
|
28
|
+
onStart?: (ctx: EvalContext, input: TInput, output: TOutput) => void | Promise<void>;
|
|
29
|
+
|
|
30
|
+
/**
|
|
31
|
+
* Called after the eval handler completes successfully.
|
|
32
|
+
* Use for logging results, cleanup, or metrics.
|
|
33
|
+
*
|
|
34
|
+
* @param ctx - Eval context with logger, tracer, storage access
|
|
35
|
+
* @param input - The agent's input (before middleware transform)
|
|
36
|
+
* @param output - The agent's output (before middleware transform)
|
|
37
|
+
* @param result - The eval handler's result
|
|
38
|
+
*/
|
|
39
|
+
onComplete?: (
|
|
40
|
+
ctx: EvalContext,
|
|
41
|
+
input: TInput,
|
|
42
|
+
output: TOutput,
|
|
43
|
+
result: EvalHandlerResult
|
|
44
|
+
) => void | Promise<void>;
|
|
45
|
+
};
|
|
46
|
+
|
|
10
47
|
/**
|
|
11
48
|
* Middleware to transform agent input/output to preset eval input/output.
|
|
12
49
|
* Allows reusing preset evals across agents with different schemas.
|