@agentuity/evals 0.1.24 → 0.1.25

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.js CHANGED
@@ -66,10 +66,10 @@ export function createPresetEval(config) {
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
- const evalInput = middleware
69
+ const evalInput = middleware?.transformInput
70
70
  ? middleware.transformInput(input)
71
71
  : input;
72
- const evalOutput = middleware
72
+ const evalOutput = middleware?.transformOutput
73
73
  ? middleware.transformOutput(output)
74
74
  : output;
75
75
  return config.handler(ctx, evalInput, evalOutput, currentOptions);
@@ -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;AA4BD,MAAM,UAAU,gBAAgB,CAI9B,MAUD;IAcA,OAAO,CAAC,SAAS,EAAE,EAAE;QACpB,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,eAAe,EAAE,GAAG,SAAS,IAAI,EAAE,CAAC;QAC9E,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,MAAM,SAAS,GAAG,UAAU;oBAC3B,CAAC,CAAC,UAAU,CAAC,cAAc,CAAC,KAAK,CAAC;oBAClC,CAAC,CAAE,KAAuC,CAAC;gBAC5C,MAAM,UAAU,GAAG,UAAU;oBAC5B,CAAC,CAAC,UAAU,CAAC,eAAe,CAAC,MAAM,CAAC;oBACpC,CAAC,CAAE,MAAyC,CAAC;gBAC9C,OAAO,MAAM,CAAC,OAAO,CAAC,GAAG,EAAE,SAAS,EAAE,UAAU,EAAE,cAAc,CAAC,CAAC;gBAClE,8DAA8D;YAC/D,CAAC,CAAQ;YACT,OAAO,EAAE,cAAc;SACvB,CAAC;IACH,CAAC,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;AA4BD,MAAM,UAAU,gBAAgB,CAI9B,MAUD;IAcA,OAAO,CAAC,SAAS,EAAE,EAAE;QACpB,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,eAAe,EAAE,GAAG,SAAS,IAAI,EAAE,CAAC;QAC9E,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,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;gBAC9C,OAAO,MAAM,CAAC,OAAO,CAAC,GAAG,EAAE,SAAS,EAAE,UAAU,EAAE,cAAc,CAAC,CAAC;gBAClE,8DAA8D;YAC/D,CAAC,CAAQ;YACT,OAAO,EAAE,cAAc;SACvB,CAAC;IACH,CAAC,CAAC;AACH,CAAC"}
package/dist/types.d.ts CHANGED
@@ -6,6 +6,7 @@ export type BaseEvalOptions = {
6
6
  /**
7
7
  * Middleware to transform agent input/output to preset eval input/output.
8
8
  * Allows reusing preset evals across agents with different schemas.
9
+ * At least one transform must be provided.
9
10
  *
10
11
  * @template TAgentInput - Agent's input type (inferred from generics or `any`)
11
12
  * @template TAgentOutput - Agent's output type (inferred from generics or `any`)
@@ -14,6 +15,9 @@ export type BaseEvalOptions = {
14
15
  */
15
16
  export type EvalMiddleware<TAgentInput, TAgentOutput, TEvalInput, TEvalOutput> = {
16
17
  transformInput: (agentInput: TAgentInput) => TEvalInput;
18
+ transformOutput?: (agentOutput: TAgentOutput) => TEvalOutput;
19
+ } | {
20
+ transformInput?: (agentInput: TAgentInput) => TEvalInput;
17
21
  transformOutput: (agentOutput: TAgentOutput) => TEvalOutput;
18
22
  };
19
23
  //# sourceMappingURL=types.d.ts.map
@@ -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;AAGxC,eAAO,MAAM,kBAAkB,EAAE,aAA0C,CAAC;AAE5E,MAAM,MAAM,eAAe,GAAG;IAC7B,KAAK,EAAE,aAAa,CAAC;CACrB,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,MAAM,cAAc,CAAC,WAAW,EAAE,YAAY,EAAE,UAAU,EAAE,WAAW,IAAI;IAChF,cAAc,EAAE,CAAC,UAAU,EAAE,WAAW,KAAK,UAAU,CAAC;IACxD,eAAe,EAAE,CAAC,WAAW,EAAE,YAAY,KAAK,WAAW,CAAC;CAC5D,CAAC"}
1
+ {"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAGxC,eAAO,MAAM,kBAAkB,EAAE,aAA0C,CAAC;AAE5E,MAAM,MAAM,eAAe,GAAG;IAC7B,KAAK,EAAE,aAAa,CAAC;CACrB,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/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@agentuity/evals",
3
- "version": "0.1.24",
3
+ "version": "0.1.25",
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": "0.1.24",
30
- "@agentuity/runtime": "0.1.24",
31
- "@agentuity/schema": "0.1.24",
29
+ "@agentuity/core": "0.1.25",
30
+ "@agentuity/runtime": "0.1.25",
31
+ "@agentuity/schema": "0.1.25",
32
32
  "@ai-sdk/groq": "latest",
33
33
  "ai": "^6.0.0"
34
34
  },
35
35
  "devDependencies": {
36
- "@agentuity/test-utils": "0.1.24",
36
+ "@agentuity/test-utils": "0.1.25",
37
37
  "@types/bun": "latest",
38
38
  "bun-types": "latest",
39
39
  "typescript": "^5.9.0"
package/src/_utils.ts CHANGED
@@ -150,10 +150,10 @@ export function createPresetEval<
150
150
  description: description ?? config.description,
151
151
  // eslint-disable-next-line @typescript-eslint/no-explicit-any
152
152
  handler: (async (ctx: EvalContext, input: any, output: any) => {
153
- const evalInput = middleware
153
+ const evalInput = middleware?.transformInput
154
154
  ? middleware.transformInput(input)
155
155
  : (input as InferSchemaOutput<TEvalInput>);
156
- const evalOutput = middleware
156
+ const evalOutput = middleware?.transformOutput
157
157
  ? middleware.transformOutput(output)
158
158
  : (output as InferSchemaOutput<TEvalOutput>);
159
159
  return config.handler(ctx, evalInput, evalOutput, currentOptions);
package/src/types.ts CHANGED
@@ -10,13 +10,19 @@ export type BaseEvalOptions = {
10
10
  /**
11
11
  * Middleware to transform agent input/output to preset eval input/output.
12
12
  * Allows reusing preset evals across agents with different schemas.
13
+ * At least one transform must be provided.
13
14
  *
14
15
  * @template TAgentInput - Agent's input type (inferred from generics or `any`)
15
16
  * @template TAgentOutput - Agent's output type (inferred from generics or `any`)
16
17
  * @template TEvalInput - Eval's expected input type
17
18
  * @template TEvalOutput - Eval's expected output type
18
19
  */
19
- export type EvalMiddleware<TAgentInput, TAgentOutput, TEvalInput, TEvalOutput> = {
20
- transformInput: (agentInput: TAgentInput) => TEvalInput;
21
- transformOutput: (agentOutput: TAgentOutput) => TEvalOutput;
22
- };
20
+ export type EvalMiddleware<TAgentInput, TAgentOutput, TEvalInput, TEvalOutput> =
21
+ | {
22
+ transformInput: (agentInput: TAgentInput) => TEvalInput;
23
+ transformOutput?: (agentOutput: TAgentOutput) => TEvalOutput;
24
+ }
25
+ | {
26
+ transformInput?: (agentInput: TAgentInput) => TEvalInput;
27
+ transformOutput: (agentOutput: TAgentOutput) => TEvalOutput;
28
+ };