@wix/evalforge-evaluator 0.41.0 → 0.42.0
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/build/index.js +19 -11
- package/build/index.js.map +3 -3
- package/build/index.mjs +19 -11
- package/build/index.mjs.map +3 -3
- package/build/types/run-scenario/environment.d.ts +4 -2
- package/package.json +4 -4
|
@@ -3,11 +3,13 @@ import type { EvaluatorConfig } from '../config.js';
|
|
|
3
3
|
/**
|
|
4
4
|
* Prepare the working directory for a scenario.
|
|
5
5
|
* If a template is provided, download and extract it.
|
|
6
|
+
* If no template, create an empty directory so file creation/modification can be tracked.
|
|
6
7
|
*
|
|
7
8
|
* @param config - Evaluator configuration
|
|
8
9
|
* @param evalRunId - The evaluation run ID
|
|
9
10
|
* @param targetId - The target ID (prompt agent, skill, or agent)
|
|
11
|
+
* @param scenarioId - The scenario ID (for per-scenario isolation when no template)
|
|
10
12
|
* @param template - Optional pre-fetched template entity
|
|
11
|
-
* @returns The working directory path
|
|
13
|
+
* @returns The working directory path
|
|
12
14
|
*/
|
|
13
|
-
export declare function prepareWorkingDirectory(config: EvaluatorConfig, evalRunId: string, targetId: string, template?: Template): Promise<string
|
|
15
|
+
export declare function prepareWorkingDirectory(config: EvaluatorConfig, evalRunId: string, targetId: string, scenarioId: string, template?: Template): Promise<string>;
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@wix/evalforge-evaluator",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.42.0",
|
|
4
4
|
"description": "EvalForge Evaluator",
|
|
5
5
|
"bin": "./build/index.js",
|
|
6
6
|
"files": [
|
|
@@ -13,14 +13,14 @@
|
|
|
13
13
|
"build:types": "tsc --emitDeclarationOnly --outDir ./build/types",
|
|
14
14
|
"build": "yarn run clean && yarn run build:cjs && yarn run build:esm && yarn run build:types",
|
|
15
15
|
"lint": "eslint .",
|
|
16
|
-
"test": "
|
|
16
|
+
"test": "node --import tsx --test tests/**/*.test.ts"
|
|
17
17
|
},
|
|
18
18
|
"dependencies": {
|
|
19
19
|
"@ai-sdk/anthropic": "^3.0.2",
|
|
20
20
|
"@anthropic-ai/claude-agent-sdk": "^0.2.12",
|
|
21
21
|
"@anthropic-ai/claude-code": "^2.0.76",
|
|
22
22
|
"@wix/eval-assertions": "0.1.0",
|
|
23
|
-
"@wix/evalforge-types": "0.
|
|
23
|
+
"@wix/evalforge-types": "0.14.0",
|
|
24
24
|
"ai": "^6.0.6",
|
|
25
25
|
"tar": "^7.5.3",
|
|
26
26
|
"zod": "^4.3.5"
|
|
@@ -57,5 +57,5 @@
|
|
|
57
57
|
"artifactId": "evalforge-evaluator"
|
|
58
58
|
}
|
|
59
59
|
},
|
|
60
|
-
"falconPackageHash": "
|
|
60
|
+
"falconPackageHash": "057d6d34ae475738de64112e3b7e26f9a9a059acd289d3e6bf0210a3"
|
|
61
61
|
}
|