@wix/evalforge-evaluator 0.204.0 → 0.206.0
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.
|
@@ -16,6 +16,10 @@ import type { EvaluationData } from '../fetch-evaluation-data.js';
|
|
|
16
16
|
* @param workDir - Optional working directory for the scenario
|
|
17
17
|
* @param pushEvent - Optional callback pushing trace events to the backend
|
|
18
18
|
* (remote job execution, where stdout is not accessible)
|
|
19
|
+
* @param executionPromptSuffix - Optional text appended to the prompt *sent to
|
|
20
|
+
* the agent* only (e.g. a provisioned "Site ID"). It is intentionally not part
|
|
21
|
+
* of the recorded result.triggerPrompt, so per-run additions don't fragment
|
|
22
|
+
* prompt-grouped result views.
|
|
19
23
|
* @returns Partial result without assertion fields
|
|
20
24
|
*/
|
|
21
|
-
export declare function runAgentWithContext(config: EvaluatorConfig, evalRunId: string, scenario: TestScenario, evalData: EvaluationData, workDir?: string, pushEvent?: (event: LiveTraceEvent) => void): Promise<PartialEvalRunResult>;
|
|
25
|
+
export declare function runAgentWithContext(config: EvaluatorConfig, evalRunId: string, scenario: TestScenario, evalData: EvaluationData, workDir?: string, pushEvent?: (event: LiveTraceEvent) => void, executionPromptSuffix?: string): Promise<PartialEvalRunResult>;
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@wix/evalforge-evaluator",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.206.0",
|
|
4
4
|
"description": "EvalForge Evaluator",
|
|
5
5
|
"bin": "./build/index.js",
|
|
6
6
|
"files": [
|
|
@@ -22,9 +22,9 @@
|
|
|
22
22
|
"@ai-sdk/openai": "^3.0.39",
|
|
23
23
|
"@anthropic-ai/claude-agent-sdk": "^0.2.63",
|
|
24
24
|
"@anthropic-ai/claude-code": "^2.1.63",
|
|
25
|
-
"@wix/eval-assertions": "0.
|
|
26
|
-
"@wix/evalforge-github-client": "0.
|
|
27
|
-
"@wix/evalforge-types": "0.
|
|
25
|
+
"@wix/eval-assertions": "0.77.0",
|
|
26
|
+
"@wix/evalforge-github-client": "0.76.0",
|
|
27
|
+
"@wix/evalforge-types": "0.101.0",
|
|
28
28
|
"ai": "^6.0.107",
|
|
29
29
|
"diff": "^7.0.0",
|
|
30
30
|
"tar": "^7.5.3",
|
|
@@ -72,5 +72,5 @@
|
|
|
72
72
|
"artifactId": "evalforge-evaluator"
|
|
73
73
|
}
|
|
74
74
|
},
|
|
75
|
-
"falconPackageHash": "
|
|
75
|
+
"falconPackageHash": "ff9d5554b6b06c20352da4522a7555b464effa2f2126657e1577f142"
|
|
76
76
|
}
|