@wix/evalforge-evaluator 0.213.0 → 0.215.0

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.
@@ -17,9 +17,9 @@ export interface EvaluatorConfig {
17
17
  evaluationsDir?: string;
18
18
  /**
19
19
  * Legacy REST URL for pushing trace events (set by the backend job runner
20
- * for remote jobs). Its presence enables remote trace push; the push itself
21
- * now goes through the gRPC `PushTraceEvent` RPC, so the URL value is only
22
- * used by the (disabled) diagnostics module.
20
+ * for remote jobs). Only its *presence* matters now: it flags the run as a
21
+ * remote job, which enables trace push via the gRPC `PushTraceEvent` RPC.
22
+ * The URL string itself is no longer dereferenced.
23
23
  */
24
24
  tracePushUrl?: string;
25
25
  /**
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@wix/evalforge-evaluator",
3
- "version": "0.213.0",
3
+ "version": "0.215.0",
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": "0.78.0",
26
- "@wix/evalforge-github-client": "0.78.0",
27
- "@wix/evalforge-types": "0.102.0",
25
+ "@wix/eval-assertions": "0.79.0",
26
+ "@wix/evalforge-github-client": "0.79.0",
27
+ "@wix/evalforge-types": "0.103.0",
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": "36cc03df7be04e44d3b4a2bffead6b62eba81da5a7b29f44c0d08f83"
75
+ "falconPackageHash": "528d97433d269d9c91ad06f34bd474e31877c318cb8be8b3fdefaa17"
76
76
  }
@@ -1,48 +0,0 @@
1
- /**
2
- * Diagnostics Module
3
- *
4
- * Runs comprehensive environment diagnostics before executing evaluations.
5
- * All results are reported via trace events for visibility in the UI.
6
- *
7
- * Each test produces REAL output that can be inspected in the logs.
8
- * Test failures don't throw - they just return a failed status and continue.
9
- *
10
- * IMPORTANT: All diagnostic output is sent to the backend via trace events
11
- * so you can see them without accessing the dev machine logs.
12
- */
13
- import type { EvaluatorConfig } from './config.js';
14
- /**
15
- * Result of a single diagnostic test.
16
- */
17
- export interface DiagnosticResult {
18
- /** Name of the test */
19
- name: string;
20
- /** Whether the test passed */
21
- passed: boolean;
22
- /** Detailed result information - contains REAL command outputs */
23
- details: Record<string, unknown>;
24
- /** Error message if the test failed */
25
- error?: string;
26
- /** Duration of the test in milliseconds */
27
- durationMs: number;
28
- }
29
- /**
30
- * All diagnostic results from a run.
31
- */
32
- export interface DiagnosticReport {
33
- /** Timestamp when diagnostics started */
34
- startedAt: string;
35
- /** Timestamp when diagnostics completed */
36
- completedAt: string;
37
- /** Total duration in milliseconds */
38
- totalDurationMs: number;
39
- /** Individual test results */
40
- tests: DiagnosticResult[];
41
- /** Summary counts */
42
- summary: {
43
- total: number;
44
- passed: number;
45
- failed: number;
46
- };
47
- }
48
- export declare function runDiagnostics(config: EvaluatorConfig, evalRunId: string): Promise<DiagnosticReport>;