@fede0089/skill-eval 3.2.0 → 3.3.0

package/dist/reporters/index.js

@@ -1,2 +1 @@
 export { HtmlReporter } from './html-reporter.js';
-export { JsonReporter } from './json-reporter.js';

package/dist/utils/ndjson.js

@@ -62,6 +62,25 @@ export function parseStreamResult(output) {
         (typeof resultEvent.response === 'string' ? resultEvent.response : '');
     return { response: text };
 }
+/**
+ * Counts tool invocations and reads the terminal status from a stream-json blob.
+ * Both are needed to tell a legitimate short answer apart from an agent that
+ * stopped early or degenerated: a trial with tool calls and a 'success' status
+ * that still produced two characters of text is a very different failure from
+ * one that never started.
+ * Returns toolCalls: 0 and an undefined status when the blob carries no events.
+ */
+export function parseStreamStats(output) {
+    let toolCalls = 0;
+    let status;
+    for (const event of parseNdjsonEvents(output)) {
+        if (event.type === 'tool_use')
+            toolCalls += 1;
+        else if (event.type === 'result')
+            status = event.status;
+    }
+    return { toolCalls, status };
+}
 /**
  * Extracts token consumption stats from a Gemini CLI stream-json stdout blob.
  * Looks for a result event with a stats.total_tokens field.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fede0089/skill-eval",
-  "version": "3.2.0",
+  "version": "3.3.0",
   "description": "CLI to evaluate agent skills triggering and functionality",
   "main": "dist/index.js",
   "bin": {

package/dist/reporters/json-reporter.js

@@ -1,10 +0,0 @@
-import fs from 'fs';
-import path from 'path';
-import { Logger } from '../utils/logger.js';
-export class JsonReporter {
-    generate(report, runDir) {
-        const jsonPath = path.join(runDir, 'summary.json');
-        fs.writeFileSync(jsonPath, JSON.stringify(report, null, 2), 'utf-8');
-        Logger.write(`\n   Report: file://${jsonPath}\n`);
-    }
-}