chore: gpt review and gpt fix

Author: WillieRuemmele

src/commands/agent/test/run-eval.ts

@@ -62,14 +62,21 @@ async function callEvalApi(org: Org, payload: EvalPayload): Promise<{ results?:
 
 async function resolveAgent(org: Org, apiName: string): Promise<{ agentId: string; versionId: string }> {
   const conn = org.getConnection();
-  const botResult = await conn.query<{ Id: string }>(`SELECT Id FROM BotDefinition WHERE DeveloperName = '${apiName}'`);
+
+  // Escape single quotes to prevent SOQL injection
+  const escapedApiName = apiName.replace(/'/g, "\\'");
+
+  const botResult = await conn.query<{ Id: string }>(
+    `SELECT Id FROM BotDefinition WHERE DeveloperName = '${escapedApiName}'`
+  );
   if (!botResult.records.length) {
     throw messages.createError('error.agentNotFound', [apiName]);
   }
   const agentId = botResult.records[0].Id;
 
+  // Filter to published/active versions only
   const versionResult = await conn.query<{ Id: string }>(
-    `SELECT Id FROM BotVersion WHERE BotDefinitionId = '${agentId}' ORDER BY VersionNumber DESC LIMIT 1`
+    `SELECT Id FROM BotVersion WHERE BotDefinitionId = '${agentId}' AND Status = 'Published' ORDER BY VersionNumber DESC LIMIT 1`
   );
   if (!versionResult.records.length) {
     throw messages.createError('error.agentVersionNotFound', [apiName]);

src/evalNormalizer.ts

@@ -342,10 +342,9 @@ export function normalizeEvaluatorFields(steps: EvalStep[]): EvalStep[] {
       normalizeScoringEvaluator(normalized, evalType);
     } else if (ASSERTION_EVALUATORS.has(evalType)) {
       normalizeAssertionEvaluator(normalized, evalType);
-    } else if (!('metric_name' in normalized) && evalType.includes('.')) {
-      // Unknown evaluator type -- just auto-inject metric_name
-      normalized.metric_name = evalType.split('.')[1];
     }
+    // Don't inject metric_name for unknown evaluator types to avoid API validation errors
+    // Unknown evaluators like bot_response_rating and planner_topic_assertion don't use metric_name
 
     return normalized as EvalStep;
   });