🛡️ fix: Guard against null content parts in formatAgentMessages payload

Sanitize each message's content array with the existing filterMalformedContentParts
helper before passing the payload to formatAgentMessages, which reads part.type
without a null guard and crashes on corrupted history (e.g. content parts persisted
from aborted/partial streams via an OpenAI-compatible proxy).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: TomasPalsson

api/server/controllers/agents/client.js

@@ -886,13 +886,21 @@ class AgentClient extends BaseClient {
         (this.agentConfigs != null &&
           Array.from(this.agentConfigs.values()).some(hasDeepSeekAgent));
       const formatOptions = needsDeepSeekFormat ? { provider: Providers.DEEPSEEK } : undefined;
+      /** Strip null/malformed content parts before formatting; `formatAgentMessages` reads
+       *  `part.type` without a null guard and crashes on corrupted history (e.g. content parts
+       *  persisted from aborted or partial streams via an OpenAI-compatible proxy). */
+      const sanitizedPayload = payload.map((message) =>
+        Array.isArray(message.content)
+          ? { ...message, content: filterMalformedContentParts(message.content) }
+          : message,
+      );
       let {
         messages: initialMessages,
         indexTokenCountMap,
         summary: initialSummary,
         boundaryTokenAdjustment,
       } = formatAgentMessages(
-        payload,
+        sanitizedPayload,
         this.indexTokenCountMap,
         toolSet,
         skillPrimeResult?.skills,