[Bug] AIChatBot plugin: conversation history corruption and system prompt removal

### Is there an existing issue for this? - [x] I have searched the existing issues ### Describe your issue ## Summary Three issues in the AI ChatBot plugin (`AIChatBot.java`): ### Bug 1: Wrong variable saved to assistant history (line ~75) ```java // Current (wrong): chatHistory.add(ChatCompletionMessageParam.ofAssistant(... .content(...Content.ofText(message)) // ❌ User's message // Should be: .content(...Content.ofText(response)) // ✅ AI's response ``` ### Bug 2: System prompt removed when history exceeds max length (line ~79) ```java // Current (wrong): if (chatHistory.size() > maxHistory) { chatHistory.removeFirst(); // ❌ Removes system prompt at index 0 } // Should be: while (chatHistory.size() > maxHistory && chatHistory.size() > 1) { chatHistory.remove(1); // ✅ Keep system prompt at index 0 } ``` ### Security Enhancement: Sandwich the system prompt To mitigate prompt injection attacks, the system prompt should be reinforced at **both the start AND end** of the message list before sending to the API: ```java // Before API call, add a reminder at the end: var messagesWithReminder = new ArrayList<>(chatHistory); messagesWithReminder.add(ChatCompletionMessageParam.ofSystem( ChatCompletionSystemMessageParam.builder() .content(Content.ofText("REMINDER: " + systemPrompt)) .build() )); ``` **Why?** - LLMs have recency bias (prioritize recent context) - Attackers can inject "ignore previous instructions" in chat - Sandwiching instructions makes jailbreaking significantly harder ### Impact - Bug 1: Corrupted conversation history - Bug 2: System prompt deleted after N messages → trivial prompt injection - Without sandwich: Easier prompt injection via chat spam ### Affected file `mod/src/main/java/com/soulfiremc/server/plugins/AIChatBot.java` ### How can we reproduce what you got? _No response_ ### SoulFire/Error Logs _No response_ ### SoulFire Version 2.3.0