fix(agent): prevent stuck loop by adding action history to LLM prompt

The UI agent had no memory of previous actions — each step was a fresh
single-shot LLM call. After typing and sending a message, the LLM saw
an empty text field and retyped the message in a loop.

- Add RECENT_ACTIONS (last 5 actions with text/result) to user prompt
- Add chat app completion detection rule to dynamic prompt
- Add send-success hints for WhatsApp and Messages apps
- Add git convention to CLAUDE.md (no co-author lines)

server/src/agent/llm.ts

@@ -327,6 +327,8 @@ RULES:
 - If SCREEN_CHANGE says "NOT changed", your last action had no effect — change strategy.
 - Do NOT repeat an action that already failed.
 - Say "done" as soon as the goal is achieved.
+- CHECK RECENT_ACTIONS before every step: if you already typed text and tapped send, do NOT type it again.
+- CHAT APP COMPLETION: After typing a message and tapping send in a chat app (WhatsApp, Messages, etc.), if the text field is now EMPTY and your message text appears in the conversation above, the message was SENT SUCCESSFULLY. Say "done" immediately.
 - COPY-PASTE: Use clipboard_set with text from SCREEN_CONTEXT (most reliable), then paste. Or just type directly.`;
 
   if (isStuck) {