[Security Patch] Fix RCE vulnerability, secure API key storage, and resolve UI threading bugs

[Guettaf-hossam]

Overview

This Pull Request addresses the critical security and structural vulnerabilities reported in #2. It hardens the system against potential prompt injection attacks, secures user secrets, and stabilizes the Tkinter UI.

Key Changes

1. Secrets Management (.env Migration)

• Replaced the insecure config/api_keys.json plaintext storage with a secure .env implementation using python-dotenv.
• Added .gitignore to ensure .env and __pycache__ are never committed.
• Refactored _get_api_key() in main.py to use os.getenv() and removed the redundant file-read bug.

2. Human-in-the-Loop (HITL) Sandbox for Dangerous Tools

• Implemented a _confirm() helper using tkinter.messagebox.askyesno (with a terminal input() fallback).
• Gated all destructive and OS-level commands. Tools like cmd_control, code_helper (run/build/edit), and file_controller (delete/move/write) will now pause and explicitly ask for user permission before executing.
• If the user clicks "No", the tool safely returns "Action blocked by user." to the LLM, neutralizing any prompt injection RCE attempts.

3. Tkinter Thread-Safety Fix

• Bug fixed: ui.py was being modified directly from background asyncio/threading loops (e.g., write_log and speaking states), which causes Tkinter to freeze or crash.
• Solution: Introduced a thread-safe queue.Queue(). Background threads now only enqueue state changes, and a new _poll_queue() method running on the main thread (via root.after()) safely updates the UI.

Note for the Maintainer:

Upon merging this, users will be prompted to re-enter their Gemini API key on their first launch to generate the new .env file. Read-only actions (like listing files or reading code) remain autonomous and un-gated for a smooth experience.

Please let me know if you need any adjustments to these changes!

[jas-kha]

Resolve conflicts