🤖 Android "Ghost" System: AI Task Automation Agent

Ghost System is an intelligent, automated agent framework designed to bridge the gap between unstructured mobile communication and structured productivity tools.

It autonomously navigates a real Android device to scrape event details from WhatsApp group chats, automatically schedules them into Google Tasks, and auto-joins video meetings (Zoom/Meet/Teams) while ensuring your privacy settings are active.

---

🌟 Key Features

• 📱 Real-Device Automation: Uses ADB (Android Debug Bridge) to control a physical or emulated Android device—no hacking or unofficial APIs required.
• 🧠 LLM-Powered Parsing: Leverages Google Gemini 2.5 Flash to intelligently extract dates, times, and meeting links from natural language conversations.
• Smart WhatsApp Scraper:
  • Auto-navigates to specific groups defined in config.
  • Performs intelligent scrolling (historical lookup) to capture recent context.
  • Outputs structured JSON data.
• 📅 Google Tasks Auto-Entry:
  • Handles complex UI interactions (Title, Description, Date/Time pickers).
  • Smart Time Entry: Uses a fallback strategy to switch Google Tasks' clock UI to "Keyboard Mode" for precise time setting.
• 🎥 Auto-Meeting Joiner:
  • Multi-Platform: Supports Zoom, Google Meet, and Microsoft Teams.
  • Privacy First: Automatically toggles OFF Audio and Video before joining any meeting.
  • Secure Access: autonomously handles Meeting ID and Password entry.

---

🔄 The Workflow Pipeline

Phase 1: The Scraper (scraper_agent.py)

1. Turbo Nav: Force-stops and launches WhatsApp to ensure a clean state.
2. Search & Enter: Types the group name and enters the chat.
3. Scroll & Extract: Performs vertical swipes to load history and uses LLM to parse events into JSON.

Phase 2: The Task Scheduler (set_event.jinja2)

1. Launch: Opens Google Tasks.
2. Input: Types Title and Description (including links).
3. Smart Time: Switches UI to Keyboard Input Mode to set precise deadlines.
4. Save: Commits the task.

Phase 3: The Meeting Automator (join_meeting.jinja2)

1. Direct Launch: Uses adb shell monkey to launch Zoom/Meet/Teams directly.
2. Join Flow: Inputs Meeting ID.
3. Privacy Guard:
   • Critical: Identifies and taps "Turn off Video" and "Turn off Audio" toggles.
   • Verify: Ensures switches are active before proceeding.
4. Credential Entry: Handles Password/Passcode entry if prompted by the app.
5. Finalize: Taps "Join" and waits for the "Waiting for Host" screen.

🛠️ Tech Stack

• Core Language: Python 3.10+
• AI/LLM: llama-index, Google Gemini 2.5 Flash API
• Device Control: ADB (Android Debug Bridge) via subprocess
• Agent Framework: DroidRun (Custom wrapper for agentic reasoning)
• Data Validation: Pydantic

---

📂 Project Structure

ghost-system/
│
├── agents/
│   ├── __init__.py
│   ├── scraper_agent.py      # The "Eye": Navigates WhatsApp & extracts data
│   ├── models.py             # Pydantic models (GroupScrapeResult) for validation
|   |── meeeting_agent.py
|   |── event_agent.py
│   │
│   └── prompts/              # Prompt Engineering Hub
│       ├── __init__.py
│       ├── prompts.py        # Python loader/manager for templates
│       ├── set_event.jinja2   # Template for Task Creation
│       ├── scrape.jinja2   # Template for WhatsApp Parsing
│       └── join_meeting.jinja2  # Template for Zoom/Meet Automation
│
├── data/                     # Output folder for scraped JSON files
│
├── droidrun/                 # Core Agent Framework (Config, Agent, Executors)
│
├── main.py                   # The "Brain": Router logic & Agent orchestration
├── groups.json               # Config: List of WhatsApp groups to monitor
├── .env                      # Config: API Keys (GEMINI_API_KEY)
├── requirements.txt          # Python dependencies
└── README.md                 # This file