awesome-geelark-skill（很棒的吉拉克技能）

GeeLark Cloud Phone API Skill

⚠️ EXPERIMENTAL PROJECT: This skill is designed for AI agents. It contains safety mechanisms but should NOT be used directly in production without testing.

Configuration: assets/config.json (auto-loads token and base URL)

First Time Setup

1. Initialize configuration:

   python3 scripts/init_config.py

1. This creates assets/config.json with your credentials (protected by .gitignore).
2. All scripts automatically load credentials from this file.
3. Install Dependencies

    pip install uiautomator2 --break-system-packages

Quick Start

Basic Usage

List Cloud Phones

from scripts.geelark_boot_helper import list_cloud_phones

# Token auto-loads from config
phones = list_cloud_phones()
for phone in phones:
    print(f"{phone['serialName']} - {phone['id']}")

Boot Cloud Phone and Connect

from scripts.geelark_boot_helper import boot_and_connect

# Token auto-loads from config
adb_info = boot_and_connect("phone_id")
if adb_info:
    print(f"ADB: {adb_info['ip']}:{adb_info['port']}, PWD: {adb_info['pwd']}")

Core Workflow

Step 1: Pre-check (Balance & Cloud Phone Confirmation) ⭐⭐⭐

Before operating cloud phones:

1. Check account balance - Confirm sufficient balance
2. Query available cloud phones - Confirm which phone to operate

from scripts.geelark_client import GeeLarkClient
from scripts.geelark_boot_helper import list_cloud_phones

# Initialize client (auto-loads config)
client = GeeLarkClient(task_name="my_task", phone_id="batch")

# Check balance
wallet = client.wallet()
balance = wallet['data']['balance']
gift_money = wallet['data'].get('giftMoney', 0)

if balance <= 0 and gift_money <= 0:
    print("❌ Insufficient balance! Please recharge.")
    exit(1)

# List phones
phones = list_cloud_phones()
for phone in phones:
    print(f"  {phone['serialName']} (ID: {phone['id']})")

Step 2: Create Cloud Phone

# Create single phone
response = client.call("/open/v1/phone/addNew", {
    "mobileType": "Android 13",
    "data": [{"profileName": "MyPhone"}]
})

# Create multiple phones (requires Pro plan)
response = client.call("/open/v1/phone/addNew", {
    "mobileType": "Android 13",
    "data": [
        {"profileName": "Phone1"},
        {"profileName": "Phone2"}
    ]
})

Step 3: Boot and Enable ADB

from scripts.geelark_boot_helper import boot_and_connect

# This function handles:
# - Balance check
# - Cloud phone status check
# - Start cloud phone (if stopped)
# - Enable ADB (if disabled)
# - Return ADB connection info
# Token auto-loads from config

adb_info = boot_and_connect("phone_id")

Step 4: Install Application ⭐⭐⭐

Critical: Must use appVersionId, NOT appName

# Step 1: Find app and get appVersionId
response = client.call("/open/v1/app/installable/list", {
    "envId": "phone_id",
    "name": "TikTok",
    "page": 1,
    "pageSize": 20
})

app_version_id = response['data']['items'][0]['appVersionInfoList'][0]['id']

# Step 2: Install using appVersionId
response = client.call("/open/v1/app/install", {
    "envId": "phone_id",
    "appVersionId": app_version_id
})

Step 5: Manage Session with PhoneManager

Recommended: Use PhoneManager for session tracking and auto-close.

from scripts.phone_manager import PhoneManager

# Method 1: Context manager (recommended)
with PhoneManager(timeout_minutes=5) as manager:
    manager.start_monitor()
    
    # Connect using phone_id (required)
    d = manager.connect_device("phone_id_123", ip, port, pwd, name="Android15")
    
    # Perform operations
    d(text="Submit").click()
    manager.record_activity("phone_id_123")

# Automatically stops all phones and saves logs on exit

# Method 2: Manual control
manager = PhoneManager(timeout_minutes=5)
manager.start_monitor()
# ... operations ...
manager.stop_all()
manager.client.save_log()

Safety Mechanisms

Cloudphone Operations

Dependencies

pip install uiautomator2 --break-system-packages

Agent Workflow ⭐⭐⭐

Since this skill is designed for AI agents that need to recognize the current screen state, the recommended workflow is:

1. Identify UI Structure: Use uiautomator2 to dump the current page hierarchy (dump_hierarchy()), enabling the agent to understand the screen layout.

- Parse the XML hierarchy to locate target elements by text, resource-id, content-desc, or class - Extract element coordinates (bounds) for verification or fallback operations - If dump_hierarchy() fails or returns empty, proceed to Step 3 immediately

1. Prioritize uiautomator2 Operations: If elements can be identified and interacted with via the UI hierarchy, use uiautomator2 for clicks, inputs, and swipes.

- Click: d(text="Button").click() or d(resourceId="com.example:id/btn").click() - Input: d(text="Username").set_text("hello") - Swipe: d.swipe(x1, y1, x2, y2) or d.swipe_ext("up") - Wait: d(text="Expected Text").wait(timeout=10) before interaction - Always verify the element is visible and enabled before interaction

1. Verify & Retry: After each operation, execute dump_hierarchy() again to confirm the previous action succeeded.

- Check if the expected UI state changed (e.g., new screen appeared, text updated, dialog dismissed) - If verification fails, retry the operation up to 3 times with brief delays (2-3 seconds) - If all retries fail, log the error and proceed to Step 4

1. Fallback to ADB Screenshots: If elements cannot be recognized or operated via uiautomator2, use ADB screenshots to visually confirm the state and perform coordinate-based operations.

- Capture screenshot: d.screenshot("/tmp/screen.png") or via ADB screencap - Use computer vision or LLM to analyze the screenshot and identify target coordinates - Click coordinates: d.click(x, y) or subprocess.run(['adb', '-s', 'device', 'shell', 'input', 'tap', x, y]) - Swipe coordinates: subprocess.run(['adb', '-s', 'device', 'shell', 'input', 'swipe', x1, y1, x2, y2]) - Note: Coordinate-based operations are less reliable; always prefer uiautomator2 when possible

Basic Operations

import uiautomator2 as u2
import subprocess

# Connect
d = u2.connect(f"{adb_info['ip']}:{adb_info['port']}")

# Authenticate (MUST DO IMMEDIATELY)
subprocess.run(['adb', '-s', f"{adb_info['ip']}:{adb_info['port']}",
                'shell', 'glogin', adb_info['pwd']], capture_output=True)

# Get UI hierarchy (Step 1: Identify)
xml = d.dump_hierarchy()

# Find and click element (Step 2: Prioritize uiautomator2)
d(text="Submit").click()

# Input text
d(text="Username").set_text("hello world")

# Screenshot (Step 3: Fallback)
d.screenshot("/tmp/screen.png")

# Or using ADB
subprocess.run(['adb', '-s', f"{adb_info['ip']}:{adb_info['port']}",
                'shell', 'screencap', '-p', '/sdcard/screen.png'])
subprocess.run(['adb', '-s', f"{adb_info['ip']}:{adb_info['port']}",
                'pull', '/sdcard/screen.png', '/tmp/screen.png'])

See: references/cloudphone_operations.md for complete guide.

RPA Tasks

GeeLark provides built-in RPA tasks for 9 platforms:

• TikTok (login, follow, like, comment, edit, delete)
• Instagram (login, publish reels, warmup)
• Facebook (login, publish, auto-comment, active account)
• YouTube (publish short, publish video, active account)
• Reddit (publish image/video, warmup)
• Threads (publish image/video)
• Pinterest (publish image/video)
• X/Twitter (publish)
• Google (login, app download)

See: references/api_reference.md for complete endpoint list and parameters

Reference Documentation

Common Error Codes

See: references/api_reference.md for complete list.

🤖 AI Agent Execution Rules

Critical Directives:

1. Pre-flight Check: Always verify account balance (wallet()) before initiating any cloud phone operations.
2. Safety Enforcement: Direct deletion is blocked. You must use scripts/delete_helper.py for double-confirmation.
3. Identifiers: Use phone_id (UUID) for all API calls, not serialName.
4. RPA Parameter Quirks:

- YouTube: video parameter is a string. - Others (IG/FB/Reddit/etc.): video/images parameters are arrays. - Instagram: Field is Image (capital I), not images. - Facebook: Login field is Email, not account.

1. Lifecycle Management: You must stop a running phone before deleting it (otherwise error 43009).
2. Auto-Configuration: Credentials load automatically from assets/config.json. Do not pass token manually.
3. Logging: Always save operation logs to logs/cloudphone/ upon task completion.

Official Documentation: https://github.com/GeeLark/geelark-openapi