[Bug] Parser fails to parse GLM-4.6v output - Box tags not recognized

[ZiryaNoov]

System Info / 系統信息

• OS: Windows 10
• Python: 3.14.0
• Model: GLM-4.6v (via API)
• Device: Samsung Android (ADB connected)

Who can help? / 谁可以帮助到您？

No response

Information / 问题信息

• The official example scripts / 官方的示例脚本
• My own modified scripts / 我自己修改的脚本和任务

Reproduction / 复现过程

Description

The parse_action() function in phone_agent/actions/handler.py fails to parse outputs from the new GLM-4.6v model because the model wraps its responses in special tokens (<|begin_of_box|>).

Steps to Reproduce

1. Clone the repository and install dependencies.
2. Install ADB Keyboard on the device.
3. Run the agent using the GLM-4.6v model endpoint:

   python main.py --base-url [https://open.bigmodel.cn/api/paas/v4/](https://open.bigmodel.cn/api/paas/v4/) --apikey YOUR_API_KEY --model glm-4.6v --lang en "Open Settings"
   

Expected behavior / 期待表现

The parser should strictly clean and strip the <|begin_of_box|> and <|end_of_box|> tags before attempting to parse the action.
Also, it should extract the command starting with "do(" or "finish(" to ignore any "thinking process" text that the GLM-4.6v model generates before the command.

I have temporarily fixed this locally by updating phone_agent/actions/handler.py to strip these tags and search for the command substring.

[ZiryaNoov]

Error Message
ValueError: Failed to parse action: <|begin_of_box|>I can see the Android home screen...<|end_of_box|>
<|begin_of_box|>do(action="Launch", app="Settings")<|end_of_box|>