[UX] Auto-appending /v1 to api_base is not intuitive and causes connection errors

[Code-MonkeyZhang]

Problem Description:

When configuring the api_base in config.yaml with provider: "openai", the current implementation automatically appends /v1 to the URL.
However most LLM providers (e.g., SiliconFlow, OpenAI, DeepSeek) provide full API endpoints in their documentation that already include the version suffix (e.g., https://api.siliconflow.cn/v1). Users intuitively copy and paste these full URLs. The current behavior results in an invalid double-suffix (e.g., .../v1/v1), causing connection failures (404 Not Found).

Steps to Reproduce

1. Open config.yaml.
2. Set provider to "openai".
3. Set api_base to a standard full URL, such as https://api.siliconflow.cn/v1.
4. Run mini-agent and attempt to send a message.

Actual Behavior

The agent fails to connect to the LLM provider. The logs show a 404 Not Found error because the client attempts to connect.

Expected Behavior

The LLMClient initialization logic should check if api_base already ends with /v1. If the suffix is present, it should use the URL as-is without appending it again. This is more intuitive workflows for users just read provider's documentation.

###Environment Information

• Python version: 3.11
• Operating system: macOS (Apple Silicon)
• Mini-Agent version: Current main branch (Dec 2025)