[BUG]litellm.BadRequestError - LLM API Rejects Messages Containing <channel> Tags

[GiriRaj249]

Describe the bug When using Strix with certain LLMs (e.g., GPT-5, GPT-4.1, or LiteLLM proxy), the model returns an error:

Invalid request: You have passed a message containing tags in the content field.

Strix appears to send internal ChatGPT-style tags such as:

<analysis>
<final>
<assistant>
<user>
<channel>

These tags are not valid in OpenAI-compatible APIs, causing a 400 Bad Request.

To Reproduce

• Create a Strix agent (e.g., PentestAgent or custom agent).
• Execute a task that produces chain-of-thought–style output or internal tags.
• Strix sends the full output (including / ) to the LLM backend.
• LLM responds with an error: Invalid request: message contains <channel> tags

Expected behavior A clear and concise description of what you expected to happen.

Screenshots Added screenshot for refernce

System Information

OS: e.g., Kali Linux 2024.3 Strix Version / Commit: e.g., 0.1.18 Python Version: e.g., 3.12 LLM Used: e.g., GPT-5 via LiteLLM (OpenAI-compatible endpoint)

• Python Version: [e.g. 3.12]
• LLM Used: "openai/openai/gpt-oss-20b"

[GiriRaj249]

Any idea why this is happening here?