fix(github_copilot): preserve system prompts and auto-add headers

[Chesars]

Relevant issues

Fixes #19873

Pre-Submission checklist

• I have added testing in the tests/litellm/ directory
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

🐛 Bug Fix

Changes

GitHub Copilot system prompts are not passed when routing requests because LiteLLM was converting role: system to role: assistant by default.

Why? This was a legacy workaround when the GitHub Copilot API didn't support system prompts.

Solution

1. Remove system→assistant conversion - The API now supports system prompts natively for all models (Claude, GPT, etc.)

2. Auto-add Copilot headers - Required headers (editor-version, user-agent, etc.) are now automatically injected for /chat/completions, same as /responses endpoint.

3. Deprecate flag - disable_copilot_system_to_assistant is now deprecated

Before

# Required extra_headers, system prompt ignored for Claude
response = completion(
    model="github_copilot/claude-haiku-4.5",
    messages=[
        {"role": "system", "content": "You are a coding assistant"},  # ❌ Converted to assistant
        {"role": "user", "content": "Hello"}
    ],
    extra_headers={  # ❌ Required manually
        "editor-version": "vscode/1.85.1",
        "Copilot-Integration-Id": "vscode-chat"
    }
)

After

# No extra_headers needed, system prompt works
response = completion(
    model="github_copilot/claude-haiku-4.5",
    messages=[
        {"role": "system", "content": "You are a coding assistant"},  # ✅ Works
        {"role": "user", "content": "Hello"}
    ]
)

Backwards Compatibility

• Manual extra_headers still work (user headers override defaults)
• disable_copilot_system_to_assistant flag exists but is ignored (deprecated)

Docs

• GitHub Copilot Provider

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Jan 31, 2026 5:26am

[nadavsinai-philips]

thanks for the quick iteration! i didn't even have time to validate first...
I now got our devops to merge the addition of disable_copilot_system_to_assistant: True to the config and I find that now Both Claude and GPT models seem to work fine, I guess that Github resolved the issue with the need for "assistant" label, so the whole option can be removed IMHO.

[Chesars]

You're welcome!. That's exactly what this PR does — it removes the system-to-assistant conversion by default, so you don't need to set disable_copilot_system_to_assistant: True. The flag is kept but marked as deprecated so users configs and code ( litellm.disable_copilot_system_to_assistant = True) don't break. Thanks!