Is it possible to set custom system instructions to the agent?

[zavocc]

I've been finding ways trying to shape the agent's behavior.. including when to use tools, how it should address itself, etc.

For instance, I want it to refer itself to something else instead of "GitHub Copilot CLI" since it wouldn't make sense to address itself that fits into CLI environment for let's say when I'm interacting with the agent within different contexts such as in Discord, Slack, or other GUI-based interactions.

In this sample code, I figured that I could use and set the system_message parameter in create_session method:

import asyncio
import copilot

async def main():
    client = copilot.CopilotClient()
    await client.start()

    session = await client.create_session({
        "session_id": "test-session-123",
        "model": "gpt-4.1",
        "system_message": "Your name is Jakey the Discord bot. [+ other instructions...]"
    })
    response = await session.send_and_wait({"prompt": "What is your name?"}) # Still says 'GitHub Copilot CLI' instead of 'Jakey the Discord bot'

    print(response.data.content)

    await session.destroy()

    # Delete session
    await client.delete_session("test-session-123")
    
    # Stop
    await client.stop()

asyncio.run(main())

However, the problem is even setting system_message parameter which it should call itself "Jakey the Discord Bot", it seems ignored... still falls back to default response style and still addresses itself as "GitHub Copilot CLI"

Would really be nice if there's a way we can add system-level instructions so the agent behavior can be consistent when interacting with it.

[github-actions]

Thanks for opening this issue! 👋

I've labeled this as:

• sdk/python — Python SDK specific
• question — About SDK capabilities
• enhancement — Feature request for custom system instructions

Your use case for customizing the agent's identity in different contexts (Discord, Slack, etc.) is clear, and you've provided a helpful code example. The team will review whether the system_message parameter should support this customization or if there's an alternative approach.

In the meantime, the maintainers may have additional insights on the intended behavior of system_message and whether custom agent identity is currently supported.

AI generated by Issue Triage Agent