Are multiple system prompt messages "normal"?

[diego898]

Question

I noticed @agent.system_prompt creates another role:system message - meaning you could end up with several, depending on how many dynamic pieces you define.

I don't think this is documented anywhere explicitly, and I actually can't find a good source saying whether models (or which models) are able to reliably capture all these system messages if they are separate.

In addition, I'm across several pieces of code/issues that implicitly rely on the fact* that there would be a single system message that appears first, or not at all.

[github-actions]

This issue is stale, and will be closed in 3 days if no reply is received.

[DouweM]

@diego898 Can you please share the code/issues that implicitly rely on there being just a single system prompt message?

Our system prompts example shows off this behavior by using multiple @agent.system_prompt calls, which can be a mix of static and dynamic (context-dependent): https://ai.pydantic.dev/agents/#system-prompts.

We're not aware of any issues with multiple role: system messages in the models where we currently use them, but if you have come across across a model that doesn't like this, please let us know: the way PydanticAI models are implemented makes it easy to merge multiple subsequent SystemPromptParts into one if necessary, and we already do this for e.g. Anthropic.