Can before_agent_invocation modify prompt

[femto]

Hello, I want some custom behavior
before run_async, say adding some messages
(say if 10 steps not using Todo, nudge the agent to use Todo),
so basically the before_agent_invocation needs to modify prompt
and return modifid_prompt(actually should be a list of messages
that can be passed into llm).
Can before_agent_invocation do this?

[njbrake]

Hi, this is related to #609. We were waiting to hear about a user needing the feature, so now that we have some indication it would be useful, we're happy to work on it 🚀 .

I'll add it on my list of to-dos for the next couple weeks. The good thing is that in the case of before_agent_invocation you could also probably write a simple wrapper around the run_async call to manage the prompt modification you want outside of the agent loop.

Just to be clear about that callbacks meaning:

If you had

output = await agent.run_async(....)
output = await agent.run_async(....)
output = await agent.run_async(....)

The before_agent_invocation would be called a total of three times, once at the beginning of each run command.

If you wanted your logic to be checked in the middle of the run_async command, you may be looking for the before_llm_call or before_tool_execution callback triggers.