[Feature Request] Implement a reflection/self-critique module for CodeAct

[ryanhoangt]

What problem or use case are you trying to solve?

I see this as a way for the agent to potentially recover from bad states, e.g.,

• After getting stuck in a loop
• For some swe-bench instances where the agent just went down the rabbit hole doing certain things. This can be a way for the agent to look back to the past after simply "reacting" in the moment.

Describe the UX of the solution you'd like

After some defined schedule, prompt the agent to do a self-reflection (probably as a user message).

Do you have thoughts on the technical implementation?

Describe alternatives you've considered

Additional context

Some relevant references:

• https://blog.langchain.dev/reflection-agents/

[ryanhoangt]

Relevant issue: #2221