Refactor microchain agent creation

[gabrielfior]

See comment in the bottom of this description for a more detailed explanation. tl;dr is to refactor build_agent and make engine creation, function registration and agent creation independent from one another.

Also relevant here is to refactor agent_functions, which depend on the agent itself. One idea that comes to mind is to expand the PromptHandler and have it manage the state of the prompt, so that the agent always reads from that before each run (instead of directly changing the agent's prompt inside a function).

          > however not sure how to handle the fact that this assignment ...

Not sure I understand. That could just stay where it is?

So before:

def build_agent(
    system_prompt: str,
    ....
):
    engine = Engine()
    agent = Agent(engine, ...)
    agent.system_prompt = system_prompt.format(engine_help=engine.help)

After:

engine = ...

def build_agent(
    system_prompt: str,
    engine: Engine,
    ....
):
    agent = Agent(engine)
    agent.system_prompt = system_prompt

But my point is that sequence of operations should be:

• Build the engine
• Build the system prompt, either by formatting TRADING_AGENT_SYSTEM_PROMPT or by fetching a prompt from the db (not by looking for a particular string in the prompt)
• Build the agent.

With maybe_append_function_definitions_into_prompt the hackyness is just moved somewhere else.

_Originally posted by @evangriffiths in https://github.com/gnosis/prediction-market-agent/pull/238#discussion_r1639831484_

[evangriffiths]

@gabrielfior we think this has been resolved by the later changes you made in https://github.com/gnosis/prediction-market-agent/pull/238. Do you agree, and we can close it?

[gabrielfior]

Yes, this has been solved. Closing.