phase managers unclear to me

[sanderrenes]

Here I find the naming confusing, but I 'm not sure whether this is my lack of understanding or a need to clarify further.

is this the continuous-time manager for the phase?
https://github.com/iwanalabs/econagents/blob/154a422c01d3fdcc320d2caf391be47e461ee1ff/docs/source/Managing_Agents.rst?plain=1#L29

So how do these agent managers work in the code.
you have agent manager per agent, or per agent-phase?
How is the abstraction between phases and games implemented?

if we assume phase-based manager, that would imply that (in a multi-phase) game, an agent manager has a phase-manager per phase (so several).
Then I see TurnBasedPhaseManager, but what is the manager for the ContinuousTime phases?
Also, continuous has a standard meaning in some types of games (game theory), in that definition the tax declaration is a continuous choice (any number allowed). To prevent this linguistic issue, please differentiate between continuous-time and turn-based phases.

What happens here if there is a turn-based phase?
https://github.com/iwanalabs/econagents/blob/14b0b7af72a1cfb2bc3cecc21bbd7d0699a9c3ae/docs/source/Managing_Agents.rst?plain=1#L135

[dylanjcastillo]

Sorry, the naming is a bit confusing. It's an agent manager per agent. The PhaseManager is just a subclass of the AgentManager that includes the phase handling.

[sanderrenes]

more notes on the API:

The names of the game and phase handlers seem to indicate that you assume phases are turnBased, but if they are continuous you can call hybrid manager and indicate which are not turnBased.
If this is correct, we need to clarify this in the agent/game handler api documentation and in the definitions files.

Handle_phase:
"
This method will use a phase-specific handler if registered, otherwise it falls back to the default implementation using the LLM.
"

"handle_phase_with_llm

"Handle the phase using the LLM."
"This is the default implementation that uses the LLM to handle the phase by generating prompts, sending them to the LLM, and parsing the response."

--->Since all of our agents are LLMs too, we need to be more clear what happens here.

'Using the LLM' or 'uses the LLM to handle the phase by generating prompts' is not explained anywhere that I no off, link to that description, probably in a different definition/assumptions/workflow doc

--> We need to make a page with basic definitions (see definitions document)
Naming in game, role-phase, task/action,
With a picture showing the agent manager, with the phase managers under need for the prisoner's dilemma game, market game, and some hybrid game.
--> Do we need to add this to the graphic user interface?

--> Need to add the definitions of 'system prompt' and 'user prompt' as well, not every user will know LLMs and the current standard of dealing with prompts

--> Here we can also go into the gamestate information as well.
--> graphically difficult, can de the default LLM thing handle this?
--> can we give a list of information that the LLM should look for?

"send_messages"
"Send a message through the transport layer."
--> I do not know what this means, the transport layer is not explained anywhere