Feat/genlayer plugin

[AgustinRamiroDiaz]

This is my first time contributing to this project. I've based my code on the docs, but mostly copying other plugins

Please let me know anything that can be improved :)

Relates to:

#972

Risks

None that I can think of

Background

What does this PR do?

Adds a plugin to be able to interact with GenLayer

What kind of change is this?

Feature

Why are we doing this? Any context or related work?

I want to be able to interact with GenLayer with eliza

Documentation changes needed?

Is there a section for plugin documentation?

Testing

None.
Is there a way to do some integration tests with the agent? Like being able to prompt it and go through the whole flow?

Where should a reviewer start?

packages/plugin-genlayer/src/index.ts

Detailed testing steps

As a user

• create a character with the genlayer plugin
• pnpm build && pnpm start
• prompt it to interact with GenLayer
• you should be able to interact with any of the actions added

Discord username

agustinramirodiaz

[AgustinRamiroDiaz]

Hey everyone!

What's the recommended way to extract parameters from an Action?

I saw some examples like in buttplug

https://github.com/ai16z/eliza/blob/7fb73d209a5d6d09a039515a088ff14b48691511/packages/plugin-buttplug/src/index.ts#L359-L373

where they add it to the examples inside options, but that didn't work for me (I was always getting empty options)

Then I saw other examples like in the discord client
https://github.com/ai16z/eliza/blob/7fb73d209a5d6d09a039515a088ff14b48691511/packages/client-discord/src/actions/joinvoice.ts#L147-L157
where there's an ad-hoc crafted state in order to call an LLM with composeContext + generateText

I like the latter solution more, and I could make it work with single messages that have the whole context like read the contrac A on method B with arguments C. I'd like to improve the state to also take into account memories and previous messages so that the user can have a conversation with the agent, instead of having to pass all the information on one message. The thing is that when I've passed the whole state I got answers that were really related to the personality of the character, and I got answers that refused to help like "I'm not here to do that".
How can I incorporate the state's memory without the personality of the character to gather the parameters for my action?

[AgustinRamiroDiaz]

Also somewhat related to my previous comment:
I'm currently using callbacks to print messages to the user based on the actions that I took. These messages are now static like Here is your output: 42. Is there a recommended way to pass this into the agent and let the agent answer back with it's character's tone and personality?

[lalalune]

@AgustinRamiroDiaz state is definitely where it should go, we're looking to refactor this to be a bit more extensible and less hardcoded in v2

[AgustinRamiroDiaz]

@lalalune thank you very much for that information.

And about my other question, is there a reason why callbacks are not stored on the state? Is there a recommended way to store it?