New LLMSingleActionAgent seems not working as expected

[ezg]

I ran the new examples from here: https://js.langchain.com/docs/modules/agents/agents/custom_llm_chat

Putting in some debug statements in CustomOutputParser.parse it looks like it never actually uses a tool, but rather (at least in the example) gets the Final answer in the first LLM call.

I'm suspecting this has to do with the fact that stop words are not properly stored in the constructor of LLMSingleactionAgent here: https://github.com/hwchase17/langchainjs/blob/fc2987cb746d4d25ae55178e08336163b7277317/langchain/src/agents/agent.ts#L101

Or in its base class:
https://github.com/hwchase17/langchainjs/blob/fc2987cb746d4d25ae55178e08336163b7277317/langchain/src/agents/agent.ts#L28