So here is V1:
Advancing Agent State:
- forward_model:
- this moves the agent one state change forward by querying the model with the current state history
- put current state into self.history and calls the self.history constructor to parse it into a prompt —>
message = "\n".join(messages)
logger.info(f"🤖 MODEL INPUT\n{message}")
self.history.append({"role": "user", "content": message, "agent": self.name})
return self.model.query(self.local_history)
- but what is the prompt?
- what is inside state_vars???
Version 2:
Advancing Agent State:
- forward_model:
- this moves the agent one state change forward by querying the model with the current state history
- put current state into self.history and calls the self.history constructor to parse it into a prompt —>
message = "\n".join(messages)
logger.info(f"🤖 MODEL INPUT\n{message}")
self.history.append({"role": "user", "content": message, "agent": self.name})
return self.model.query(self.local_history)
- but what is the prompt?
- what is inside state_vars???
If you copy and paste v1 and v2 into obsidian, you will observe that the last two lines of v2 are in fact no treated as bullet points as expected, but rather as a continuation of the previous code block. Would like to bring this bug to your attention
). To put a code block in a code block, use more backticks (or tildes, which I’m using because they’re easier to type on my phone).