improve documentation #15
Description
Could somebody please improve the documentation?
From what I can gather, a plugin must extend AutoGPTPluginTemplate and set:
- self._name
- self._version
- self._description
It should also implement can_handle_xxxx methods, returning True only if they do something other than pass in xxxx, the options of which are:
post_prompt
This method is called just after the generate_prompt is called, but actually before the prompt is generated.
...why?
This seems to be the only method that any of the existing plugins currently implement.
I can see that they register commands that they can support, presumably this is only called once. I'm not sure how the class registers it self though.
I can see that they define the args that are provided to the function being registered. "imap_search_command": "<imap_search_criteria_command>", seems to break a convention - any reason, or is the string content just arbitrary?
Parameters:
- prompt: PromptGenerator - okay, this is actually fairly well documented. The prompt is sent to the LLM
on_response
Parameters:
- response - I guess we're expecting it's as per below?
- *args - is this
{"arg name": "value"}? - *kwargs ...I've got no idea
Return
A string - but what should it look like? Who/what reads it and when/why?
...Is this just a string, as you'd see in a response from ChatGPT?
This method is called when a response is received from the model
response format
{
"thoughts": {
"text": "thought",
"reasoning": "reasoning",
"plan": "- short bulleted\n- list that conveys\n- long-term plan",
"criticism": "constructive self-criticism",
"speak": "thoughts summary to say to user",
},
"command": {"name": "command name", "args": {"arg name": "value"}},
}on_planning
This method is called before the planning chat completion is done
...Who is involved in this "planning chat"?
...What are the messages? Are they just {role: "", content: ""}? Can somebody please provide a realistic example?
Parameters
- prompt: PromptGenerator
- messages: List[Message]
Returns
Optional string - what/why?
post_planning
This method is called after the planning chat completion is done.
Parameters:
- response: str
Returns
- str - what should this be?
pre_instruction
This method is called before the instruction chat is done
Parameters
- messages (List[Message]): The list of context messages.
...can we get have an example?
Returns
the same messages?
on_instruction
This method is called when the instruction chat is done.
Parameters
- messages (List[Message]): The list of context messages.
...what do the context messages look like? Are they for humans or the system?
post_instruction
This method is called after the instruction chat is done
Args:
response (str): The response.
Returns:
str: The resulting response.
pre_command
This method is called before the command is executed
...some example use cases for some of these would be useful
Args:
command_name (str): The command name.
arguments (Dict[str, Any]): The arguments.
post_command
This method is called after the command is executed.
Args:
command_name (str): The command name.
response (str): The response.
Returns:
str: The resulting response.
handle_chat_completion
This method is called when the chat completion is done.
...the whole session or one turn?
Args:
messages (List[Message]): The messages.
model (str): The model name.
temperature (float): The temperature.
max_tokens (int): The max tokens.
user_input
This method is called to request user input to the user
- user_input (str): The question or prompt to ask the user.
returns the user input as a string
report
Generates a report to the user
...what's in the message?
- mesage: str