Scripting: Answer Yes/No dynamically

[ryanpeach]

Issue

Right now InputOutput.confirm_ask is being used as part of a single "run" given a user input.

aider/aider/coders/base_coder.py

Line 815 in 77d2bc5

self.io.user_input(with_message)

aider/aider/coders/base_coder.py

Line 864 in 77d2bc5

message = self.preproc_user_input(user_message)

aider/aider/coders/base_coder.py

Lines 855 to 858 in 77d2bc5

	self.check_for_file_mentions(inp)
	inp = self.check_for_urls(inp)
	return inp

aider/aider/coders/base_coder.py

Line 1598 in 77d2bc5

if self.io.confirm_ask(f"Add {rel_fname} to the chat?", group=group, allow_never=True):

However in scripting this is inconvienient, because you can only use InputOutput(yes=True), which you have to declare BEFORE Coder.create or Coder.run

Example as is:

from aider.coders import Coder
from aider.models import Model
from aider.io import InputOutput

# This is a list of files to add to the chat
fnames = ["greeting.py"]

model = Model("gpt-4-turbo")

# Create an io object (required) and optionally enable confirmations with yes=True
io = InputOutput(yes=True)

# Create a coder object
coder = Coder.create(main_model=model, io=io, fnames=fnames)

# This will execute one instruction on those files and then return the last LLM message
result = coder.run("make a script that prints hello world")
print(result)

Instead, logically, you should be able to run something like this to see if you want to respond yes or not:

fnames = []  # Pretend we didn't give it the filename here
io = InputOutput(yes=None)  # Don't set yes
...
result = coder.run("make a script that prints hello world. Use greetings.py")  # Instead we gave it here
print(result)

result: Do you want to add greetings.py to the context? [Y/n]

result = coder.run("yes")  # Or maybe coder.answer to provide verification

Now result should equal result from the previous run.

Further, we should be able to load this from history!

the chat.history.md file should allow "unanswered" steps:

aider/aider/coders/base_coder.py

Lines 464 to 468 in 77d2bc5

	if not self.done_messages and restore_chat_history:
	history_md = self.io.read_text(self.io.chat_history_file)
	if history_md:
	self.done_messages = utils.split_chat_history_markdown(history_md)
	self.summarize_start()

aider/aider/utils.py

Line 145 in 77d2bc5

def split_chat_history_markdown(text, include_tool=False):

This function could be modified to scan for answered Y/n questions to things such as adding a file or url, and auto-rerunning those commands.

But further, if there is an unanswered "Y/n" followed by a user "/yes" or "/no" then it adds that to the chat history file as the answer, and then proceeds.

I am willing to implement the demo.

Version and model info

No response

[ryanpeach]

I've reviewed everywhere confirm_ask is used:

1. Offer to open a url seen in the messages
2. Offer to add a file seen in the messages to context
3. Run pip install
4. Install playwright
5. Create a git repo
6. Add aider files to gitignore
7. Add command output to chat
8. Run shell command
9. Allow edits to files not added to chat
10. Create new file
11. Add file name to chat
12. Attempt to fix test errors
13. Attempt to fix lint errors
14. Try to proceed anyway?
15. Add url to chat

It would be great if we could individually answer these yes-no for scripting purposes at the least.