Run same prompt on different files

[daliusd]

Hey,

I wonder if it is possible to run same prompt on different files without approval (e.g. I want to instruct AI to remove specific code). I found how to disable approval, there are events I can react to. So it seems it is doable. Before investing any time to that I wonder if someone already has done something like that?

[olimorris]

I don't understand what you're asking for. Can you explain in terms of a real use case?

[daliusd]

OK. Let's say I have following jsx file:

import React from 'react';

function HelloWorld() {
  return (
    <div>
      <h1>Hello, World!</h1>
      <p>This is paragraph</p>
    </div>
  );
}

export default HelloWorld;

I have following prompt:

#buffer
@editor

Wrap all strings with `t` function from react-i18next. Assume react-i18next is installed.

Result is following:

import React from 'react';
import { useTranslation } from 'react-i18next';

function HelloWorld() {
  const { t } = useTranslation();
  return (
    <div>
      <h1>{t("Hello, World!")}</h1>
      <p>{t("This is paragraph")}</p>
    </div>
  );
}

export default HelloWorld;

Let's assume result is good for me and I want now to do the same in remaining 200 jsx files in the project. I don't care if that will be done 100% correct - I will review diffs later. Can this be done using codecompanion?

[olimorris]

I understand. It might be possible with the @files tool. It could bulk update all the files in the repo without using the @editor

[daliusd]

But it will send one big request (e.g. with 200 files) instead of 200 requests? I guess LLMs might be inefficient with that. Or I am wrong?

[daliusd]

In the end codecompanion.nvim is not necessary right tool for all the tasks. I will investigate further idea updating multiple files via LLM from CLI which seems to be more viable option. There are some CLI apps to talk with different models, e.g. I found these two useful:

• https://github.com/kardolus/chatgpt-cli
• https://github.com/github/gh-models

https://ollama.com/ might be used as well BTW if running LLM locally is an option.

Now since I am forced to use GitHub Copilot in some situations I have created https://github.com/daliusd/copilot-cli where I will experiment further.