Roadmap

[alexeykudinkin]

Let's this be a meta-issue.

[alexeykudinkin]

Don't hesitate to push here the stuff you're want to see working up to Saturday morning.

[alexeykudinkin]

Some cases that would be nice to see:

• Ok, Google, what is a HashMap (or whatever)?
  • Some fancy response like ("Ok, Alexey, you're supposed to know what a HashMap is") would be super-cool
• Go to the assignment of the field 'X'
  • Here we'd need to maintain context (that we're inside some method of class that has field named 'X', etc., that'd Yaroslav shining moment)

[nalbion]

I've created a branch for integration with OpenAI.

The gpt-developer GitHub action provides an example of a prompt which can be used to programmatically interact with an LLM:

Commands:
read: Read files
    read <comma separated filenames>
write: Write contents to a file a file
    write <filename> <<EOF
    <contents>
    EOF
remove: Remove files
    remove <comma separated filenames>
commit: Set the commit message
    commit <message>
comment: Set the comment message
    comment <<EOF
    <message>
    EOF
exit: Exit the program
    exit

Existing files: {files}
You are interacting with software. Solve the issue detailed above using the commands documented above. You use commands to `read`, `write`, `remove` files in the code and then `comment` on the issue or `commit` the changes. `exit` to stop.
You need to use the commands. The text you write is being parsed by a custom software that executes the commands. There's no human on the other end.