Better support for different types of responses and interaction

[dave1010]

Clipea is designed to give you a quick 1 line shell command to run. This should always be the priority.

With the shell integration, Clipea currently exits as soon as it gets the first shell command from the LLM.

But ideally Clipea would handle other responses better:

• More than 1 command to run
• explanatory text before or after the command

And also handle other interaction modes

• Continue chat mode, to refine the response, without starting from scratch.
• a REPL-like mode, where the LLM continually monitors the shell (expensive due to context length!)
• a multi stage mode. This would let the LLM get input and/or parse output
• re-act autonomous agent mode?

Need to consider how this works with shell integration. Easier if Clipea spawns sub processes but not ideal.

Could save the LLM output to a file and then stream it again if you run ?? by itself

[vividfog]

I asked ?? write a simple template for a docker build file and found out that indeed line-by-line is a different use case. As for explanation, perhaps this would help

?? -v delete all files that are cache files from this subfolder

The -v would make it explain what it's about to do and why.

There's a slippery slope in going too deep into the multiline and chat rabbit hole though.

There's open-interpreter, which is basically that. A chatbot with execution capabilities. But that same thing that makes it powerful, also makes it needy. Embracing the premise of being a Really Smart Autocomplete is what brought me here, so personally I'd like to see that all extensions always remain true to that premise.