Cache commands so that they're instantly reusable

If you type “convert to mp4” on more than one file, you shouldn’t need to ask an LLM how to do this more than once, Substage should be able to cache the result of the command, and reuse it.

There are issues with this though…

Currently we rely on the LLM to insert the input file name(s) in the correct place in the Terminal command. For example:

ffmpeg -i input.mov output.mp4

Instead, we’d probably need to get the LLM to output:

ffmpeg -i $SELECTED_FILE $SELECTED_FILE_BASE_NAME.mp4

Which sounds fine, but:

• It gets more complicated when you have 2+ files selected.

• We can’t always ask the LLM to output variable file, since it would make certain things impossible: “rename these files, giving them more human friendly names”

• Even something simple like zipping up files would mean that it wouldn’t know a sensible filename to give the final zip; it would have to be archive.zip, output.zip or something equally generic.

For large LLMs, it’s easy to give a more nuanced prompt, saying “if it’s this type of request, use variable file names”. For smaller LLMs (especially local ones), this doesn’t work so well at all.

Needs some thought!