Ironically the thing that makes this even remotely secure is that it doesn't look like it runs the command immediately, it copies it to clipboard. Ironic because that makes it not really any different than asking ChatGPT or similar to give you a command to use.
With that said, if it did run automatically that would be a compliance and security nightmare.
Overall, these days most of my terminal usage is inside my IDE which has a shortcut to use AI to write a commands and I imagine that will become a standard feature in all terminals soon enough so I don't see too much use in this.
Point is to actually avoid hallucinations, Think I didnt focus on it as much as I wanted.
To avoid hallucinations, make sure you use strict prompt when creating the chatbot/api key.
Combination of rag + strict prompt will make sure that AI does not go out of bounds of it contextualised knowledge and avoids using LLM's internal knowledge.
With that said, if it did run automatically that would be a compliance and security nightmare.
Overall, these days most of my terminal usage is inside my IDE which has a shortcut to use AI to write a commands and I imagine that will become a standard feature in all terminals soon enough so I don't see too much use in this.
Good work, it's just not for me.