Copilot Agent shouldn't run command inside terminal emulator

[brianhuster]

Problems

1. Copilot will wait for the command to exit before it can even continues. But sometimes, it runs an interactive command, and then ... it will wait for forever.
2. VSCode's built-in terminal emulator uses interactive shell by default, and interactive shell is designed for humans, not AI. The shell could contains feature that AI doesn't know, for example, I use Vi mode in Bash, and the AI doesn't know that it must enter insert mode first (by pressing i or a) before it can writes command. I also uses ble.sh, and the AI doesn't know that to run multi-line command, it must press Ctrl-j instead of Enter. Anyway, interactive shell is not friendly to AI agent in my opinion

My proposal

• Should uses Node's API to run shell commands instead. The stdout and stderr of shell command could be streamed to users (and AI as well)

[iwangbowen]

The copilot agent should read terminal output and input commands automatically if necessary

[brianhuster]

The copilot agent should read terminal output and input commands automatically if necessary

Of course it needs to read command output, but that doesn't mean it needs to use the terminal emulator with interactive shell

[Tyriar]

This doesn't work the way that Copilot tries to use the terminal, it's "acting as a human" and needs an interactive session to run successive, dependent commands. For example it may cd into a different directory and run something there.

Commands getting stuck is a known issue that needs to get resolved separately. If you see this your best bet would be to make sure shell integration is working well (though it's meant to work regardless of this).