Copilot-like features

NeuroPilot/EvilPilot icon + light/dark mode change animation by Pasu4

Copilot Chat

The following features send a chat message to Neuro:

• Ask Neuro to fix/explain (using the /fix and /explain commands respectively)

The user can also send a message to Neuro directly using either @neuro, @evil, or @neuroapi (as appropriate). Doing so will reroute the prompt to Neuro and will await for her response.

Chat features will register a disposable chat action and force an action from Neuro with only the chat action. The state will be the relevant code (even if it is in a Neuro-unsafe path), and the query will be dependent on what subcommand and prompt was chosen/typed (if any). The action force’s state will always be not ephemeral.

Note

If you type the wrong name’s @, the prompt will be sent to GitHub Copilot, even if the @ is still highlighted in the chatbox (which can happen if you don’t clear the chatbox before changing the active name).

Inline Completions

When completions are requested, a new disposable complete_code action will be registered, and then Neuro will be forced to complete the code using the action. Similar to the chat features, the state will be the nearby code, and the context from the action force will always be not ephemeral.

The query (and the action description) will prompt Neuro to suggest only code, never worded responses.

You can adjust how many lines are sent to Neuro using the Before/After Context setting.

You can also adjust when inline completions are triggered with the Completion Trigger setting.

Tip

Unfortunately, requesting inline completions will trigger all completion providers (including Copilot), and since Copilot is required for the Copilot Chat window, you cannot simply disable it. You can, however, work around this by modifying User/Workspace Settings to make Copilot unable to talk to the API.

Simply paste this into your settings.json file:

"github.copilot.advanced": {
    "debug.overrideEngine": "someRandomString"
}