add amend command, allow non command autocompleted prompts #466
Conversation
There was a problem hiding this comment.
LGTM - performs as expected and really useful to have.
1 question: I couldn't parse exactly when/why command_autocomplete would be False?
1 comment: In the future I wonder if it'd make sense to build our own wrapper around the ChatCompletionMessage, which could include a get_text_content method, and also track message metadata for history/sampler like the latest commitID, the git diff since the last user message (this would simplify undo a lot), etc. Could also have a single .create classmethod that selects the right OAI class based on role, which would save a ton of imports.
|
We don't want it to be true when asking y/n questions, when asking y/n/i, etc. We only want it to be true when we're actually checking for commands. |
mentat/command/commands/amend.py
Outdated
| @classmethod | ||
| def help_message(cls) -> str: | ||
| return ( | ||
| "Removes all context since last user request. Will not undo any edits made." |
There was a problem hiding this comment.
| "Removes all context since last user request. Will not undo any edits made." | |
| "Used to amend a previous user request. Works by reseting context to the state it was at the last request and prefills user input with the last request. Does not undo any edits." |
Fixes #462. Adds amend command and adds kwarg for having command autocomplete on a prompt.
Pull Request Checklist