Automatic bypass for edit tool

[chezsmithy]

Validations

• I believe this is a way to improve. I'll try to join the Continue Discord for questions
• I'm not able to find an open issue that requests the same enhancement

Problem

When I choose to allow the edit tool to work in automatic mode, I’m still asked to approve each change before the tool proceeds.

Solution

Update the edit tool, so that when in automatic mode each of the diff apply questions is bypassed and each change is auto applied. This will implement full YOLO mode.

[chezsmithy]

I’m noting I’m willing to submit a PR to solve for this issue. I’m also interested in other solutions or suggestions on implementation.

[RomneyDa]

@chezsmithy this would be interesting! Thinking to keep uniform behavior for tool policies, this could be a settings toggle in the Settings Page. Potentially could also apply to chat codeblocks that match an existing file.

Top setting name candidates

• Full YOLO mode
• Edit Tool Auto-Apply
• Automatically Apply Agent Edits

I think a good start would be to

1. Add an "Automatically Apply Agent Edits" or similar settings toggle in the settings page that sets a new shared config value "autoApplyAgentEdits" or similar
2. In the Client Tool Implementation of the edit tool, reference the value of autoApplyAgentEdits to conditionally accept all pending diffs after applies complete (see accept all button). Then, streaming should automatically continue after complete without changing any other logic!

[RomneyDa]

Fixed by #5657

[dy]

Hmm, I don't see that option in my installed version 1.2.22. Quite annoying that z.ai asks every edit the approval command

[Foorack]

@dy Seems like it was removed here? #8310

[Foorack]

I cannot find any information AT ALL why they removed this feature. Makese no sense to me. Luckily I found you can configure Ollama as a provider in Copilot extension.

[nateriverdev]

I'm using the Continue VSCode extension and really need a YOLO mode so it can automatically approve all permissions, instead of me having to sit there pressing Enter for every single step
Additionally, if possible, please add auto-retry when errors occur, along with suggestions and a task list when interacting with it. :)
Thank you to the author, this is the best extension I've used so far for running local LLMs with VSCode.