Before submitting your bug report
Relevant environment info
Description
Validations
Problem
The original "Auto-Accept Agent Edits" setting (#5657) was removed in #8310. While #8310 made file edit diffs auto-apply (which is great), the agent still pauses for approval on other actions like terminal commands. This means I'm still sitting there pressing Enter/approve for every single step during an agent session.
For those of us running local LLMs or working in low-risk environments, we want a true "YOLO mode" where the agent runs fully autonomously — file edits, terminal commands, and all other tool invocations — without stopping for approval at each step.
Solution
Bring back a global "Auto-Accept All Agent Actions" toggle (or similar) under Experimental Settings that bypasses approval prompts for all tool types, not just file edits. The original #5657 implementation was a great start but only covered edit tools. Ideally this would cover:
Keep it marked as experimental/dangerous with a warning, as the original implementation did. Power users understand the risk.
Related
To reproduce
No response
Log output
Before submitting your bug report
Relevant environment info
Description
Validations
Problem
The original "Auto-Accept Agent Edits" setting (#5657) was removed in #8310. While #8310 made file edit diffs auto-apply (which is great), the agent still pauses for approval on other actions like terminal commands. This means I'm still sitting there pressing Enter/approve for every single step during an agent session.
For those of us running local LLMs or working in low-risk environments, we want a true "YOLO mode" where the agent runs fully autonomously — file edits, terminal commands, and all other tool invocations — without stopping for approval at each step.
Solution
Bring back a global "Auto-Accept All Agent Actions" toggle (or similar) under Experimental Settings that bypasses approval prompts for all tool types, not just file edits. The original #5657 implementation was a great start but only covered edit tools. Ideally this would cover:
Keep it marked as experimental/dangerous with a warning, as the original implementation did. Power users understand the risk.
Related
To reproduce
No response
Log output