feat: add /yolo toggle to auto-approve permissions in TUI

[fakecore]

Problem

In interactive TUI mode, permission prompts (bash, edit, read) interrupt workflow. --dangerously-skip-permissions only works for the headless run command — there is no way to skip permission prompts in the TUI.

Proposed Solution

Add a /yolo slash command and --yolo CLI flag that auto-approves all permission requests in the TUI. Explicit deny rules in config remain respected.

This mirrors the approach in upstream anomalyco/opencode#11831 and PR anomalyco/opencode#27855.

Use Cases

• Power users who trust the agent and want uninterrupted sessions
• Rapid prototyping where speed matters more than review
• Developers stepping away while the agent works

[entrptaher]

The cli already has a flag. --never-ask-questions, how does --yolo differs?

[fakecore]

The cli already has a flag. --never-ask-questions, how does --yolo differs?

They are different:

• --never-ask-questions: skips the question tool. When the agent wants to ask a clarification/decision question, it auto-resolves by letting the model choose. Permission prompts still remain.
• --yolo: skips permission prompts. When permission.asked is emitted, the TUI auto-approves it once. Explicit deny rules still apply.

So for example, with only --never-ask-questions, a destructive command like rm still goes through the bash permission flow and asks for approval. --yolo is what auto-approves that permission request.

[entrptaher]

Wow that's great explanation. +1