Feature Request: Granular Auto-Approval Rules (like Claude Code)

What feature would you like to see?

Currently, YOLO mode (default_yolo = true or /yolo command) is all-or-nothing:

• OFF: Ask for approval on all non-readonly tool calls
• ON: Auto-approve ALL tool calls

I would like granular auto-approval controls similar to what Claude Code offers, where I can:

1. Auto-approve read-only commands (ReadFile, Glob, Grep, Shell read operations) while requiring confirmation for modifications
2. Configure approval rules based on tool paths/arguments (e.g., auto-approve writes to certain directories but not others)

Proposed configuration (Granular, like Claude Code):

[auto_approval]
default = "ask" # "ask", "allow", "deny"

[[auto_approval.rules]]
tool = "WriteFile"
path = "/tmp/**"
action = "allow" # Auto-approve writes to /tmp

[[auto_approval.rules]]
tool = "WriteFile"
path = "**/*.md"
action = "ask" # Always ask before writing .md files

[[auto_approval.rules]]
tool = "Shell"
pattern = "git *"
action = "allow" # Auto-approve git commands

[[auto_approval.rules]]
tool = "Shell"
pattern = "rm *"
action = "ask" # Always ask before rm commands

Use case: When working on large projects, I need to explore many files (reading) without constant interruptions, but I want to be asked before any file modifications - especially in sensitive areas like production config files or outside the working directory.

This would provide a safer middle ground between the current binary YOLO on/off options.

Additional information

No response

No files, tests, or entry points are named. Start by locating the current default_yolo configuration and /yolo command handling, then trace how tool approval decisions are made. Done means granular rules can distinguish read-only operations, paths or arguments, and allow, ask, or deny actions with coverage for the proposed cases.