ARIA's autonomy level controls how much agents ask for user confirmation before acting.
Levels
| Level | Derived From | Behavior |
|---|---|---|
interactive |
user_skill_level: beginner |
Always asks for confirmation before proceeding |
balanced |
user_skill_level: intermediate |
Auto-proceeds when the next step is obvious; asks on ambiguous actions |
yolo |
user_skill_level: expert |
Fully autonomous -- agents proceed without asking |
How It's Set
Autonomy is derived from user_skill_level in module.yaml, which is set during /aria-setup:
- beginner →
interactive-- recommended for new users or unfamiliar projects - intermediate →
balanced-- recommended for most users - expert →
yolo-- recommended for experienced users who trust the workflow
You can override autonomy_level directly in module.yaml if you want a different mapping.
What Each Level Controls
Step Menus
At natural pause points, agents may present an A/P/C menu:
- [A] Advanced -- configure step details or explore options
- [P] Party -- discuss the step in multi-agent mode
- [C] Continue -- proceed with defaults
| Autonomy | Menu Behavior |
|---|---|
| interactive | Always shown; waits for user input |
| balanced | Shown at key decision points; auto-selects Continue for routine steps |
| yolo | Never shown; always auto-continues |
Technique Selection
In brainstorming and research workflows:
| Autonomy | Behavior |
|---|---|
| interactive | User selects techniques manually |
| balanced | AI recommends techniques; auto-proceeds |
| yolo | AI selects and auto-proceeds |
Scope Confirmation
In PRD, architecture, and planning workflows:
| Autonomy | Behavior |
|---|---|
| interactive | Confirms scope, constraints, and goals with user |
| balanced | Infers scope from context; confirms only when ambiguous |
| yolo | Infers everything from context; proceeds directly |
Safety Overrides
Regardless of autonomy level, these actions always require user confirmation:
- Course corrections -- mid-implementation changes that affect scope or direction
- Destructive operations -- deleting issues, removing labels, resetting state
- Git operations -- force pushes, branch deletions, merging to main
These safety overrides cannot be bypassed by any autonomy setting.