Autonomy Configuration¶
BMAtlassian supports three levels of agent autonomy, controlling how much user confirmation is required during workflows.
Configuration¶
Set autonomy_level in _bmad/atlassian/module.yaml:
Levels¶
Interactive (Default)¶
Agents always ask before proceeding. Every story selection, sprint decision, and scope confirmation requires explicit user input. Best for new projects or when learning the BMAD workflow.
Balanced¶
Agents auto-proceed when the choice is obvious (e.g., only one story in Review, only one epic incomplete) but ask for confirmation on ambiguous decisions. Balanced mode briefly reports what it auto-selected, giving you a moment to override.
YOLO¶
Fully autonomous. Agents select stories by priority, create sprints with defaults, and proceed through workflows without pausing. Only destructive operations (course corrections) still require explicit approval.
Behavior Matrix¶
| Scenario | Interactive | Balanced | YOLO |
|---|---|---|---|
| Select story (1 candidate) | Ask | Auto-select | Auto-select |
| Select story (N candidates) | Ask | Auto-select first by priority | Auto-select first by priority |
| Select epic for retrospective | Ask | Auto-detect + confirm | Auto-detect + proceed |
| Sprint planning decisions | Ask all | Ask sprint creation only | Auto-create + auto-assign |
| Document project scope | Ask | Ask | Auto: generate all |
| Brainstorming technique | Ask | AI-recommended | AI-recommended |
| PRD/Architecture confirmation | Ask (domain expertise) | Ask (domain expertise) | Ask (domain expertise) |
| Course correction approval | Ask (destructive) | Ask (destructive) | Ask (destructive) |
Always Interactive¶
Some workflows remain interactive regardless of autonomy level:
- Course Correction -- involves destructive scope changes that require explicit approval
- PRD/Architecture Creation -- these are collaborative workflows where the AI needs the user's domain knowledge; the "confirmation points" are actually collaborative input, not simple approvals
- Advanced Elicitation -- interactive creative tool where the user interaction IS the value
Orchestrator Integration¶
When running via /bmad-atlassian-orchestrate, the autonomy level determines how many confirmations the orchestrator pauses for. In yolo mode, the orchestrator runs continuously, dispatching agents and processing stories without interruption until it encounters a course correction or completes all available work.