Governed Autonomy

Governed Autonomy in Practice

Proworkbench follows a simple execution doctrine: propose first, execute second.

Core principle

The model can suggest actions and assemble plans, but risky side effects are routed through explicit controls.

Policy model

• Default posture is conservative for risky capabilities.
• Effective state is derived from global + per-risk + per-tool rules.
• High-impact operations can require explicit approval.
• Execution rights are separated from conversation.

Execution boundaries

PB keeps risky execution in WebChat governed flows with explicit invoke states. Telegram and Slack are communication surfaces, not execution surfaces.

Operational guardrails

• Doctor checks prerequisites and guides remediation.
• Runtime status makes provider/model state visible.
• Approvals queue centralizes high-risk decision points.
• Canvas preserves output history and continuity.

What this model does not claim

PB does not claim perfect security. It reduces operational risk through explicit controls, clear defaults, and runtime visibility.