Required inputs
- The agent or workflow being governed
- Permission scopes and risk levels
- Evaluation, monitoring, and incident signals
Harness · blueprint
Human Approval Harness
Routes consequential agent actions through scoped permissions, approval thresholds, and rollback controls.
PolicyApprovalsRollback
Intended outcome
Faster automation without giving up human accountability.
This component is described as an implementation contract: the outcome, inputs, controls, and acceptance signals should be agreed before deployment.
Operating controls
- Policy enforcement before action
- Approval gates and least-privilege access
- Traceable logs and a tested rollback path
Acceptance signals
- Blocks work outside the approved boundary
- Routes exceptions to the right human
- Preserves an inspectable decision trail
Access path
Start with a bounded implementation brief.
Define the business outcome, systems involved, permissions, acceptance tests, and accountable owner before requesting delivery.
Related building blocks
View all components