Action Governance

The Action Governance module controls what actions AI agents can take. It validates tools against an allowlist, scores risk, and enforces approval workflows for sensitive operations.

Features

• Tool allowlist: Only pre-approved tools can be invoked
• Parameter validation: Required parameters checked before execution
• Risk scoring: Composite score based on tool, action, and parameters
• HITL approval: High-risk actions require human approval
• Dry-run mode: Evaluate without executing
• Idempotency: Replay protection via idempotency keys

High-Risk Action Examples

• delete_record -- Permanent data deletion
• refund_approval -- Financial authorization
• bulk_update -- Mass data mutation
• customer_entitlement_override -- Access control changes

Endpoint

POST /v1/actions/authorize

Source

src/agentguard/action_governance/