Risk Guardrails & Approvals

Overview

Tenant risk configuration governs caps, approval requirements, and global settings.
Executor enforces guardrails before apply; Approval Manager gates high‑risk actions.
Safety Profiles: Tenants select from predefined profiles (Conservative, Balanced, Aggressive) or configure Custom settings.

Configuration model

Defaults in src/core/config.py:DEFAULT_RISK_CONFIG.
Tenant overlays via ConfigurationService.get_tenant_risk_config().
Risk Engine: AdvancedRiskEngine calculates risk scores dynamically based on:
- Profile Rules: Explicitly allowed/forbidden operations.
- Inherent Risk: Metadata from the operation registry (Low/Medium/High/Critical).
- Context: Cost impact, historical failure rates, and time of day.
No Magic Inference: Previous "Industry Template" inference (guessing risk based on org name) has been removed in favor of explicit Profile selection.

Safety Profiles

Profile	Description	Key Behaviors
Conservative	Maximum safety; manual approvals for all stateful changes.	Blocks sensitive ops (e.g., `grant_invoker`); requires approval for DB/BQ changes.
Balanced	The sweet spot; approvals for DBs, autonomous tuning for stateless.	Allows standard scaling; gates destructive ops.
Aggressive	Full autopilot; executes playbooks automatically.	Minimal gates; relies on rollback safety nets.

Rollback Safety

Default: Rollbacks are not auto-approved by default (auto_approve_rollbacks: False).
Intelligent Assessment: The Risk Engine analyzes rollback checkpoints:
- Safe: Compute scaling (e.g., Cloud Run memory) is Low Risk.
- Destructive: Data ops (e.g., BigQuery table revert, schema changes) or massive scale events are High Risk and require approval.

Approval flow

Guardrail examples