Taxonomies Overview
A Mandaitor taxonomy is a structured definition of what an AI agent is allowed to do within a specific industry. It provides the shared vocabulary for creating precise and verifiable delegation mandates.
Each taxonomy consists of four building blocks:
| Building Block | Purpose | Example |
|---|---|---|
| Actions | Operations that can be delegated | construction.validation.approve |
| Resource Patterns | URI templates scoping where actions apply | construction:project:{projectId}/* |
| Constraint Templates | Reusable boundary conditions (time, budget) | Invoice approval limit of EUR 5,000 |
| Mandate Templates | Pre-built delegation blueprints | "Automated Plan Validation" template |
Why Taxonomies?
Taxonomies solve a critical problem in delegation: ambiguity. Without a shared, standardized vocabulary, the scope of a delegation can be misinterpreted. A taxonomy ensures that when a Principal grants permission for an action like construction.validation.approve, all parties (the Principal, the Delegate, and the Mandaitor verification service) have the exact same understanding of what that action entails.
Integration
Taxonomies are published as individual NPM packages (e.g., @mandaitor/taxonomy-construction). The Mandaitor API loads these packages into its Taxonomy Registry at startup. When a mandate is created, its scope is validated against the definitions in the specified taxonomy, ensuring correctness and preventing invalid permissions.