Data-model governance
Stops each squad's agents inventing incompatible vocabularies and unsafe data handling across the portfolio. The rules every data model follows.
The rules
- Naming conventions — consistent entity/field naming across products.
- Mandatory PII tagging — every field classified (public / internal / client-confidential / regulated-PII); nothing untagged.
- Data minimisation — collect only what's needed; justify sensitive fields.
- ACORD alignment — map to the standard where it applies, so products interoperate.
- Versioning & migration — schema changes are versioned; migrations are irreversible actions needing sign-off.
The design context pack
The design system encoded machine-readably — which components for which patterns, token-only styling (no raw hex/px), copy tone, worked good/bad examples — is loaded into every UI-producing agent session and enforced by lint. It's the highest-leverage artefact for UI consistency, and a fragment of the repo context & memory standard. (The pack is instantiated with the design-system foundation — a Design + Engineering task once the design-tooling ADR lands.)
Standards referenced: ACORD, UK GDPR, W3C DTCG, WAI-ARIA APG.