Agent-Native Spine is the architecture I use across my projects. People and agents use the same product operations through different interfaces.
I define each product action once. The app, HTTP API, CLI and MCP tools call the same implementation, with shared validation and access rules.
Architecture across my projects
The common architecture behind my projects: an operation catalog, shared product logic, caller permissions and explicit checks before startup.
Why it matters
Agents need a clear model of what a product can do. Shared operations keep the product rules consistent as interfaces and features change.
What I built
- Operations before interfaces. Each product action has an identity, input and output contracts, an owner and a scope. Interfaces connect to that operation model.
- One implementation of the product rules. A shared core holds the product behaviour. Thin adapters connect the app, HTTP, CLI and MCP to that core.
- Permissions for each caller. People and agents use the same access checks. Each caller has its own identity, capabilities and permitted scope.
- Checks before startup. Each host connects checks for missing handlers, operation ownership, permissions and scope coverage. A failed check can stop startup.
- A structure I reuse. Each project supplies its own product logic, storage and adapters. The operation model and checks remain consistent across projects.
Where I use it
- ShappyOS. Tasks, Bots and Chat use shared operation contracts, caller permissions and scope checks. Product changes run through Dromio workflows.
- This portfolio. The content editor, HTTP API, CLI and MCP tools call the same content service. Saves validate the input and reject stale revisions.
