|

Every vertical — telecom, water, electric, defense, logistics — looks different on the surface but shares the same structural pattern: entities connected by relationships, constrained by physics, optimized for cost. The agent skill system captures this pattern formally. It is the type system that makes Nexma domain-agnostic, and it is the reason we can enter new markets without writing new code.

What an agent skill defines

An agent skill defines the entities in a domain, their typed properties and spatial geometry, and the relationships between them. It defines the constraints that govern validity — capacity limits, separation distances, operational thresholds. It defines computed expressions derived from other values. And it defines how all of this renders and how the operator interacts with it. Everything the user sees and everything the agent reasons about flows from the agent skill.

Why this architecture

The agent skill system captures the structural commonality across domains and makes it programmable. The platform reads the agent skill and generates everything the operator needs, including the agent’s domain knowledge, the solver’s constraint model, and the map’s rendering rules. This is not abstraction for its own sake. It is the core architectural decision that allows Nexma to serve defense, telecom, utilities, construction, and logistics from a single codebase. Without it, we would need to build and maintain a separate product for each domain. With it, we add domains by adding definitions.

Agent skill as product

Adding a new vertical to the platform does not require code. It requires an agent skill. Define the entities, their relationships, and their constraints. The platform handles the rest. This is how we go from one domain to dozens without proportionally scaling the engineering team. The agent skill is the product. The platform is the engine. And the separation of those two concerns is what makes the entire system scale.