Documentation menu
Concepts

How Skaftor works

The end-to-end architecture: how intent, blueprints, work orders, agents, and code connect through the engineering memory graph and governance gates.

Skaftor's architecture is a governed pipeline with a knowledge graph running through it. Every artifact is a node; every relationship is a real, queryable edge.

Intent → blueprint → work orders → code

Business intent (requirements) and technical intent (architecture blueprints) are captured as structured artifacts. Approved intent compiles into work orders — the agent-executable unit of work. Agents execute work orders and the resulting code links back to the intent it satisfies.

The engineering memory graph

Every cross-entity link — requirement→blueprint, work order→requirement, feedback→work order, code→requirement — is written as a graph edge. Because the edges are real, the graph answers 'what depends on this?' and 'why does this exist?' by traversal, and derives signals like blast radius and traceability.

Governance

Approval gates block progression until their condition is met, producing an audit trail. Automated checks — blast radius, spec drift, and release confidence — run continuously so agent velocity never outruns governance.

The knowledge graph is what makes agent orchestration safe: an agent picking up a work order inherits the same context a senior engineer would.

Related