Swamp.club is commonly perceived as a Kubernetes/infrastructure automation tool, but its core primitives — Zod-validated models, immutable versioned data, method execution, and composable DAGs — are domain-agnostic. Using the @dougschaefer/writing-voice extension as a case study, the post demonstrates how the same architecture that structures pod health data can store an organization's editorial voice profile, complete with tone tiers, prose rules, anti-patterns, and kill lists. Agents load this typed, versioned context before generating content, producing consistent on-brand output instead of generic AI prose. The broader thesis is that any unstructured context an agent needs — incident playbooks, hiring rubrics, product positioning — can be expressed as a schema and managed in the same system as infrastructure state.
