Skip to content

Concepts

Understanding-oriented explanation — the why behind DNA. Read these when you want the mental model, not a task recipe.

Start with the thesis; the rest expand its pieces.

  • The thesis — CRDs, but for agentic behavior — the primer. A short normative core (RFC-2119) plus the teaching that anchors DNA to the Kubernetes CRD model: the owner names the schema, spec is authored intent, behavior is derived.
  • Kinds — identity and composition — how (apiVersion, kind), dep_filters and prompt templates turn cross-references into a single composed prompt.
  • The built-in Kinds — a catalog — the commented catalog of the composition-plane Kinds beyond the core prompt family: collaboration, safety & governance, preferences, testkit.
  • The microkernel and its five ports — the closed core that knows no Kinds, and the ports extensions plug into.
  • Market fidelity — how "consume standards byte-faithful under their owner's namespace" is enforced against real marketplace bundles.
  • Tenancy and layers — scopes, the orthogonal tenant dimension, and layer overlays.
  • Search & memory — semantic recall and agent memory as two kernel ports with pluggable adapters: offline-first sqlite-vec + FTS5 + RRF, pgvector for scale, and memory as the Kinds you already have.
  • The reference graph — what points at this instance? spec.relations declarations become stored edges the write path produces while it validates them, walked by one recursive CTE — plus the four signals that keep the answer honest about its own gaps.
  • Agent-facing knowledge — why DNA represents knowledge as curated, cited Kinds rather than generated wiki prose.
  • Composing with GitHub Spec Kit — the layering model: DNA names Spec Kit the supported spec-driven flow and sits beneath it, adding portability, memory, governance and tracking without changing how Spec Kit runs. The full decision is the Spec Kit adoption ADR.

For the procedures that put these ideas to work, see the How-to guides.