Skip to content

How-to guides

Task-oriented recipes for people who already know the basics. Each answers a specific how do I…? Read the Concepts first if you want the model behind them; do the tutorial first if you have never loaded a scope.

The CLI

  • A tour of the dna CLI — every command group in five minutes, one executed example each, linked to the generated reference.
  • How to install bundles from a repository — dna install github:owner/repo[/subdir][@ref]: fetch, reader-driven detection, validation of untrusted manifests, conflicts, and the installed.lock provenance record.
  • Scaffolding a repo — dna solution new/update — generate a real app (package, Dockerfile, version floor, deployment wiring) from a Copier template, and roll it forward when the template moves: the four silent Copier behaviours the command reports, and the two wiring places no template can reach.

Authoring

  • How to author an agent — dna new scaffolding, named layouts (order the persona by name, no Mustache), and single-file souls.
  • How to add a Kind — ship a new Kind + Extension (or a record-style Kind as pure data) in about thirty minutes.

  • How to read instance data — the blessed query surface: the one documented way to read manifest data in either SDK.

  • Copilot kits — a complete flow installed as instances — dna copilot install <dir>: the Kind, the Agent and the Copilot (with its surfaces[] flow) written through the kernel door, with the human approval gate intact (--approve stamps the operator).

Adapters and formats

Spec-driven methodology

  • Spec Kit — the supported spec-driven flow — run GitHub Spec Kit untouched, then dna specify import to durably capture the run as DNA Kinds (and dna specify export to project them back byte-faithfully). DNA composes underneath Spec Kit, adding portability, memory, governance and tracking.

Search and memory

  • How to use semantic recall & memory — install the extras, search a scope with dna recall, drive the memory verbs, and register embedding/search providers programmatically.

Quality

  • How to evaluate agents — author EvalCases and EvalSuites, run them offline with dna eval run (the composed prompt is the default target), pin an EvalBaseline and gate CI on regressions.

DNA Cloud

  • The DNA Cloud starter catalog & BYO-DNA — the three _lib starter agents every tenant inherits, and how a Pro tenant authors + pushes their own agent as a tenant overlay (gated by PricingPlan caps).

Lifecycle

Operations