Kerno

Backend validation · Backend services · stage: backend-validation

Deterministic backend scenario re-runs with greybox database checks — no LLM in the loop.

Backend changes fail in ways unit tests don’t see: a handler that returns 200 but writes the wrong row, a migration that silently breaks a flow. The Kerno plugin adds a backend-validation stage that re-runs your committed Kerno scenario suite (.kerno/scenarios/) against the app running in the agent’s slot — deterministically, with no LLM in the loop.

Scenarios use the slot’s own database for greybox checks and report pass/fail with a full evidence trail, so an agent’s backend change is proven against real requests and real state.

What it installs

A backend-validation test stage registered in .har/stages.json.
A fail-fast lock that serializes backend validation across slots (Kerno runs one agent per machine); frontend stages still run concurrently.
An adaptation guide at .har/stages/KERNO.md.

Requirements

Kerno CLI (npm install -g @kerno/cli) and Docker
A Kerno agent bound to the slot’s worktree (kerno init)
A committed scenario suite — validation re-runs an existing suite, it does not generate one

Artifacts & evidence

Scenario results and the evidence trail land under .har/artifacts/backend-validation/.

Adapting it to your repo

The stage never starts or rebinds the Kerno agent — binding is a one-time setup per machine.
See .har/stages/KERNO.md for the full setup and adaptation guide.

Kerno Plugin install guide