MCKRUZ · Delivery Standard · v1.0
Agents write code faster than anyone can read it. That doesn't make delivery faster on its own. This standard moves the work to the two things that now decide the outcome: saying clearly what you want, and proving that what came back is right.
When code gets cheap, the hard part moves to two places: describing the work clearly enough to test, and checking the result before anyone trusts it.
So the work runs as one short loop, repeated for every piece of work from a one-line fix to a whole feature. Gated phases open and close the engagement; the loop fills the middle. No checking is just vibe coding — the loop closes it.
Decide what you want, then write it as a short spec you can test — "returns 401 for a bad token," not "make auth good."
The agent plans first — you approve before any code — then builds and self-checks inside the scope, context, and permissions you set.
Mechanical gates, a separate grader that didn't write the code, and a non-author human decide it's good — with a named sign-off on anything high-risk.
Discern just means working out whether the work is actually good before anyone trusts it. It's the half of the method most teams skip — and the reason this exists. The build loop, in full → · Or watch it happen on film →
Frame the problem and build the factory — repo, pipeline, and the rules agents work inside. Each phase ends at a gate a human signs.
The middle isn't phased. Every story runs Intent → Delegate → Discern and ships before the next starts — checking happens per change, never piled up for later.
Document the system, ship to production, and hand everything over — until the client can run the loop without us.
Want to go deeper? Each phase has a plain-language interactive walkthrough — the idea, the precise mechanics, and a complete worked example, all on one page.