Pick a prompt to watch Alfred work, sped up. What it made shows on the devices.
The design system has two command lines. Alfred answers while the work happens, about one thing at a time. Concierge inspects the work afterward, across every run. Alfred builds. Concierge inspects. You decide.
Both tools are private. They run inside the studio’s repo, so this page shows how they work, with real output, and not something to install.
Answers from the design system while you work, one question at a time.
Alfred is the command line for the design system. It reads the contracts (each component’s written rules), usage rules, vocabulary, brand packs (one brand’s small set of overrides) and tokens, and runs the checks that already exist. It keeps no copy of any of them, so its answer is always the one the system gives today.
A design system fails when its rules live in a document nobody opens. Alfred puts them where the work happens, so a designer, an engineer and a coding agent ask the same question and get the same answer.
A need in plain words, and exactly what Alfred printed. It names the pattern the need belongs to, then the component to use and the one not to.
The prompts in the replay above are three real runs of Alfred building on the system. These are the numbers each run produced.
These are single runs. The time to fix is a coding agent’s, not a person’s. The first check on the wallet page was clean because the agent had the system’s rules in front of it, so that run shows no failure.
A prototype sits outside the enforced checks, so its clean result is labeled advisory, not a pass. Each device shows a picture of a real render, not a screen to use.
Alfred’s suggestions are leads, not verdicts: on a sign-up flow its top three named the right component for 7 of 18 elements, and the history filled in the rest. Of the foundation audits, only space is enforced in CI today.
Alfred’s check can report another contract’s failure as the checked component’s, so trust Concierge’s components table until that is fixed.
Inspects the agents’ work after the fact, across every run.
Concierge is the checker for agentic design-system work. The agents that work in the design system, compose, the art department and design-dna, each write a record of the run: the brief, what they used, what they checked and what they skipped. Concierge reads those records and checks them itself, so no run grades its own work.
It checks again later, because a record describes the system as it was on the day it was written. It lives apart from the agents on purpose: a checker that ships in the same repo as the agents it checks changes in the same commits as them.
Compose records name where a screen lives, not the file, so output freshness only works for art department runs. A compose record may leave out its brand pack, so its pack cannot be checked. Only design-dna runs carry a cost and a time.
Concierge does not yet read GitHub’s check results, and its browser check does not cover the art department’s runs.
Home About Using AI Ask the system Raptive Onboarding Nextdoor Faves Nextdoor Business Design system Trulia Synthesizer Alfred CLI Resume