1DesignToolDownload

Steering · since v1.0.0

A design kit the agent has to pass

The gate is the difference between 'generated' and 'shipped': kit checks run on every turn, and in Reviewed or Council mode independent critics judge the frozen evidence before the result counts.

DownloadRead the docs

Why it matters

Every AI tool can produce output; almost none can say no to it. The gate is the studio's answer — a set of automated checks the kit defines per direction (spacing on scale, contrast, type rules, the tells that make AI work look AI) plus a review tier where critics evaluate the frozen capture. A failed turn shows what failed and can repair — a green check can't quietly close a bad build.

How it works

  1. 1Every turn gets checkedAfter the agent writes, kit scripts verify the files against the direction's rules — deterministic checks, not a vibe.
  2. 2Pick your review depthFast is build → checks → done. Reviewed adds one independent critic and a repair pass. Council runs Visual, UX, Copy and Rights critics in parallel on the frozen capture.
  3. 3Failures get namedA failed check reports the rule id (tells.spacing-off-scale), not a shrug — and a repair pass is bounded, so loops end.
  4. 4Compare widens the checkWith Compare on, an edit re-renders at a second device or theme — a variant that fails rolls the whole edit back and names itself.

Capabilities

  • Deterministic kit checks on every turn — spacing, type, contrast, the AI tells
  • Three quality modes: Fast, Reviewed, Council
  • Council: four critics (Visual, UX, Copy, Rights) review one frozen capture in parallel
  • Stable rule ids — findings name the check that fired, and the auto-fix prompt uses it
  • Bounded repair passes: a failing build can't loop forever
  • Rollback on variant failure — the bad render names itself as the cause

Without it

Hope the model's taste held, open the result, squint, paste the screenshot back into chat asking 'is this good?'.

With 1DesignTool

The gate answers before you look: checks with names, critics with disciplines, a verdict you can read.

FAQ

What do the checks actually verify?

Kit rules per direction — spacing on scale, type hierarchy, contrast, plus the known AI tells (generic gradients, cliché layouts). Each finding carries a stable rule id.

What's the difference between Reviewed and Council?

Reviewed sends the frozen evidence to one independent critic with up to one repair pass. Council runs four discipline critics — Visual, UX, Copy, Rights — on the same capture, aggregates their verdicts, then a bounded repair. Council costs more tokens and takes longer.

Can I turn the gate off?

Fast mode is the light setting — automated checks still run, but no critic reviews. The kit checks themselves are what 'done' means; skipping them isn't a mode.

Does the gate work on my real codebase?

Codebase projects verify differently — scoped to the diff (repo_diff + tells on changed files), since your repo's conventions outrank a fresh project's gate.

In the docs

Related features

Design systemsAI design system generatorPick one of the 17 built-in directions — Swiss grid to Night console — or save your own. Either way the design kit compiles it into the rules, tokens and checks the agent designs under.Read more →SteeringVariants and blind compareAsk for variants and each option is planned as its own direction, built independently, and shown unnamed — you pick with your eyes, not your biases.Read more →AgentsYour coding agent is the designerClaude Code, Codex, Cursor Agent, Gemini, Devin and eleven more — whichever CLI you trust, 1DesignTool drives it as a designer and holds its output to the same visual bar.Read more →

Design with the agent you already have.

Free to try, buy once — the studio and your own coding CLIs do the rest.

Download