AGENT-DESIGN-STUDIO DIR. A · SPECIMEN SHEET
SHEET 01 / 01 — PROCESS SPECIFICATION
An agent skill · MIT · Seven steps · Two gates

Your agent can already write UI. This teaches it to design.

A seven-step process an AI coding agent runs end to end — intake, fidelity, wireframe, three real directions, a token system, motion, review. Every step leaves an artifact on disk. Two steps stop and wait for a person.

Fig. 1 — the seven steps
01 INTAKE

Subject, audience, and the one job the screen must do.

02 FIDELITY

How finished this needs to be — the first design decision.

03 WIREFRAME

Structure only. States and viewports declared here.

04 DIRECTION

Three real renders. A person chooses.

05 SYSTEM

Tokens derived, not picked. Then the build.

06 MOTION

Every animation earns its place or is refused.

07 REVIEW

Scored against a floor. Fixes verified.

00 MODE

Static artifact, or a screen inside a real codebase.

Fig. 2 — deliverables

Every run leaves a paper trail

00-brief.md
01-fidelity.md
02-wireframe.html + .png
03-directions/{a,b,c}.html + .png
03-direction-decision.md ← gate 1
04-design-system/tokens.json
    components.md · assets-manifest.md
05-implementation.md
06-motion-spec.md
07-uat-report.md ← gate 2
STATUS.md ← read this one

STATUS.md is the one page for everyone else.

Which step is done, which is waiting, who it is waiting on, and a link to every artifact. Written for someone who does not read code — so the person paying for the work can see where it stands without asking.