Skip to content
Back to skills

Design

ASecurity

Settles what an interface should look like before it is built - captures the current screens, writes design notes, makes three options per round (images on the real screenshots when an image tool exists, otherwise HTML/CSS rebuilt from the product's real components, styles, icons and fonts), critiques each against design-quality.md and the current screens, relays them for a pick, and records the approved design in mockups/. Never edits product source. Used by the bigbrain designer; the verifi...

  • 9 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 10, 2026
ai-agentsgo

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned November 10, 2026

npx -y skills add greenfield-inc/agent-farm --skill design --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Design?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Design
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/greenfield-inc-design/badge)](https://www.skillsdirectory.com/skills/greenfield-inc-design)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: design
description: Settles what an interface should look like before it is built - captures the current screens, writes design notes, makes three options per round (images on the real screenshots when an image tool exists, otherwise HTML/CSS rebuilt from the product's real components, styles, icons and fonts), critiques each against design-quality.md and the current screens, relays them for a pick, and records the approved design in mockups/. Never edits product source. Used by the bigbrain designer; the verifier reads its design-quality reference.
---

# Design

The approved design is the reference the build is checked against: the verifier compares built screens with it, screen by screen. Your starting message names the bundle, the screens and states, what must change, what must stay, and your status file. You reach the person only through `needs_you` (`contract.md`).

## 1. Capture

Run the dev build (with `verify-app`'s setup rules: reuse what is running, start only what is missing, test data only) and screenshot each affected screen in every relevant state (default, empty, loading, error, long content), theme and viewport the work touches. Save them under `mockups/current/` named `<screen>-<state>-<viewport>-<theme>.png`. If a screen cannot be reached, report what is missing as a `needs_you` instead of inventing it.

## 2. Design notes

Write `mockups/design-notes.html` from the `documents` skill's `design-notes` template, once per product area, then reuse and extend it: the screens captured; grid, max widths, breakpoints and gutters; the type scale and where each size is used; tokens from code when they exist (primitive → semantic); colour pairs for light and dark with their contrast; the components to reuse, with source paths; density and copy voice; do and don't rules seen in the product, each with a screenshot. Read the source for tokens and components; do not guess them from pixels when the code says.

## 3. Options, three per round

This is greenfield's `ui-mockup` method. Pick the method that fits, and say which one you used on the gallery:

- **Images**, for exploring looks quickly or changing an existing screen, when you have an image generation or editing tool: edit the real screenshots from step 1.
- **HTML and CSS**, when exact text, real data, responsive behaviour or a later pixel comparison matters, and whenever you have no image tool. A missing image tool is not a blocker. Rebuild the screen from the product's own source, not from memory or by eye: its real components and markup, its styles (CSS, SCSS, tokens), its icon assets (inline the real SVGs) and its fonts, so layout, type, spacing, colours and density match the current screenshots. Write it by hand or copy from source; do not generate approximate chrome with a script.

For round N, make three genuinely different options (not three shades of one idea), labelled 1 to 3. Save them as `mockups/<screen>-v<N>-<option>.png|html`.

- Change only what is in scope. Keep layout, type, density, icons and unaffected regions as they are.
- Use the product's own tokens and components from the design notes.
- Look at every result. Check its text and look for unintended changes.
- **Fidelity.** Put each option beside the current screenshot of the same screen and state, and compare every region the work does not change (app chrome, logo, navigation and sidebar, header, icons, fonts) region by region. A placeholder logo, an approximated or "simplified" icon, a substitute font, or chrome that differs from the screenshot is a defect: fix it from the real asset before showing the option.
- Critique each option against [references/design-quality.md](references/design-quality.md): the principles, and the jank items that apply to a still image (overflow, alignment, spacing, contrast, touch targets, dark mode, phone width, missing states). Fix what you can before showing it.
- List each option's invented details (text, data, icons nobody asked for); they are illustrative, not requirements.
- Ask, through `needs_you`, before using any paid fallback.

Write or update `mockups/index.html` from the `mockups` template: the brief, the method used (images or HTML/CSS rebuilt from which components and styles), the current screenshots, this round's options with a one-line approach, the self-critique (fidelity and design quality) and the invented details for each.

## 4. Relay the pick

Write a `needs_you` of kind `choice`: one question "Which option for <screen>?", the options with your recommendation first and why, and the three files (plus the current screenshot) in `files` so the orchestrator opens them beside the person. Keep working on the next screen while you wait. A favourite with changes starts round N+1 from that option, keeping earlier rounds on the gallery. Only an explicit approval selects the final design.

## 5. Record the approved design

- The approved design is what the build is checked against, so make it exact. If it is an image with invented details, either rebuild it in self-contained HTML/CSS from the real components and styles (`mockups/<screen>-v<N>-<option>.html`) so it can be opened, resized and compared, or list which details stay illustrative.
- On the gallery: the approved version, "approved via Q-n → D-n", the **must-match** list (screens and states the build is checked against, each naming the J-n that will check it), annotations only where tokens and components do not already say it, the illustrative details, and the rejected options marked rejected. Status `ready-for-build`.
- Run `check-bundle`, set your status to `done` with the approved paths, and report. The planner cites them in the cover sheet and in `control-flow.json`'s `qa.mockups`; mid-build, the orchestrator passes them to the implementer.

## Never

- Edit product source to make a mockup.
- Present a favourite as approved.
- Show an option with a placeholder logo, approximated icons or chrome that does not match the current screenshot.
- Put a question to the person in your own tab; every question is a `needs_you`.

Files in this skill

  • SKILL.md5.9 KB
  • references/design-quality.md4.5 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.