Bezalel · Vitruvius · Rams · CRISP

Design Intelligence for AI Agents

CContextual
RResponsive
IIntelligent
SSeamless
PPowerful

AI has given every developer a design tool. It gave none of them a design standard.

Design has always needed a language. CRISP is that language — for the age of AI.

Start free audit → Install the skills

01 — Install in 30 seconds

One command.
Fourteen skills.

The installer auto-detects your AI harness and writes the skills into your project as ordinary files you own.

zsh — crisp-install
$ npx skills add @laith-wallace/crisp
# Auto-detects Cursor, Claude Code, Antigravity, VS Code, Gemini CLI ✓ Claude Code detected ✓ Installing 14 CRISP skills... ✓ Done. Run /crisp-teach to get started.

Then run /crisp-teach once per project. It learns your users, design system, and references, and writes .crisp.md — the context file every other command reads automatically. Your context is never lost between sessions.

SUPPORTED TOOLS
Claude Code
Cursor
Antigravity
VS Code Copilot
Gemini CLI
Codex CLI

The files land where your harness expects them: .claude/ for Claude Code, .cursor/ for Cursor, .agents/skills/ for Antigravity. Nothing updates behind your back.

02 — Why these skills exist

Agent-generated design fails
in four predictable ways.

You think your agent's design problem is prompt quality. It isn't. The agent has no way to judge its own output — so it ships the first thing that compiles. Every era solved this with a named standard. Ours hadn't, until now.

I — 1446 BC

It doesn't know who it's building for.

"I have filled him with wisdom, with understanding, and with knowledge, and with all kinds of skill."

Exodus 31:3 — on Bezalel, the first named designer

The oldest design standard in history is not about skill. Bezalel's third gift, da'at, is relational knowledge — knowing who the work serves and why. An agent dropped into a repository has none of it. So it builds for nobody in particular: empty states that say "No data available", headings that could belong to any product, one settings page for the first-week novice and the five-year power user alike.

The fix is to give the agent that knowledge once. /crisp-teach interviews you about your users, your design system, and what good looks like for your product, then writes .crisp.md so every later command starts oriented. /crisp-brief turns a vague request into a scoped brief with observable success criteria. /crisp-research surfaces the competitive patterns before you design.

II — 25 BC

Functional is not the same as good.

"Well building hath three conditions: firmness, commodity, and delight."

Vitruvius — De Architectura

Code that compiles is firmitas alone — one condition of three. Agents stop there. The interface works, and it also shows a spinner on every filter change, presents "1,247" with no comparison or consequence, and forces a portal redirect for a one-click approval. Technically correct. Experientially hollow.

The fix is evaluation with teeth. /crisp-audit scores all five dimensions with P0–P3 severity and a prioritised fix list. /crisp-review returns a grade and the top three issues in thirty seconds. /crisp-ux-laws and /crisp-a11y hold the work to the laws of cognition and WCAG 2.2 AA.

III — 1976

Agents add. They never subtract.

"Good design is as little design as possible. Less, but better."

Dieter Rams

Left alone, an agent answers every request by adding: another toggle, another card, another modal, all twenty-three settings visible to every user at all times. Rams built his ten principles on the opposite instinct — that the discipline is in what you remove.

The fix is subtraction as a command. /crisp-redesign modernises an existing UI without breaking what already works. /crisp-copy strips microcopy to the words that orient. /crisp-design-eng governs motion and polish so nothing moves without a reason.

IV — 2026

Nothing survives the session.

"Your AI agent can generate code. It can't evaluate design."

CRISP

Close the session and the context evaporates. The next session re-decides everything: a new spacing scale, a different empty-state pattern, a fourth shade of grey. Three thousand years of design standards were written down precisely so the knowledge would outlive the practitioner. Your agent's knowledge dies in hours.

The fix is a written record the agent maintains. .crisp.md persists your context across sessions and updates. /handoff turns a reviewed design into a specification another agent can build. /feature-design records the rationale behind every flow decision. /crisp-ai extends the standard to AI-native surfaces, and /crisp-funnel turns it into a build.

Bezalel · Vitruvius · Rams · CRISP — 3,472 years of the same question.

03 — The Framework

Five dimensions.
One clear standard.

CRISP gives your AI agent a shared design language. Each dimension maps to a specific failure pattern and a specific fix.

C
Contextual
Can the user tell where they are and what this page does within 5 seconds?
FAILS WHEN
Dead-end empty states · missing breadcrumbs · unclear orientation · generic headings
R
Responsive
Does the UI update immediately on every interaction?
FAILS WHEN
Spinners on every action · click-wait-update patterns · no hover feedback · delayed confirms
I
Intelligent
Are we showing insights, not just raw data?
FAILS WHEN
Numbers without meaning · blank forms · no next-best-action · we know the user but pretend we don't
S
Seamless
Are we fitting into their day — not forcing them into ours?
FAILS WHEN
Unnecessary logins · portal redirects · custom components that break habits · friction on routine tasks
P
Powerful
Is complexity hidden appropriately for each user type?
FAILS WHEN
Expert settings shown to novices · no keyboard shortcuts · no undo · all options always visible

04 — Before & After

What changes when
your agent understands design

AI agents produce technically functional interfaces. CRISP evaluates whether they're actually good — and why they aren't.

WITHOUT CRISP
  • CEmpty state says "No data available" — user doesn't know why or what to do
  • RSpinner appears on every filter change — user waits while nothing seems to happen
  • IRaw metric: "1,247" — no context, no comparison, no suggested action
  • S"Open in portal" — forces a redirect and new login for a simple approval
  • PAll 23 settings visible to every user at all times — cognitive overload
WITH CRISP
  • C"Your Q3 pipeline looks healthy — 12 deals active. Add your first deal to get started."
  • ROptimistic update — result appears instantly, background sync happens silently
  • I"1,247 leads — up 23% this month. 3 deals need your attention before Friday."
  • SInline approval card — one click, no redirect, no re-authentication required
  • P3 most-used actions surfaced. Everything else collapsed behind "Advanced settings"

BENCHMARKED AGAINST Stripe · Linear · Notion · Asana · Slack

05 — The Skills

Fourteen skills.
One standard.

Every skill evaluates against the same five dimensions and reads your project context from .crisp.md. Every skill has its own page.

06 — Try it free

No CLI. No install.
Just paste your URL.

Get a full CRISP audit of your product in seconds — free. No account required.

Start free audit →

07 — The CRISP Letter

Design evaluation,
in writing.

Occasional letters on making AI agents produce work worth shipping. New skills announced here first. No noise.

No spam. Unsubscribe any time.

08 — FAQ

Common questions

How is CRISP different from running /audit in Impeccable?
Impeccable improves how your AI generates frontend code — typography choices, layout patterns, visual polish. CRISP evaluates whether the design is actually good — whether a user knows where they are, whether the interface feels instant, whether it surfaces insight instead of data. Different layer entirely. They're complementary, not competing.
Do I need to run /crisp-teach before every session?
No. Run it once per project. It writes .crisp.md to your project root — every CRISP command reads from it automatically in subsequent sessions. Your context is preserved across updates too; .crisp.md is never overwritten when you update the skills pack.
Who is CRISP for?
Product designers, design leads, and engineers who use AI agents to design and ship software. If you've ever received a "functional but flat" design output from an AI — technically correct, but experientially hollow — CRISP is the missing evaluation layer.
What does CRISP actually stand for?
Contextual, Responsive, Intelligent, Seamless, Powerful. Five dimensions of world-class product design, each with a specific test question and a specific set of failure patterns. The framework is benchmarked against Stripe, Linear, Notion, Asana, and Slack.
Can I use CRISP without /crisp-teach?
Yes. /crisp-audit and /crisp-review work on any design screenshot or description, without project context. /crisp-teach just makes them significantly more accurate — the agent knows your users, your design system, and what "good" looks like specifically for your product.
Where do the files go when I install?
npx skills add @laith-wallace/crisp auto-detects your AI harness and places files in the correct directory: .claude/ for Claude Code, .cursor/ for Cursor, .agents/skills/ for Antigravity, and so on. If you prefer, download the universal ZIP and extract to your project root manually.

Start with an audit.
Stay for the standard.

Run a free audit on your product right now — or install CRISP skills and give your AI agent a design standard.

Start free audit → Install the skills