Spruce.
The only AI design reasoning system that starts with users, not pixels.
Installs into the AI coding tools you already use — so the design is grounded in real people and the jobs they’re doing, not the average of training data.
Philosophy · A revision
The problem isn’t that AI can’t design. It’s that all AI designs the same way.
The cause is structural. AI models are trained on the same corpus of design examples, and inside that corpus, certain choices appear far more often than others. When a model generates new design, it reaches for those defaults reflexively — not because they’re right for your product, but because they’re statistically most likely.
We’ve cataloged the patterns we keep seeing. They’re everywhere because they’re always available.
Inter for everythingreplaced with A typeface for this productBlue and purple gradientsreplaced with Warm neutrals, one accentThree equal cardsreplaced with Asymmetric, editorial8px corners by reflexreplaced with Restrained, specific“Let’s get started!”replaced with Direct, no throat-clearingLinear easing, 300msreplaced with Custom curves, paced
Spruce is the alternative. It starts with users, not pixels.
Most AI design tools generate from prompts and fill in the decisions invisibly. Spruce starts upstream — with named personas, the jobs they’re hiring the product to do, the moments they encounter the design. Captured into context files every command reads from, so every decision can be tied to a real person doing a real thing.
From that grounding, Spruce composes the design reasoning that follows — typography, color, spacing, components, motion, voice. The fundamentals good design has always rested on, now calibrated to a specific product for a specific audience. And it gives you tactical control through a set of commands that let you direct, refine, or rethink any decision.
The grounding makes the reasoning specific; the reasoning runs in the background; the commands put you in the chair.
What you get back is design that happens to be made with AI rather than AI design.
The product gets to look like itself.
The catalog
Twenty-five commands across five tiers. Each one is a specific conversation with your AI tool — and a specific kind of impact on the project.
Browse by tier
Reasoning about who the design serves, before reasoning about how it should look. Optional but every other tier reads better with it in place.
Reasoning about who the design serves, before reasoning about how it should look. Optional but every other tier reads better with it in place.
- /personas
Establish who the design is for, not just what it should look like.
Maya
Quiet evaluator
- /jtbd
Name the jobs the personas hire the product to do.
When a quiet morning startsI want to settle before kids wakeSo I can begin the day grounded - /journey
Map a persona’s path through a real flow, with emotional arc and friction.
- /scenarios
Anchor design decisions in concrete moments, not abstract use cases.
“Maya opens the app at 5:43 a.m., still in bed.”
Maya · Morning grounding
- /audit
HCD-grounded evaluation against named personas and their jobs.
Affects Maya
- Onboarding skips Maya's job
- Voice misreads quiet audience
- Notifications too frequent
The loop
Commands run together as a loop on a foundation.
Set context
/spruce-up
Discover
/personas
Generate
/design
Refine
/typeface
Review
/critique
Install · One command
One command. Spruce installs into your AI coding tool and stays out of your way until you ask for it.
After installing
The workflow, in four moves. Context first, users next, design from there.
Set up your project.
Run /spruce up once. It walks through a short interview about your product, audience, and direction, then writes a context file that every subsequent command reads. Your design language travels with the project.
FAQ · Common questions
What visitors ask most. Answered directly.
Releases · Recent shipping
The fourth and fifth harnesses: Spruce now installs into VS Code (with GitHub Copilot) and Gemini CLI in addition to Claude Code, Cursor, and Codex CLI — completing the originally-planned multi-tool roadmap. The same `npx spruce-skill add` command auto-detects which AI coding tool the project uses and installs the right per-harness bundle. **Spruce works in all five major AI coding tools.**
- v0.4.0 — Codex CLI supportMay 6, 2026
The third harness: Spruce now works in Codex CLI in addition to Claude Code and Cursor. Codex picks up Spruce skills natively from the project's `.agents/skills/` directory — Codex's primary skill-loading path — and follows each skill's conversation pattern correctly. Verified end-to-end via real Codex CLI smoke tests.
- v0.3.0 — Cursor support May 4, 2026
The second harness: Spruce now installs into Cursor in addition to Claude Code. Same install command, same 25-command catalog, same slash invocation pattern (`/decide`, `/typeface`, `/personas`) — no mental-model translation between harnesses.
- v0.2.0 — Discovery tier May 3, 2026
The first major system addition: a new Discovery tier between Setup and Generative that grounds design work in named users and the jobs they're hiring the product to do.