When I'm shipping an app, I live in two places: designs in Daisy, code in Claude Code. The work is fine. The handoff is the slow part — copy a screen, drop it into the agent, ask for the build, jump back to fix a color, repeat.

We just shipped something that kills the handoff. It's called agent-skills: a single SKILL.md your coding agent loads that teaches it to run Daisy for you. You stay in the terminal. Screens materialize on your canvas, and the agent saves rendered PNGs of each one right next to your code.

What you're installing

A skill is a Markdown file that teaches your agent how to do one specific thing. This one is the agent's working manual for the Daisy API: create projects, fire generation runs, tune themes, read back each screen's rendered HTML, and render screens to images. It works with Claude Code, Cursor, Codex, and any runtime that reads skills.

After you install it, you just ask:

you → your agent

Design onboarding for a meditation app: welcome, breathing-style picker, and notifications opt-in

Daisy plans the flow, generates the screens in parallel, and your agent hands you the results:

what you get back

✓ Welcome                 → ./welcome.png
✓ Breathing-style picker  → ./breathing-picker.png
✓ Notifications opt-in    → ./notifications.png

Those are real renders of the screens now sitting on your canvas. Drop them into a PR, a chat, wherever the review happens.

Setup

Two steps in your terminal:

terminal

npx skills add daisy-now/agent-skills
export DAISY_API_KEY="dsy_live_..."

Mint the key in Settings → API keys — the API needs a Pro or Max plan. Put the export in your shell profile so the agent always finds it. In Claude Code the skill loads on the next launch; run /skills and look for design-mobile-apps-with-daisy.

Keys are scoped, so you decide how much rope the agent gets. A full_access key lets it design. A read_only key lets it inspect your projects without ever spending a credit.

What you can ask for

The skill covers the whole loop, not just generation:

"Set a darker theme for this project and regenerate the home feed." Themes live on the project, so set one before generating — every screen inherits it and the app comes out looking like one app.
"Add a profile screen to the Runr project. Avatar, stats grid, and an edit button." It finds the project you already have. No starting over.
"On the settings screen, change the primary button to red and make the destructive action stand out more." Edits land on the existing screen, in place — not a regenerated copy.
"What did I create on Daisy this week? Render each screen to a PNG."
"Cancel the run that's been going for 8 minutes."

Runs are async by default: the agent fires one off, keeps working, and polls until Daisy finishes. It also reports the credits each run actually charged, so there's no guessing what a design session cost — and reading screens or rendering images is free.

Why this matters

Daisy exists to delete the boring part of design: producing mockups by hand. But I was still the delivery person, carrying screens between tools. That's a job for the agent.

With the skill installed, the agent is the designer. I describe the feature once. It plans the screens, generates them in Daisy, and saves the renders next to my code. I commit them with the PR. Done.

What it can't replace is taste — someone still has to call what's good enough. That part stays yours. The courier work doesn't.

Try it

If you're already building with Claude Code, Cursor, or Codex — and you're on a Pro or Max plan — install it and ask for something:

npx skills add daisy-now/agent-skills

The repo is at github.com/daisy-now/agent-skills — issues, questions, and contributions all live there.