For developers

Start with a screen audit.

From your app directory, run the free CLI. It captures the screen at desktop and phone widths and writes a report you can inspect beside the code.

Read a sample audit

Terminal App directory

$ npx unvibed audit

AUDIT_REPORT.md: Screen grades and evidence
.unvibed/audits/: Captures and the HTML report

Give your agent a design contract.

DESIGN.md
tokens.css
CLAUDE.md

Choose a direction in the Playbook. Save its design contract beside your package.json, import its tokens and append the project rule to CLAUDE.md or AGENTS.md. The next session reads the same decisions.

Explore the Playbook

Give the next session a starting point.

BACKLOG.md
AGENTS.md

Shipbook pairs an acceptance check in the backlog with instructions for the agent that verifies it. Choose your tools and download the starter, including a first-week plan and a worked handoff.

Explore Shipbook

Review a pull request.

Pull request
Before and after
DESIGN.md

For a Rescue, name three screens and the tasks that must still work. A designer reviews the rebuild. You get a pull request against your repo, captures at both widths and the contract to keep using after you merge.

See the Rescue intake

What we measured.

We tested an instruction kit on ten apps. With the same render budget, the plain agent won six apps, the kit won one and three were unresolved. A separate test of the strong agent alone against its full repair loop split four wins each, with two unresolved. The strong agent alone beat the smaller model using the loop on seven of ten apps.

That is why we sell the contracts you can inspect and a designer’s review of the changes.

Read the ballots and action checks