S Shipwright

Launch QA for vibe-coded projects

Before you ship the AI-built app, make it survive first contact.

Paste a GitHub repo or demo URL. Shipwright generates a launch-readiness report in seconds — browser health, README friction, fake-complete features, and the next fixes that matter.

Try a sample:
Read the guide · View on GitHub
10
reusable skills
10 min
first report target
0
fake green checks

Choose the job

Four doors. One clean launch path.

Shipwright is no longer just a pretty audit page. Pick the work you need now: shape an idea, inspect a project, install reusable skills, or run the full agent workflow.

01 Plan from a rough idea

Let the AI Planner ask follow-up questions and turn a vague website concept into a buildable plan.

Start with an idea 02 Audit a repo or demo

Paste a GitHub URL or live demo and get a launch-readiness report with the next patch.

Run launch QA 03 Install the workflow

Use the open-source skills inside Codex or Claude Code for evidence-backed local checks.

Browse skills 04 Ship with a sequence

Move from agent claims to browser proof, install proof, release copy, and public launch.

See workflow

The pain

AI agents are great at producing finished-looking work.

01

The README lies by omission.

It says "npm install" but forgets env vars, version constraints, and what success looks like.

02

The browser was never opened.

Console errors, broken routes, missing assets, and mobile overflow hide behind a clean diff.

03

The launch page has no sharp promise.

Strangers need a quickstart, proof, limitations, and GitHub metadata before they trust it.

Try it now

Generate a launch QA report.

Paste any GitHub repo URL or owner/repo. Doctor calls the GitHub public API directly from your browser, analyzes the README, and produces a fix plan you can paste into Claude Code or Codex. Nothing is sent to any server.

Try a sample:
Launch channel

Detecting mode… Enhanced mode runs richer GitHub checks via the Doctor backend (CI/CD, SECURITY, package.json, URL probe). If the backend is unreachable from your network it falls back to in-browser checks automatically.

Paste a repo URL and click Analyze to generate your launch QA report.

What does Shipwright actually check?

Agent output review

  • Hallucinated feature claims
  • Fake buttons and dead links
  • TODO / FIXME / placeholder strings
  • Unverified integration references

Browser launch audit

  • Console errors and warnings
  • Network 4xx/5xx failures
  • Interactive element functionality
  • Mobile layout at 390px and 768px

README install audit

  • Missing env vars or API keys
  • Version constraint gaps
  • Expected output not shown
  • First-run path ambiguity

GitHub release checklist

  • Repo topics and description
  • Release notes completeness
  • Launch post copy quality
  • Contributor-facing next issues

Usage guide

Use Shipwright in two modes.

The website is the product demo and report builder. The installed skills are the real agent workflow for evidence-backed browser, README, and release audits.

01

Paste the thing you want to ship.

Use a GitHub repo, owner/repo, a localhost URL, or a public demo URL. Shipwright is built for web apps, skill packs, MCP servers, CLIs, and templates.

02

Generate the launch risk report.

Choose the project type and launch channels. The demo report shows the exact categories Shipwright checks: fake-complete work, browser health, install path, and GitHub packaging.

03

Turn findings into the next patch.

Copy or download the Markdown report, then give it to Codex or Claude Code as the fix plan before posting the project publicly.

Who is this for

Built for builders who ship fast and want it to actually work.

Vibe coders

You built it with Claude, Cursor, or Copilot in a weekend. Before you tweet the link, make sure it survives a stranger clicking "Get started."

Indie hackers

You're launching on Product Hunt next week. Shipwright catches the broken mobile layout and missing env vars before your first users do.

Open source maintainers

Your README says "easy setup" but a new contributor gives up at step 3. Shipwright walks the install path like a stranger and reports where it breaks.

AI skill / MCP authors

Your skill pack or MCP server looks done in the editor. Shipwright checks trigger descriptions, install paths, and security boundaries before publish.

Inspection radar

Plot the launch route before the public countdown.

A quiet control-room view of the checks Shipwright runs: each waypoint maps to a launch risk, the evidence it needs, and the fix signal a builder should trust.

Launch Map 04 Signal: evidence required
Evidence

Every green light needs an observed browser state, command output, or file reference.

Risk

Markers call out launch blockers before they become public support requests.

Patch

The route ends with the smallest next fix, not a vague quality score.

How it becomes real

From static demo to agent-powered launch gate.

1

Agent output review

Catch hallucinated claims, fake buttons, TODOs, and unverified integrations.

2

Browser launch audit

Open the app, check console/network health, interactions, mobile layout, and trust gaps.

3

README install audit

Follow the public install path like a stranger and record exactly where adoption breaks.

4

GitHub release checklist

Generate topics, description, release notes, launch post, and next contributor issues.

Honest status

What this demo does and does not do.

This demo does

  • Generate a structured launch QA report based on project type
  • Show the exact audit categories the real skills check
  • Produce a copyable/downloadable Markdown report for your agent
  • Offer AI-powered website planning (BYOK — bring your own API key)

This demo does not

  • Clone your GitHub repository or read its files
  • Open a real browser to check console errors or mobile layout
  • Store your API key on any server (key stays in your browser only)
  • Make requests without your explicit API key configuration

For real, evidence-backed audits, install the open-source skills and run them locally with Claude Code or Codex.

All 10 skills

From trend discovery to launch.

Discovery

github-radar

Scan GitHub Trending and separate tools, opportunities, and hype traps.

Discovery

trend-to-product

Turn a hot repo or trend into a differentiated product opportunity.

Planning

idea-to-prd

Convert a rough idea into a lean PRD with scope and acceptance criteria.

Planning

prd-to-issues

Break a PRD into GitHub-ready issues ordered by delivery sequence.

QA

launch-readiness

Audit a project for README, install, demo, trust, and conversion gaps.

QA

browser-launch-audit

Verify a web app in a real browser before shipping.

QA

readme-install-audit

Test if a first-time user can install from the README alone.

QA

agent-output-review

Catch hallucinated or fake-complete AI-generated work.

Launch

github-release-checklist

Package a repo for public release with metadata and launch copy.

Meta

workflow-to-skill

Turn a repeated workflow into a clean, installable skill.

AI-powered planning

Start messy. Leave with a build brief.

Tell the AI Planner your website idea. It asks targeted questions about users, features, proof, stack, and visual direction, then turns the answers into a complete website creation plan.

BYOK — key stays in localStorage and only calls your chosen API endpoint.
Ask Clarify the real user

Audience, pain, proof, constraints, and what the first visit must accomplish.

Shape Turn answers into scope

MVP features, page map, content blocks, design direction, and launch risks.

Review Send it through Shipwright

Use the generated brief as the input, then audit the finished site before sharing.

Open source first

Ship fewer beautiful half-products.

Start with the skills today. Turn this website into the hosted version when the workflow proves people keep coming back before every launch.

View on GitHub Try the demo

AI Planner Settings

Your API key is stored in your browser only (localStorage). It is never sent to any server other than the API endpoint you specify above.