Ship and Test with Agents

A ready-to-fork project template for building software with an AI coding assistant. It gives the assistant written rules, specialist helpers, automatic safety checks, and a test setup, so its work comes back checked instead of only claimed done.

By Stylus Nexus

Fork the Full Starter

4 agents. 13 hooks. Playwright E2E. CI workflows. Skills via agent-plugins.
Get started in 5 minutes ↓

Agents

Hooks

CI Workflows

Tool Setup Docs

What you get

Everything a production team uses to ship software with AI coding agents, in one fork-ready repo.

Development Infrastructure

AGENTS.md — project brain with rules, commands, and earned gotchas, read by every tool
Domain guidances — AI safety, architectural decisions, database patterns, session hygiene, shared primitives, testing strategy, unattended agents (loaded on demand, not always)
Specialized agents — backend, docs, test review, exploratory QA, and two domain templates
Agent memory — empty scaffold directories, ready to fill in as you work
Skills — pulled live from the agent-plugins marketplace instead of a fixed bundle that goes stale
Automated hooks — context injection, test guards, coverage checks

Testing Infrastructure

AI mock fixtures — deterministic test data instead of expensive live AI calls
Auth and profile mocking — test every subscription tier without real accounts
Visual verification — screenshot baselines catch what assertions miss
Verify loop — build + test + lint before agents claim "done"
Experiment-as-test — quality regression detection with baseline comparison
Tiered CI — smoke on PR, regression nightly, visual on UI changes (all manual-trigger by default)

These patterns come from a team that ships a production SaaS with AI agents every day. Not theory — patterns extracted from real failures, real fixes, and real workflow evolution over months of daily use.

Curious to know more?

Full Starter and agent-plugins are the two living pieces. (The two starters this repo replaced — agent-starter and test-starter — are archived.)

Marketplace

Agent Plugins

The review-merge-ship loop, evidence-before-done verification, and dozens more skills, installable directly — the source this starter's own skill recommendations point to. Works in Claude Code, Codex, and ~70 other agents.

Is this for you?

Starting a new web app or API in Node/TypeScript? Begin with the rules, checks, and tests already in place.
Adding structure to a project you already have? The setup runbook scans your project first and fills in the rules.
A small team sharing one AI workflow? Everyone's assistant follows the same written rules.

What it is not for

Not an app, framework, or library. There is no product code in it.
Not a replacement for human review. The checks catch common mistakes; they don't prove code is correct or safe. The security scan is a report-only baseline.
Not a compliance guarantee. Setup writes your legal and privacy limits into AGENTS.md. Enforcing them is up to you.
Not built for non-web or non-Node projects as it stands. The tests drive a web app over HTTP, and the hooks use npm and jq. We haven't tested the Python, Rust, or Go paths.
Not a managed service. Your fork is yours. Updates from us don't arrive on their own.

Get started in 5 minutes

Fork and install. Fork stylusnexus/full-starter on GitHub, clone your fork, and run npm install. You need Node (the version in .nvmrc) and jq.
Open the clone in your AI tool (Claude Code, Codex, Cursor, or another).
Paste the one-line command for your tool, shown below.
Answer the questions it asks. “Not sure” is a fine answer. It writes your AGENTS.md and fills in e2e/app-contract.ts.
Run ./scripts/verify.sh. Until your app has a dev script, expect 1 passed, 1 skipped. The security scan's dependency audit needs internet. The first browser test run also needs npx playwright install chromium.

Works with your AI tool

The full starter includes setup docs for each tool. Point your assistant at the right file.

Claude Code

CLAUDE-SETUP.md

Paste: Read CLAUDE-SETUP.md and set up my project.

Marketplace, hooks, agents, memory

Codex

CODEX-SETUP.md

Paste: Read CODEX-SETUP.md and set up my project.

AGENTS.md, plugins, MCP servers

Cursor

CURSOR-SETUP.md

Paste: Read CURSOR-SETUP.md and set up my project.

.cursorrules, Marketplace, @references

Using a different tool? The patterns are universal and MCP servers work everywhere — paste Read SETUP.md and set up my project. for a tool-agnostic runbook.