A portable, token-optimized AI workflow for React projects. One
AGENTS.md
gives Claude Code, OpenCode, Copilot, and Codex the same shared
conventions, curated skill libraries, and structured ticket pipeline —
so every agent ships with the same discipline, every time.
# Clone the workflow starter$ git clone
https://github.com/wannacry081/react-workflow-starter.git$ cd
react-workflow-starter# Bootstrap into your React project$
node scripts/ai/init-project.mjs
/path/to/your/project# Select skills, sync, and prune starter files$ npm run skills:suggest$ npm run sync-skills$ npm run prune:starter -- --dry-run$ npm run prune:starter$
node scripts/ai/generate-context.mjsThis project is inspired by everything-claude-code and reworked into a frontend-first system you can actually run from your terminal. Instead of hard-coding behavior into one giant prompt, you compose your workflow by installing focused skills via Context7 and skill.sh.
The flow stays disciplined:
Ticket → Plan → Work → Review → Ship → Done.
Each phase produces a concrete artifact in
docs/project/,
so handoffs are clear, review is faster, and context does not get
lost across sessions. You get practical structure without slowing
down delivery.
The repository also ships with code-review-graph as part of the default setup. Agents are expected to use the graph first for exploration, impact analysis, and review context because it is faster, cheaper on tokens, and more precise than broad grep passes.
If you want your own version of AI-assisted development, this is a clean baseline: install the skills you need, swap in your stack standards, and ship with a repeatable loop built for real projects.
Pick a starter based on your product surface. Each workflow follows the same skill-driven, plan-first, ship-fast discipline.
Backend
API-first workflow with clear contracts, validation, and review gates.
Mobile
Structured setup for React Native and mobile-first product teams.
Mobile
Cross-platform mobile development with plan-first workflow discipline.