Configure your full-stack AI coding agent

A ready-to-use bundle of files — the exact set depends on your stack(s), your generation mode, and which AI coding tool(s) you pick. Every file is built from the same consolidated Master Agent content: your stack's 10 Core Operating Directives, real best practices and anti-patterns, the Universal Guardrails (code quality, error handling, token efficiency, security), and a mandatory end-of-session Task Summary & Impact chart — all enforced together instead of scattered across separate instructions/agents/rules files.

Combined (the default, recommended for multi-stack picks) merges every selected technology into one cross-stack-aware agent — plus a universal .agent.md at your repo root — so it understands how a frontend change reaches a backend endpoint or a database schema. Modular gives each stack its own independent file per tool instead, useful when different files should stay ownable by different people or teams. Tools with only one fixed root file (Windsurf, Aider, AGENTS.md) always get the merged content, since their format has no per-stack convention to split into.

Four rules embedded in every generated file, combined or modular, regardless of stack: (1) Code Optimization & Quality — clean, production-ready code with no dead imports or leaked resources; (2) Error Handling & Validation — defensive programming and explicit edge cases at every boundary; (3) Token Efficiency — concise, structured responses with no filler; (4) Security & Safety — injection prevention, no exposed secrets, deliberate CORS/auth, and input validation on every API route and form. A fifth standing rule, Mandatory End-of-Session Reporting, requires a markdown Task Summary & Application Impact table at the end of every response.

Zero hallucination & codebase verification, strict type safety, bug prevention, token efficiency, comprehensive unit testing, OWASP & security standards, performance & resource management, architectural consistency, structured error handling, and — mandatory on every task — automatically syncing README.md with whatever changed.

No — and we're not going to pretend otherwise. The bundle content is generated by deterministic templates, not a language model. There's nothing to wait for, nothing that can hallucinate, and nothing that needs an API key.

GitHub Copilot, Claude Code, Cursor, Windsurf, Cline, Continue.dev, Aider, and the open, multi-vendor AGENTS.md convention. Pick as many as your team actually uses — each one gets its own correctly-formatted file(s) from the same underlying content. See the Guides page for exactly how each tool reads its file.

22 stacks across six categories: Languages (Python, Java, Go, Rust, TypeScript), Backend & APIs (FastAPI, Django, Node.js, Spring Boot, Ruby on Rails, C#/.NET), Frontend (React, Next.js, Vue, Angular), Database (PostgreSQL, MongoDB, SQL/Prisma), Mobile (Flutter/Dart), and Tools (Docker/Kubernetes, Terraform, Tailwind CSS). Check as many as apply to build a real-world combination like Python + FastAPI + React + MongoDB — there's no limit on how many you combine.

Whenever your selection spans multiple layers, a Cross-Stack Guardrails section is added automatically — generic rules for any frontend+backend, backend+database, or full-stack combination, plus curated guidance for specific, common pairings like React+FastAPI, Next.js+Prisma or FastAPI+MongoDB. It's how the agent knows a database migration isn't done until the backend model and the frontend that depends on it are updated too.

Yes. Every file sits at its real project-root-relative path inside the zip — extracting it directly into your project root populates .github/, CLAUDE.md, .cursor/, .agent.md, etc. immediately, with no extra wrapper folder.

The generator itself, no — stack selection, mode, target selection, bundle generation and the zip file are all built entirely in your browser, and your last picks are remembered in your browser's local storage only. The one exception: the site owner records anonymous, aggregate usage — page paths, a bare referrer hostname, and which stacks/tools/mode a generated bundle used — visible only on a password-protected internal dashboard. No file content, no custom team-rules text, no IP address, and no cookie or fingerprinting is ever collected.

Yes, and you should. They're a strong, opinionated starting point, not a rulebook carved in stone — trim, extend or reword any directive to match how your team actually works. The Guides page has tips on what's worth customizing first.

Every file we generate is stamped with the content version it came from. AI coding tools change their own conventions constantly, so if you come back after we've shipped new guardrails, you'll see a banner here telling you exactly what changed and offering a one-click regenerate — instead of silently working off a rules file that's fallen behind. See the Changelog page for the full history.