Avant Garde — Security skill for Claude Code

Avant-Garde

**Think before you build. Guard before you ship.**

*Avant-Garde* is a skill for Claude (claude.ai & Claude Code) that stops your AI from writing a single line of code for a new website or app until the **foundations are written down** — and refuses to call it "ready to ship" until the **security gate** is passed.

Built for vibe coders, indie hackers, students and solo founders who build with AI — and for the AI itself. These documents are not paperwork: they are the **shared memory of the human–AI duo**. Every session is a fresh instance; the docs are what lets the pair resume exactly where they left off.

The name is a French double entendre: *avant* (the foundations **before** the code) + *garde* (the security **guard** before going live). And yes, "avant-garde" is the same word in English.

---

Why

When you ask an AI to build an app without telling it precisely what you want, it doesn't stop to ask twenty questions — it decides for you. And when non-developers ship AI-built apps, the same disasters repeat: open databases where any user can read everyone's data, admin keys shipped to the browser, `.env` files committed to public repos, AI routes with no rate limit burning the creator's credit card.

Avant-Garde fixes both problems with one method: **six short documents + a ten-point security gate**, written together with Claude in about thirty minutes, before any code.

What it does

When the skill is active and you ask Claude to build a site or app, it:

1. Triages your project (30 seconds): static page → light track (3 docs); accounts, database, payments or AI calls → full track (6 docs + security gate, non-negotiable).
2. Interviews you document by document — max 5 questions at a time, plain language, nothing invented, doubts marked [TO CONFIRM].
3. Writes the six documents into docs/:

#	Document	Answers
01	Product (PRD)	What are we building, for whom — and what are we NOT buil