A new open-source project called Full Stack HQ is aiming to solve one of the most frustrating problems with AI coding assistants: unpredictability. The configuration kit, posted to Hacker News on May 14th by developer sabahattink, packages together CLAUDE.md and GEMINI.md configurations along with a full agent stack—10 specialist personas and 28 deep knowledge modules—that install in under 30 seconds via curl or PowerShell.

Permission-First Workflow Changes the Dev Loop

The core innovation here isn't another AI model—it's discipline. Full Stack HQ enforces what it calls a "permission-first workflow" where the agent plans before acting, shows you exactly what it intends to do, and waits for explicit approval. When you ask it to add user authentication with JWT, it responds with a phased breakdown like "Phase 1: Create auth module + JWT strategy" followed by an [APPROVAL NEEDED] prompt. The agent only implements one phase at a time, then stops and reports back. Valid approval keywords include PLAN APPROVED, IMPLEMENTATION APPROVED, PROCEED, and DO IT—anything else keeps the agent waiting.

Ten Specialist Agents Handle Domain-Specific Tasks

The project ships with ten distinct AI personas triggered by phrases like "Use the frontend-specialist to..." or "Use the security-auditor to..." The roster includes dedicated agents for React/Next.js/Tailwind work, NestJS backend tasks, Prisma and PostgreSQL migrations, system architecture decisions, code review, testing with Vitest/Jest/Playwright, OWASP security audits, Docker/CI/CD deployment, bundle optimization, and technical writing. Each agent has a specific trigger phrase designed for natural integration into your workflow.

Tech Stack Opinions and Slash Commands

Full Stack HQ comes pre-tuned for what the maintainer calls "modern production stacks": Next.js 15 with App Router, TypeScript 5, Tailwind CSS v4 on the frontend; NestJS, Node.js 22+, BullMQ, and Redis on the backend; PostgreSQL 16+ with Prisma 6+ for data; JWT with refresh token rotation for auth; Vitest/Jest/Playwright for testing; and Docker/GitHub Actions/Vercel for infrastructure. The project also includes ten slash workflow commands including /plan, /brainstorm, /debug, /create, /enhance, /test, /status, and /orchestrate—designed to slot into Claude Code's native command system.

Cross-Platform Install with Granular Control

Installation targets both Mac/Linux (curl -fsSL pipeline) and Windows PowerShell (irm command). The install script supports flags for --only-antigravity or --only-claude if you only use one IDE, plus a --force flag to overwrite existing configs. After installation, you restart your IDE and the configurations activate automatically, dropping files into ~/.gemini/ and ~/.claude/ directories with agent definitions, skill modules, and workflow scripts.

Open Source with Community Contributions Welcome

Full Stack HQ carries an MIT license and explicitly welcomes contributions. The CONTRIBUTING.md guide walks through adding new skills—create a directory under skills/, add a skill.md file, and open a PR. The maintainer credits inspiration from vudovn/antigravity-kit, sickn33/antigravity-awesome-skills, Vercel Labs official skills, and Anthropic's own official skills. At time of publication, the Hacker News post had received 6 points with no visible comments.

Key Takeaways

  • Full Stack HQ enforces structured permission workflows before any AI action executes—solving the "surprise file creation" problem that plagues many developers
  • The project targets both Claude Code and Google Antigravity IDE with near-identical configuration sets
  • 10 specialist agents + 28 skill modules ship out of the box, covering frontend, backend, database, security, DevOps, and documentation domains
  • MIT licensed, community-contributable, with a one-command install process for Mac/Linux/Windows

The Bottom Line

This is exactly the kind of infrastructure project the AI coding ecosystem needs right now—opinionated defaults that make agents actually useful in production environments rather than just impressive demos. The permission-first approach won't be for everyone (some will find it tedious), but for teams working on critical systems where surprises cost money, Full Stack HQ offers a disciplined framework that other config kits should copy.