When I first started using AI coding, I was just like everyone else. Very excited. "Wow, it writes code really fast!" Then three months later... looking back at the codebase made me want to cry. Technical debt piled up. Zero tests. Security holes I didn't know about. Deploying manually. Every time I opened a new chat, the AI completely forgot everything we discussed. I realized: the problem wasn't that the AI was bad, but that the AI lacked a process, like hiring an extremely skilled developer without a PM, no QA, no code reviews, no architect. It writes quickly but the quality is... So I started to build Forgewright, which wasn't a one-shot deal. It was a loop. Repeating over and over. Tiring but each loop was better than the last.
Gathering. I searched for best practices from everywhere. How Google does code reviews. How Stripe designs APIs. How game studios manage their pipelines. OWASP security. Martin Fowler's architecture patterns. And the prompt engineering sets shared by the community, I collected everything useful.
Testing in real scenarios. Not just testing on paper. I used it to build real SaaS, build real games, build real mobile apps. Noting what the AI got wrong, what it skipped, what it hallucinated.
Improving. Each failure became an additional rule. AI doesn't self-write tests? Added QA skill requiring TDD. AI forgets context? Built a memory set. AI doesn’t understand large codebases? Integrated a knowledge graph. AI writes bad UIs? Built a design database with 85 styles + a reasoning engine. And so on. Dozens of loops. Fail. Learn. Fix. Repeat. The result after many iterations: 48 specialized AI skills, each skill is an expert: BA for requirement analysis, PM for writing user stories, Architect for design.
Gate 1: You review the design. Engineer writes code (TDD-first), QA writes test pyramid, Security audit.
Gate 2: You review the code. DevOps sets up CI/CD, SRE implements monitoring.
Gate 3: You review the deployment. AI runs automatically, but you control at three crucial points called not just the web. Pipelines for all:
Game Development (Unity / Unreal / Godot / Roblox)
Mobile (React Native / Flutter)
AI/ML (RAG, fine-tuning, MLOps)
Marketing (SEO, CRO, campaigns) Now I'm sharing it back with the community.
Open source, free.
Setup with just one line: ./setup.sh install.
Then say: "Build a SaaS for [your idea]" and the pipeline runs automatically.
If you want to be even stronger, add:
GitNexus: AI understands the entire codebase, knows where changes will affect what.
mem0: AI remembers decisions across sessions, no more forgetting.
NotebookLM: research with sources, create summary podcasts.
Paperclip: multiple AI agents running in parallel like a company.
Full instructions in the README. https://github.com/buiphucminhtam/forgewright 48 skills · 19 modes · 15 protocols From idea to production, just one phrase. I'm not saying it's perfect. It is the result of many failed loops. But each loop it gets a little better. And I believe the community will help make it even better. If you find it useful, hit
Star on GitHub to help me out. Each star is additional motivation for me to keep improving. And share with anyone using AI coding, who knows it might help them to
No talks yet. Create the first talk
No reviews yet

Agent Skills and Structured Prompts for Scalable Development

NeuralMemory stores experiences as interconnected neurons and recalls

Code mọi nơi với AI Agents

Auto-resume Claude Code when your usage limit resets — single static R

LinesYu/trainingAI at hoi - dong - toi - cao

Bộ công cụ tối giản vibe kit nhiều dự án

The only AI pipeline that encompasses the entire lifecycle.
Auto-translated ·
Reviews & comments