Every time you start a new session with Cursor or Claude Code,
your AI agent forgets everything. The error you fixed yesterday.
The pattern you established last sprint. The architectural decision
you made three weeks ago.
You end up re-explaining your project constantly. Agents repeat
mistakes. Patterns get ignored. It slows everything down.
I built Agentic KDD to solve this.
What it is
Agentic KDD is an autonomous development pipeline that sits inside
your project and works with Cursor or Claude Code. You type:
And it runs completely — Orchestrator → Analyst → Front → Back →
QA with browser validation → Memory update. No supervision needed.
The KDD part
KDD stands for Knowledge-Driven Development — a methodology I
couldn't find anywhere else, so I defined it.
The idea: the accumulated knowledge of your project actively guides
every development decision. Four layers:
- trabajo.md — what's happening right now
- errores.md — error patterns to avoid (written by agents, read before planning)
- patrones.md — rules the project has established (e.g. "persistent modals")
- decisiones.md — why things are the way they are
Every agent reads these before acting. Every agent updates them after.
The system gets smarter with every task.
How to use it
- Copy the files to your project root
- Open in Cursor or Claude Code
- Type:
aa: configurar - Start building with
aa: [your task]
Works with any stack — Next.js, Laravel, FastAPI, Vue, anything.
GitHub
→ github.com/Adrianlpz211/Agentic-KDD
If it saves you time, a ⭐ goes a long way.
Top comments (0)