Knowledge Base for Agentic Systems

## Requirements

we have these purposes for this , first is to make the AI not read the full codebase and give the context from the past experience for the new requirements. agents should use the past knowledge to make the decision and ignore the same errors and problems coming again and again in system.

second purpose for this is to if a new person is onboard in the project then they get all the idea of the project from this knowledge base , now that person can be anyone like business analyst , product designer , UI/UX dev ,product thinker , backend dev, frontend dev , data manager , project manager , tester , QA , etc anyone from the team. so they get clear idea of what is project about , business flow , architecture , and anything they needed should be there for all of them.

thirst thing is we have to keep this knowledge base compatible for all different type of projects with different tech stack , different nature , and different use cases of business oriented and NGOs.

fourth thing is that we have too keep the AI don't get exhaust , from exhaust i mean in two terms first is in the token , we should use the minimal tokens to give a best response which is not bargained with the performance and response. second is if AI get very big context and context then there will be high probability of it getting hallucinate so we have to keep this both things aligned.

## Description

This issue introduces an experimental Knowledge Base system designed to support our agentic AI workflows.

As our use of AI agents grows across repositories, we need a structured and reusable context layer that captures business flows, architectural rules, risk models, and sprint state in a format optimized for machine reasoning.

Unlike traditional documentation, this knowledge base is designed to be actively maintained and programmatically updated by agents based on GitHub activity (issues, status changes, PRs). The goal is to reduce context loss, improve review quality, and make agents business-aware rather than purely code-aware.

This is an infrastructure-level experiment within engineering-recipes to evaluate whether a standardized, stack-agnostic knowledge layer can improve AI-assisted engineering across projects.