Context Engineering

Context

You are assembling the information an AI needs to work on a real task — spread across repo files, tickets, documentation and notes — and you want it reusable, not re-gathered from scratch every session.

Consider these patterns first

This is a foundational pattern — nothing needs to come before it.

Problem

How do you give an AI the right, complete context for a task without hand-pasting it every time, and without burying the model in irrelevant material?

Solution

Model context as a first-class, named artifact. Group the relevant items — repo files, Confluence pages, Jira issues, instructions, scratch notes — into a context, with mixins for fragments shared across contexts. Preview it inline to see exactly what the model will read, and serve it over MCP so any AI pulls the same curated context with a single tool call.

Consider next