// Context Engineering

An interactive mental model for agentic AI that explains how instructions, retrieved context, session state, memory, MCP servers, sandbox execution, tool results, and the Think-Act-Observe loop fit together.

Conversation state and retrieved context lead naturally into memory, but only if we're clear about what memory is for. Rules, skills, and instruction files package what you already know. Memory should capture what the work itself teaches the system, and that means reflection, verification, and forgetting matter just as much as recall.

Context engineering gets more useful when the knowledge it depends on is packaged for reuse. In this post I map the portable core (AGENTS.md and SKILL.md), Copilot-specific concepts like custom instructions, agents and prompt files, and how to decide what knowledge belongs where.

Better prompts help, but they're only part of the story. Context engineering is the craft of designing what an AI agent sees, when it sees it, and how that changes across the session. The goal isn't a bigger context window. It's a more effective one.

Chris continues building custom agents for the software development life cycle. After reviewing the changelog including Copilot Spaces updates and assigning issues to Copilot via API, he explores the awesome-copilot repository for agent inspiration, brainstorms SDLC agent ideas like test writers and security reviewers, and live-builds a GitHub Actions workflow agent with a companion instructions file.

Chris dives into Copilot custom agents and custom instructions for the turn-based game MCP server project. After reviewing the changelog including enterprise bring-your-own-key support and Claude Opus 4.5, he restructures copilot-instructions.md, creates meta instruction files for writing instructions and agents, and explores plan mode for iterating on agent designs.

Chris catches up on several weeks of GitHub releases, including new models including GPT-5.1, GPT-5.1 Codex, and GPT-5.1 Codex Mini, MCP registry controls, and isolated sub-agents. He then returns to the turn-based game MCP server project, using Copilot's agent mode to diagnose and fix a dependency upgrade failure.

In this live stream, we explore building a 3D tic-tac-toe visualization using Three.js and Copilot coding agent, demo MCP elicitation for gathering game preferences, and discuss the importance of context engineering when working with AI tools. We also cover GitHub changelog highlights including path-scoped custom instructions for Copilot code review and agents.md support.

In this episode, we recap the latest GitHub changelog including GitHub Spark entering public preview and coding agent support for instructions.md files. We introduce the 'Build for the love of code' hackathon, prototype a rubber duck debugging app with GitHub Spark, and build an MCP server from scratch using TypeScript while exploring tools, prompts, and resources.

In this stream, Chris catches up on several weeks of GitHub updates including the remote MCP server preview and Copilot coding agent for business users. The live coding session demonstrates adding internationalization to the Copilot Airways app using Copilot coding agent, custom VS Code chat modes for planning, and agent mode in Xcode for iOS development.