Context Engineering on Chris Reddington

An interactive agentic AI mental model

Thu, 16 Apr 2026 00:00:00 +0000

An interactive agentic AI mental model is an interactive guide to how an agentic AI system works. It brings together the moving parts that are easy to talk about separately, but harder to hold as one picture when you are building with agents.

The project walks through how instructions, capabilities, retrieved context, session state, memory, MCP servers, sandbox execution, tool results, and the Think-Act-Observe loop connect in practice.

Features

Interactive mental model: Explore the core components of an agentic AI system in one connected view
Context engineering focus: Understand how context windows, retrieved context, and session state shape agent behaviour
Operational building blocks: See where memory, tool use, MCP servers, and sandbox execution fit into the overall system
Practical framing: Use the Think-Act-Observe loop as a simple way to reason about how agents work in real workflows

Agentic memory: what agents should and shouldn't remember

Fri, 10 Apr 2026 00:00:00 +0000

While building some of my own AI-agent based projects, I’ve run into the same frustrating loop. An agent would uncover something useful in the work itself (a hidden dependency, an awkward repo rule, a dead end I’d already ruled out), and then forget it all in the next session. I’d steer it back on course, start fresh later, and watch it head straight for the same mistake again. All that earlier momentum had gone.

AGENTS.md and SKILL.md: building a reusable agent toolbox

Wed, 08 Apr 2026 00:00:00 +0000

I was working on a personal project and coming up with a Copilot CLI demo to show and tell at the GitHub Social Club in London yesterday. But as I started a fresh agent session, and typed the setup, I caught myself writing the same lines I’d written for some work a few days earlier.

It included a few lines around the process for writing out plans, how the agent should hand off between planning, what the implementation expectations were, and how to review the work when it was done. That repetition (me repeatedly hitting the up arrow to get to my earlier prompts) was a sign that the knowledge I was typing should be packaged as something reusable, not left as another throwaway prompt.

Context engineering: more context isn't better context

Sun, 05 Apr 2026 00:00:00 +0000

I was in a customer presentation recently where I asked the room whether they were familiar with the term prompt engineering. Almost everyone raised their hand. This wasn’t a surprise, given how it’s been used over the last 3 years to make sure that being specific, giving examples, and setting clear instructions are part of the conversation around AI tools.

But then I asked whether they were familiar with context engineering. Only a few people raised their hand, and even fewer were using it as a deliberate way to think about how they work with AI agents. I genuinely expected more people to have heard of it given how much the term has grown over the past year, even if they weren’t applying the principles just yet.

Rubber Duck Thursdays - Let's build with custom agents (again!)

Thu, 04 Dec 2025 00:00:00 +0000

Chris continues building custom agents for the SDLC, exploring the awesome-copilot repository for inspiration and live-building a GitHub Actions workflow agent with plan mode.

Topics Covered

GitHub Changelog Roundup — Blocking repo admins from installing GitHub Apps now GA, Copilot Spaces with public spaces and code view support, secret scanning updates for November 2025, GitHub Enterprise Server 3.19 RC, assigning issues to Copilot via GraphQL and REST APIs, and Claude Opus 4.5 availability across more IDEs.
Agent Inspiration from awesome-copilot — Reviewing partner-built agents and the awesome-copilot repository for ideas including test writers, security reviewers, tech debt analyzers, documentation generators, PR review assistants, and onboarding guides.
Agent Consolidation — Discussing whether to keep agents separate or consolidate them based on shared tool access, output formats, and domain overlap.
Building a GitHub Actions Agent — Using plan mode to design a custom agent for authoring and updating GitHub Actions workflows, including monorepo build order awareness and minimal permissions guidance.
GitHub Actions Instructions File — Creating a companion instructions file with repository-specific CI/CD conventions, build order, environment variables, and recommended practices for workflow definitions.
Plan Mode Workflow — Demonstrating the iterative plan-then-implement workflow where Copilot asks clarifying questions before generating code, and switching between plan and agent modes.

Rubber Duck Thursdays - Let's build with custom agents

Thu, 27 Nov 2025 00:00:00 +0000

Chris explores Copilot custom agents and custom instructions, restructuring project context files and creating meta instruction files for a more effective AI-assisted development workflow.

Topics Covered

GitHub Changelog Roundup — Enterprise bring-your-own-key for Copilot, linter integration with Copilot code review (ESLint, PMD), GitHub Actions cache size exceeding 10 GB, Claude Opus 4.5 in public preview, and secret scanning updates.
Custom Instructions Deep Dive — Restructuring copilot-instructions.md to include monorepo tech stack details, build order, service boundaries, security guidelines, and testing conventions using recommended practices from the docs.
Meta Instruction Files — Creating instruction files that teach Copilot how to write better custom instructions and custom agents, pulling in recommended practices from the GitHub and VS Code documentation.
Plan Mode for Agent Design — Using Copilot’s plan mode to iterate on ideas before implementation, demonstrating how follow-up questions help refine requirements like rubber duck debugging.
Context Engineering — Discussion on supplying the right context to Copilot, balancing context window size, and the importance of being requirements-driven for high-quality code generation.
Community Q&A — Topics including sustainable engineering pace, avoiding burnout, handling merge conflicts, and the fundamentals of software quality.

Rubber Duck Thursdays - Let's build

Thu, 20 Nov 2025 00:00:00 +0000

Chris catches up on several weeks of GitHub releases and then returns to the turn-based game MCP server project to fix a dependency upgrade issue.

Topics Covered

GitHub Changelog Roundup — Managing Copilot coding agent tasks in VS Code, new OIDC token claims for GitHub Actions, GPT-5.1 and Codex models, configuring the coding agent as a bypass actor for rulesets, MCP registry and allowlist controls in VS Code Stable, plan mode and isolated sub-agents in public preview, Gemini 3 Pro, and CodeQL updates.
MCP Server for Games — Revisiting the tic-tac-toe MCP server project and demonstrating how tool calls allow natural language game interaction with a backend API.
Fixing Dependency Upgrades with Copilot — Using agent mode across multiple models (Codex, Gemini 3 Pro) to diagnose and resolve a type check failure caused by a Dependabot SDK version bump.

Rubber Duck Thursdays - Time to build!

Thu, 04 Sep 2025 00:00:00 +0000

Join us for Rubber Duck Thursdays! A lighthearted and informal stream where we live code on some projects.

In this stream, we explore adding a 3D visualization to the turn-based game MCP server using Three.js, built with help from Copilot coding agent. We demo switching between 2D and 3D views of tic-tac-toe games while playing against the MCP server, and discuss the iterative process of prompting Copilot — including the importance of specifying requirements like piece orientation and camera controls.

Rubber Duck Thursdays - Build for the love of code

Thu, 24 Jul 2025 00:00:00 +0000

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.

Topics covered

GitHub changelog recap — M2 Pro hosted runners, GitHub Issues quality-of-life updates, Azure endpoint deprecation for GitHub Models, PR files changed improvements, Copilot code review policy changes, GitHub Spark public preview for Copilot Pro Plus, coding agent support for instructions.md custom instructions, and base branch selection for coding agent tasks.
Build for the love of code hackathon — Introduced the gh.io/love-of-code hackathon running July 16 to September 22 with six categories: hardware projects, AI agents, terminal tools, games, web apps, and wildcard entries. Brainstormed duck-themed project ideas with Copilot.
GitHub Spark prototyping — Used GitHub Spark (now in public preview) to rapidly prototype a rubber duck debugging web app with animated duck companions and different debugging personalities.
Building an MCP server from scratch — Scaffolded a TypeScript MCP server project from an empty repository, covering server initialization with the MCP SDK, registering tools with Zod schema validation, defining prompts and resources, and setting up repository foundations including dev containers, Dependabot configuration, and GitHub Actions CI workflows.

Rubber Duck Thursdays - Let's build

Thu, 03 Jul 2025 00:00:00 +0000

In this stream, Chris returns from vacation to catch up on weeks of GitHub updates and demonstrates adding internationalization to apps using Copilot coding agent, custom VS Code chat modes, and agent mode in Xcode.

Topics Covered

GitHub Changelog Review — Updates including model evaluations CLI for GitHub Models, remote GitHub MCP server in public preview, Copilot coding agent now available for business users, GitHub Models pay-as-you-go billing, Claude Sonnet 4 and Opus 4 GA, CodeQL updates for Go and Swift, Actions Runner Controller release, and Copilot code review with custom instructions support
Custom Chat Modes in VS Code — Creating a custom planning mode in VS Code Insiders to generate implementation plans and create GitHub issues, integrating the GitHub MCP server as a tool within the chat mode
Remote GitHub MCP Server — Setting up and authenticating with the new remote GitHub MCP server, eliminating the need for local Docker containers and personal access tokens via OAuth
Copilot Coding Agent for Internationalization — Assigning a GitHub issue to Copilot to add English, French, and Spanish language support to the Copilot Airways Next.js web app, reviewing the AI-generated pull request
Xcode Agent Mode — Demonstrating agent mode in Xcode to add internationalization to a Swift iOS version of the Copilot Airways app with a language selector
Issue Creation with Copilot — Using Copilot on github.com to draft well-structured GitHub issues from natural language descriptions

Rubber Duck Thursdays - Let's build with agents

Thu, 19 Jun 2025 00:00:00 +0000

In this episode, we recap the latest GitHub changelog updates including Copilot coding agent availability for pro users and the new Playwright MCP browser capability. We update Copilot custom instructions and setup steps for a GitHub Action project, use the Playwright MCP server to build interactive UI features on a trend radar app, and leverage Copilot coding agent to identify and fix technical debt.

Topics covered

GitHub changelog recap — Copilot coding agent for Copilot Pro users, GitHub Actions fine-grain permissions, upgraded Llama models, improved PR file experience, Dependabot minimum package age configuration, and Copilot coding agent Playwright MCP browser support.
Copilot custom instructions and setup steps — Updated custom instructions for a validate-file-exists GitHub Action project and created a Copilot setup steps YAML to configure the coding agent environment with Node.js and dependencies.
Playwright MCP server for UI development — Used the Playwright MCP server with Copilot coding agent to build drag-and-drop functionality and point editing on a trend radar web app, demonstrating how the agent can interact with a running application.
Technical debt with Copilot coding agent — Prompted Copilot to identify technical debt in the project, then created GitHub issues for package metadata fixes, documentation inconsistencies, and input validation improvements. Copilot coding agent autonomously submitted a PR addressing all three items with new tests.

Copilot Agent Mode is now available in GitHub Copilot for Xcode

Sat, 24 May 2025 00:00:00 +0000

GitHub Copilot for Xcode has added agent mode, bringing the same agentic capabilities available in Visual Studio Code to native Apple platform development. This video covers the key features and a practical demonstration.

Topics and demos include:

The new ask/agent mode toggle in the Copilot chat pane within Xcode—ask mode for back-and-forth conversation, agent mode for autonomous task execution that can invoke tools and run terminal commands
Configuring Model Context Protocol (MCP) via the Xcode settings page: navigating to the MCP tab, editing the configuration JSON, and reviewing available tools based on what’s already configured
Adding the GitHub MCP server to give Copilot access to GitHub repository data (issues, pull requests, etc.) directly from within Xcode
Using agent mode to look up open issues and summarize the state of work across the repository, helping prioritize what to tackle next
A hands-on fix: identifying that instant (on-first-keystroke) form validation creates a poor user experience in an iOS app, writing a clear instruction with explicit expectations for Copilot to follow, and watching it explore workspace files and apply the targeted change
Verifying the fix in the iOS simulator: confirming that input fields no longer show red validation errors on first load, and that error messages appear correctly when invalid data is entered

How to get the most out of the Copilot coding agent

Wed, 21 May 2025 00:00:00 +0000

The GitHub Copilot coding agent works best when given rich context and a consistent environment. This video breaks down four concrete practices that help set Copilot up for success before a single line of code is written.

1. Write well-scoped GitHub Issues — Can the issue be resolved in a single pull request? Does it include a clear problem description, acceptance criteria, pointers to the relevant files, and a loose implementation plan to follow?

Rubber Duck Thursdays - Building from requirements with Agent Mode

Thu, 15 May 2025 00:00:00 +0000

In this stream, Chris walks through building a GitHub CLI extension in Go from scratch, taking a spec-driven and test-driven development approach powered by Copilot Agent Mode.

Topics Covered

GitHub Changelog Review — Model updates in GitHub Models (Mistral, Cohere, Llama), GPT-4.1 as the new default Copilot model, updated rate limits for unauthenticated requests, and the VS Code April release with MCP support and prompt files
Microsoft Build Preview — A look ahead at sessions and events for the following week
Building gh-demo from Scratch — Bootstrapping a new Go-based GitHub CLI extension using VS Code’s new workspace creation with Copilot, the Cobra library, and a spec-driven approach starting from a product requirements document
Test-Driven Development with Copilot — Writing tests first for a hydrate subcommand that reads JSON files to create GitHub issues, discussions, and pull requests, with content type filtering and label collection
GitHub Actions CI — Creating a build, test, and lint workflow for the new CLI extension
Custom Instructions — Setting up a copilot-instructions.md file to guide Copilot’s behavior around path handling and project conventions

Rubber Duck Thursdays - Agent mode and custom instructions

Thu, 08 May 2025 00:00:00 +0000

Chris continues exploring Copilot Agent Mode and custom instructions by building a memory sequence game for the GitHub CLI extension project in Go. The changelog segment covers GitHub CLI accessibility improvements, Copilot Code Review expanding to all languages, and the introduction of Copilot premium requests.

Changelog Highlights

GitHub CLI accessibility improvements — speech synthesis, screen reader support, enhanced prompting and progress indicators
Copilot Code Review now supports all languages with AI-powered suggestions and contextual feedback
Copilot Autofix updates for automated pull request suggestions
Copilot premium requests — announcement of upcoming premium model usage limits and visibility controls in VS Code and Visual Studio
GitHub CLI extension repo transferred to the github-samples organization

Live Coding — Memory Game with Agent Mode

Recapped custom instructions from the previous week and their impact on Copilot output quality
Built a memory sequence game (colors: red, yellow, green, blue) as a new subcommand in the Go-based GitHub CLI extension using Copilot Agent Mode
Used the Cobra command library and Bubble Tea for terminal UI interactions
Wrote detailed specs in chat — game mechanics, lives system, sequence display timing — and let Agent Mode scaffold the implementation and tests
Played live demos of coin toss, higher/lower, and word guess games from the CLI
Discussed potential future projects including VS Code extension development

Rubber Duck Thursdays - The one with custom instructions

Thu, 01 May 2025 00:00:00 +0000

This episode dives deep into Copilot custom instructions — what they are, how to write them, and why they matter for guiding AI-assisted development. Chris reviews the GitHub Changelog and then live codes a higher/lower game for a Go-based GitHub CLI extension using Agent Mode with detailed custom instructions.

Changelog Highlights

Copilot Code Review now supports C, C++, Kotlin, Swift, and more
CodeQL improves JavaScript and Ruby analysis in version 2.21.1
GitHub Models playground adds a prompt improvement feature to refine prompts with AI
Copilot Edits for JetBrains IDEs is now generally available
Credential revocation API for exposed tokens is now GA — supports bulk revocation via REST API
Task lists retirement on April 30th — code scanning alert references deprecated
Mistral Small 3.1 replaces Mistral Small in GitHub Models
Actions Runner Controller now supports running Dependabot on self-hosted Kubernetes runners

Live Coding — Custom Instructions Deep Dive

Explored how Copilot custom instructions provide repository context, coding standards, and structural guidance to every Copilot request
Demonstrated writing detailed instructions for a Go-based GitHub CLI extension — specifying test commands, linting rules, folder structure, and game design patterns using the Cobra library
Live coded a higher/lower guessing game using Copilot Agent Mode with Claude 3.7 Sonnet, showcasing how rich instructions lead to higher quality scaffolded code
Showed the Copilot Airways app built with Agent Mode from a screenshot, demonstrating vision capabilities and custom instruction-guided development

Rubber Duck Thursdays - Building with Agent Mode and MCP

Thu, 10 Apr 2025 00:00:00 +0000

A big announcements week — this episode unpacks Copilot Agent Mode going stable in VS Code and the launch of the official GitHub MCP server, then puts both to work building a real app from scratch.

The GitHub Changelog segment covers the VS Code Copilot March release (agent mode GA, MCP support, next edit suggestions), the GitHub MCP server launch, Copilot Pro Plus tier, Copilot Code Review GA, security campaigns GA with automated issue creation, GitHub Issues and Projects improvements (sub-issues, issue types, 50K item limit), Helm support for Dependabot version updates, and more.

Rubber Duck Thursdays - Creating a brickbreaker walkthrough

Thu, 27 Mar 2025 00:00:00 +0000

Chris live-codes a GitHub-themed brick breaker game walkthrough using Copilot agent mode, demonstrating how prompt framing and context shape AI-generated code. The session covers bootstrapping project structure from an existing Game of Life template, Copilot custom instructions, and the latest GitHub changelog including enterprise rulesets, Copilot edits in JetBrains, and dependency label improvements.

In this episode, Chris sets out to build an educational walkthrough — similar to an existing Conway’s Game of Life sample — but for a brick breaker game styled with GitHub’s contribution graph colors. Using Copilot agent mode, he bootstraps the project structure by referencing the Game of Life walkthrough as context, then iterates on lesson pages focused on teaching prompt engineering for agent mode. The live coding demonstrates key concepts: how vague prompts produce incomplete results, how adding specific requirements (responsive containers, keyboard and mouse controls, bouncing physics) improves output, and how Copilot custom instructions (including a fun “talk like a pirate” experiment) modify agent behavior. Chris also briefly showcases the OctoSnap arcade concept and discusses ideas for making brick breaker use GitHub contribution data for brick colors and scoring. The changelog segment covers Mistral Small 3.1 in GitHub Models, upcoming GitHub Actions cache service changes, Copilot edits in JetBrains IDEs, enterprise custom properties and rulesets going GA, and Maven dependency labeling in the dependency graph.

Rubber Duck Thursdays - Building a scoring system

Thu, 06 Mar 2025 00:00:00 +0000

Chris walks through building “OctoSnap,” a GitHub-themed memory card matching game using Next.js and JavaScript. The session dives deep into designing a scoring model with GitHub Copilot — covering difficulty multipliers, time-based rewards, and penalty mechanics — and reviews the latest GitHub changelog including new AI models, GitHub Advanced Security pricing changes, secret scanning push protection APIs, and code scanning improvements for GitHub Actions workflows.

In this episode, Chris steps away from the Go-based CLI game extension built in previous weeks and introduces OctoSnap, a browser-based card matching game featuring artwork from the GitHub Octodex. He demonstrates the game across easy, medium, and hard difficulty modes, showing how image transformations like rotation and color filters increase challenge. The core focus is on iterating with Copilot to build a nuanced scoring system that balances mode difficulty multipliers, time duration bonuses, perfect play ratios, and random click penalties. Chris shares how he broke the problem into granular prompts for Copilot and visualized the scoring curves to ensure fair overlap between difficulty tiers. The changelog segment covers new models in GitHub Copilot and GitHub Models, GitHub Advanced Security being split into Secret Protection and Code Security standalone products, delegated alert dismissal, expanded CodeQL support for GitHub Actions and Go 1.24, and GitHub Mobile updates including Copilot chat and sub-issues.

Game of Life Walkthrough

Wed, 18 Dec 2024 00:00:00 +0000

This is a companion walkthrough for the hero Game of Life video that was published on the GitHub YouTube channel. It provides a step-by-step guide to the concepts covered in the video, including:

GitHub Actions
GitHub Codespaces
GitHub Copilot
GitHub Copilot Chat
GitHub Copilot Edits
GitHub Copilot Instructions
GitHub Copilot Extensions
GitHub Pages