Ken Muse

Learn how Agent Plugins 1.0 keeps Copilot skills portable while isolating client-specific agents and hooks.

MCP tools can return exact data while an AI assistant reports something different.

GitHub Copilot CLI includes a few special agents to help you with your tasks.

Learn how a SKILL.md path convention helps GitHub Copilot resolve bundled scripts before execution, avoiding a failed call and an extra model turn.

Bigger context windows can still fail.

A real AI-agent escape shows how to protect your dev environment with sandboxes, limited credentials, egress controls, approvals, and logs.

That 'see other file' line in your skill isn't free -- it triggers extra turns, extra tokens, and the model may never even load the referenced file.

Asking an AI model to create numeric scores is unreliable and risky.

The agent succeeded -- or did it?

Give GitHub Copilot its own microVM with a firewall and monitoring so that you can worry less about what your AI is doing.

Skills are optional -- the model decides when to load them.

Keep Copilot skills small -- use a script to deliver only the instructions that match your version and license, and why that demands supply-chain caution.

Learn to read VS Code's Copilot token logs and calculate the exact AIC cost of each request step by step.

Tired of noisy diffs full of auto-generated timestamps?

VS Code's Copilot automatically injects your workspace structure into every chat.

Learn how agentic memory helps AI coding assistants retain context across sessions, with a look at memory types and Copilot's implementation.

Learn why using multiple AI subagents improves results through cleaner context, natural output diversity, sharper task prompts, and adversarial reviews.

Pin VS Code extension versions in dev containers to prevent supply chain attacks.

Explore how rootless Docker uses user namespaces, why kernel developers have concerns about the attack surface, and what unconfined flags really disable.

Trace how Docker's daemon, socket, image builds, and BuildKit all rely on root-level Linux kernel privileges.

Build container isolation from scratch with Linux namespaces and cgroups to understand the kernel features Docker uses.

Learn how temperature, Top-K, Top-P, and Min-P control LLM output by shaping the probability distribution models use to select the next token.

Run Copilot CLI sessions in parallel without repo chaos.

Learn why focused AI agents outperform generalists.

A kitchen analogy that makes it easier to know when to use Copilot's customization features: instructions, prompts, agents, skills, MCP, hooks, and plugins.

Package and share Copilot customizations with agent plugins -- build a plugin, create a marketplace, and install it in VS Code or the CLI.

Explore VS Code's AI extension APIs and when these APIs make sense over using a standalone MCP server.

Learn how to embed an MCP server in a VS Code extension to take your development work to the next level.

Explore GitHub's new Agentic Workflows -- AI coding agents in GitHub Actions with defense-in-depth security and natural language automation.

Transform unpredictable AI behavior into reliable results using MCP tools for deterministic context.

Discover Model Context Protocol (MCP), the open standard that lets AI models connect directly to your databases, APIs, and tools for deterministic results.

Transform your dotfiles repo into a GitHub Action that secures runners without tokens or manual cloning.

Learn how to use Git notes to attach metadata like build results or review comments to commits without rewriting history or breaking signatures.

Your Perforce branch history vanishes in Git because Git has no place to store it -- here's why and how to help your team adapt.

Demystifying Git merges -- from merge commits and fast-forwards to rebases and cherry-picks.

Reviewing another year of blogging and celebrating growth, community connections, and the road ahead in 2026.

A deep dive into Git's internal storage model -- how blobs, trees, and commits work together to track your code's history without a traditional database.

Deploy persistent services on custom GitHub runner images to speed up builds, reduce egress costs, and cut external dependencies.

Cache large repositories on custom GitHub runner images to speed up clones from minutes to seconds using Git reference clones.

Learn the best way to protect sensitive information from leaking into the logs on GitHub runner custom images during both build-time and run-time.

Learn how to use OIDC tokens in GitHub Actions custom images to authenticate with private container registries without storing credentials.

Discover how to improve GitHub Actions performance and drastically cut the time and bandwidth required to use Docker images in your workflows.

Use pre-job scripts in custom runner images to enforce workflow validation and ensure only approved workflows run on your GitHub Actions runners.

Learn to layer secret management techniques through practical examples that build defense-in-depth security for GitHub Actions, containers, and more.

Move beyond static credentials with federated auth, managed identities, and secret vaults -- learn the tradeoffs and security considerations.

Explore practical approaches to storing secrets securely, from files to HSMs, with real-world hardening strategies you can implement today.

Learn how to capture custom GitHub-hosted runner images, add pre- and post-job hooks, and make them part of your daily workflows.

Use Azure flexible federated identity credentials with GitHub Actions to secure your workflows with custom OIDC claims approval expressions.

How signed commits and repository protections completed my defense against the Shai-Hulud supply chain attack.

Simple security practices that protected my dev environment from the Shai-Hulud supply chain attack -- and how you can use them too.