Chris reviews the latest GitHub changelog and explores the major VS Code December release, then continues building custom agents for a game MCP server project.
GitHub Changelog Highlights
CodeQL 2.23.6 adds Swift 6.2.1 and new C# security queries
GPT-5.1 Codex Max now in public preview for GitHub Copilot across VS Code, GitHub.com, and GitHub Mobile
Workflow dispatch inputs limit increased from 10 to 25
Copilot code generation metrics now available in the enterprise insights dashboard
Enterprise teams limits increased over 10x — up to 2,500 teams and 5,000 users per team
Dependabot-based dependency graphs for Go now provide more complete transitive dependency trees
npm classic tokens revoked — replaced with session-based and CLI token management
Repository custom properties now support GraphQL management and a URL type with built-in validation
GitHub MCP server adds tool-specific configuration, lockdown mode for untrusted contributors, and default content sanitization against prompt injection
GitHub Enterprise Server 3.19 GA with rule set history import/export, SSH/TLS cipher configuration, and OpenTelemetry metrics
Auto model selection GA in VS Code for all Copilot plans
VS Code December Release — Agent Features
Agent HQ for managing multiple coding agents — background, cloud, or local — working simultaneously
Background agents with git worktrees for isolated workspaces, enabling multiple agents to work in parallel without file conflicts
Custom agents for organizations — share agents at the org level via .github-private repositories
Sub-agents via run sub agent — delegate tasks to specialized sub-agents with their own context windows to avoid context bloat
Claude skills support — reuse existing Claude Code skills within VS Code agents
Session management improvements — integrated sessions view, compact and side-by-side layouts, and persistent local agent sessions
Live Coding — Custom Agents and Background Agents
Chris demonstrates the new VS Code agent features by continuing work on a tic-tac-toe game with an MCP server backend. He creates a testing specialist agent using a TDD workflow, experiments with background agents running in git worktrees, and explores sub-agent delegation for specialized tasks like code quality review.
