GitHub's December 2025 - January 2026: The Ships That Matter

#webdev #ai #github #git

If you took the last weeks of 2025 off (good for you!), you missed some significant launches. Here's what actually matters for your day-to-day work.

The Big Three

Copilot Skills: Teaching the Agent Your Way

You can now teach Copilot how to perform tasks your way. Skills are folders containing instructions, scripts, and resources that Copilot automatically loads when relevant to your prompt.

Create a .claude/skills directory in your repo. Add a folder with a SKILL.md file containing your instructions. When Copilot determines the skill is relevant, it follows those instructions.

# Example structure
.claude/skills/
  ├── api-tests/
  │   └── SKILL.md
  └── error-handling/
      └── SKILL.md

Every team has patterns. The way you structure tests. How you handle errors. Your API conventions. Instead of repeating these instructions every time, you codify them once as skills.

Works in Copilot coding agent, Copilot CLI, and VS Code Insiders agent mode right now. Stable VS Code support arrives early January.

Check out anthropics/skills and github/awesome-copilot for community-created skills you can use today.

Model Wars: GPT-5.2, Claude Opus 4.5, and Gemini 3 Flash

Three major models launched within 24 hours.

GPT-5.2 went generally available across all Copilot plans. You can access it in github.com, Mobile, VS Code, Visual Studio, JetBrains, Xcode, and Eclipse. Business and Enterprise admins need to opt-in first.

Claude Opus 4.5 moved from preview to GA. It cut token usage in half while surpassing internal benchmarks. Same availability as GPT-5.2.

Gemini 3 Flash hit public preview. It's optimized for speed. You can bring your own key in VS Code. Rolling out gradually to Pro, Pro+, Business, and Enterprise.

You're not locked into one model anymore. GPT-5.2 excels at complex reasoning. Claude Opus handles long-form code better. Gemini Flash gives you rapid responses when you need them.

Pick the right tool for the task at hand.

Copilot Memory: Context That Persists

Public preview for Pro and Pro+ users. Copilot now builds repository-specific memory by capturing key insights about your codebase and using them to improve how it assists you.

Not your entire codebase. Insights. Architectural decisions. Patterns.

Enable it in Settings > Copilot > Toggle "Copilot memory" ON.

It works in Copilot coding agent and code review workflows. Memory is repository-specific and doesn't share across repos. You can disable it anytime.

# Before Memory: Generic suggestions
# After Memory: Suggestions that match YOUR patterns

# Your codebase always uses this pattern:
class ServiceBase:
    def __init__(self, config):
        self.logger = setup_logger(__name__)
        self.config = config

# With Memory, Copilot suggests new services 
# using this EXACT pattern automatically

GitHub Actions: The Pricing Drama

GitHub announced two things on December 16:

20-39% price reduction on hosted runners
New $0.002/minute charge for self-hosted runners (planned for March 2026)

The community said no. Loudly.

GitHub postponed the self-hosted charge indefinitely. The hosted runner price cuts went live January 1 as planned.

Current state:

Hosted runners: 20-39% cheaper (in effect now)
Self-hosted runners: Still free
Public repos: Still free

Performance and Tooling Updates

Actions Workflows Page got a major upgrade. Workflows with 300+ jobs now render successfully with lazy loading. You can filter jobs by status (failed, in progress, etc.). If you're running complex workflows or large monorepo setups, this update removes a real pain point.

C++ Code Editing Tools hit public preview in Visual Studio 2026 Insiders. Deep symbol awareness for reliable multi-file editing. Copilot can now view references across your codebase, understand symbol metadata, visualize class hierarchies, and trace function call chains. If you write C++, this changes how Copilot understands your code at scale.

Copilot Code Review for Enterprise Cloud now supports the agentic features launched at Universe:

Rich agentic tool calling for full project context
Integration with CodeQL, ESLint, and other deterministic tools
Direct handoff to coding agent from reviews
Mission control visibility into agent sessions

The Quick Hits

When you assign an issue to Copilot, you're automatically added as an assignee too. Makes filtering with assignee:@me actually useful.

GitHub Docs now has dynamic Copilot prompts. Launch prefilled prompts directly from documentation. No more copying and pasting.

Secret scanning added 24 new secret types from Azure, Databricks, Microsoft, Paddle, PostHog, and more. Improved private key detection for Elliptic Curve and PKCS#8 formats.

Visual Studio 2026 went GA with GitHub cloud agent in public preview. Delegate repetitive tasks directly from Visual Studio.

What to Do Next

If you're on Copilot Pro/Pro+, enable Copilot Memory in Settings. Start building skills for your team's patterns.

If you're on Business/Enterprise, check your admin settings. Enable the new models (GPT-5.2, Claude Opus 4.5, Gemini 3 Flash) if you want your team to access them.

If you use GitHub Actions, your hosted runner costs dropped on January 1. Check your billing. If you use self-hosted runners, you're good for now.

If you write C++, try the code editing tools in Visual Studio 2026 Insiders.

The Bottom Line

December gave us personalization (Skills), choice (multiple models), and persistence (Memory). GitHub is betting that the future of AI-assisted development isn't about one model doing everything. It's about giving developers the right tools to teach AI how they work and choosing the best model for each task. I'm here for it!

The fundamentals still matter. AI gets better when you give it better context. Skills and Memory are both about making that context explicit and reusable.

Links: