Rubber Duck Thursdays - Let's build with GitHub Copilot CLI

Key Concepts

• GitHub Copilot CLI & SDK: Recent updates to the Copilot CLI, integrated with the gh CLI, and the release of the Copilot SDK (Node.js/TypeScript, Python, Go, .NET) enable developers to build applications leveraging Copilot’s AI capabilities.
• Plan Mode: A new conversational mode within the Copilot CLI for more agent-like task planning and execution, involving clarifying questions and iterative refinement.
• Context Engineering: Prioritizing effective context window management over prompt engineering for optimal AI performance.
• Personalized Learning Application: A custom application built with the Copilot SDK demonstrates personalized coding challenges based on user profiles and skill sets.
• AI-Powered Code Evaluation: The application utilizes Copilot to evaluate code submissions, identifying missing components and potential errors.

GitHub Copilot CLI & SDK Overview

The recent updates to the GitHub Copilot CLI and the introduction of the Copilot SDK are central to expanding Copilot’s functionality beyond the IDE. The Copilot CLI is now accessible via gh copilot install, integrating seamlessly with the standard GitHub CLI. The Copilot SDK, available for Node.js/TypeScript, Python, Go, and .NET, provides a wrapper around the CLI, allowing developers to build custom applications. A key principle highlighted is a shift from “prompt engineering” to “context engineering,” emphasizing the importance of providing targeted, relevant information to the AI rather than relying on complex prompts. The underlying platform powering Copilot is the Microsoft Copilot Platform (MCP).

Plan Mode & Advanced Reasoning

A significant new feature is “Plan Mode” (accessed via /plan), which moves beyond simple question-answer interactions. This mode allows Copilot to act more like an agent, asking clarifying questions and iterating on requirements to achieve a desired outcome. The session emphasized the ability to configure reasoning effort and leverage advanced reasoning models. A convenience flag, “YOLO Mode,” allows for automatic approval of all Copilot actions, but is cautioned against for general use.

Application Showcase: Personalized Learning

A custom application built using the Copilot SDK was demonstrated. This application focuses on providing personalized coding challenges tailored to a user’s skill profile and recent activity. The application leverages the Copilot CLI via the SDK, sending prompts and receiving responses for a chat-like experience and AI-powered code evaluation. User profiles can specify preferred languages (e.g., Go) and areas of interest (e.g., CI/CD, Kubernetes), influencing challenge generation.

The application demonstrated a challenge involving containerizing a TypeScript countdown timer and running it on Kubernetes. The AI evaluation correctly identified missing Docker build instructions and a deployment solution. A subsequent challenge, after updating the user profile to prioritize Go, recommended building a Go CLI countdown timer with CI integration, showcasing the application’s adaptability.

Development & Workflow Insights

The presenter showcased a practical workflow using the Copilot CLI to review a pull request for a community-contributed theme ("square shadow") for a countdown timer application. Debug mode reveals the underlying calls made to the Copilot CLI, showing the prompts and responses exchanged during challenge generation and evaluation, providing insight into the AI’s reasoning process. The application includes a sandbox environment with tabbed file editing, although temporary issues with syntax highlighting were noted. The presenter highlighted the rapid development cycle, having worked on the application in “typical development time hours” over recent days.

Community & Future Development

The session actively engaged with the audience, showcasing a community-contributed theme for the countdown timer application and emphasizing the value of collaborative development. The presenter acknowledged ongoing development and bug fixing, including issues with syntax highlighting, incorrect file type suggestions (TypeScript instead of Go), and an empty repository error during GitHub export. The presenter encouraged viewers to explore the Copilot CLI and SDK and share their creations, providing links to the respective GitHub repositories: https://github.com/github/copilot-cli and https://github.com/githubcopilot-SDK.

Conclusion

The GitHub Copilot CLI and SDK represent a significant step towards extending Copilot’s capabilities beyond the IDE, enabling developers to build custom applications leveraging its AI power. The introduction of Plan Mode and the emphasis on context engineering promise more sophisticated and effective interactions with the AI. The demonstrated personalized learning application highlights the potential for creating tailored developer tools that adapt to individual skill sets and project needs. While acknowledging ongoing development and bug fixes, the session conveyed a strong sense of excitement and potential for the future of Copilot-powered development.