Rubber Duck Thursdays!

Key Concepts

• GitHub Copilot CLI: A command-line interface tool for interacting with Copilot, distinct from the standard Copilot extension.
• Agentic Skills: Custom configurations, instructions, and tools (e.g., documentation writers, accessibility checkers) that can be added to Copilot.
• Plan Mode: A feature in the Copilot CLI that allows users to outline project requirements, which the AI then uses to generate a structured implementation plan.
• Autopilot Mode: A mode where the AI executes the generated plan automatically, including installing dependencies and writing code.
• Remote vs. Delegate: "Remote" allows controlling a local CLI session from other devices; "Delegate" offloads the session entirely to the cloud, allowing the local machine to be shut down.
• Generally Available (GA): A software release status indicating a product is no longer in preview and is ready for public use.
• TUI (Terminal User Interface): Libraries (like OpenTui) used to create interactive, visually appealing interfaces within the terminal.

1. Recent GitHub Releases & Updates

• GitHub CLI Skill Management: Users can now manage agentic skills via gh skill. The awesome-copilot repository serves as a community-driven hub for these custom agents.
• Rules Insights Dashboard: A new platform feature for repository rule sets that allows teams to visualize how rules are configured and identify where they are being bypassed.
• Claude Opus 4.7: Now generally available, this model is being integrated into the Copilot model picker, replacing older versions (4.5/4.6) for Copilot Pro users.
• Code Scanning Integration: Code scanning alerts can now be directly linked to GitHub Issues, streamlining the remediation process.
• Notification Sorting: A long-requested feature allowing users to sort notifications by date (newest to oldest or vice versa).
• Merge Conflict Resolution: Copilot can now assist in fixing merge conflicts, significantly reducing manual effort in complex codebases.

2. Methodologies & Frameworks

• The "Plan-Build" Workflow:
  1. Initiation: Use shift + tab to enter Plan Mode.
  2. Contextualization: Provide a natural language prompt. The AI analyzes the repository context and suggests a technical stack.
  3. Refinement: The user can suggest changes (e.g., "Use OpenTui instead of Chalk") to the plan.
  4. Execution: Use Autopilot Mode with the allow all flag to automate the installation of dependencies (e.g., bun, OpenTui) and the generation of code.
• Session Management: The CLI supports keep alive to prevent system sleep during long tasks and delegate to move active sessions to the cloud.

3. Real-World Application: Emoji List Generator

The host demonstrated building a bespoke CLI tool to convert bulleted lists into emoji-prefixed markdown.

• Tech Stack: TypeScript, Bun (runtime), OpenTui (UI), and the GitHub Copilot SDK.
• Process:
  • Used Claude Sonnet for planning and Claude Opus 4.7 for implementation.
  • Addressed technical hurdles (e.g., Bun-exclusive dependencies) by allowing the AI to adapt the plan dynamically.
  • Used OpenTui to create a clean, bordered terminal interface.
  • Implemented a "suppress warnings" instruction to clean up the UI output from experimental SQLite features.

4. Key Arguments & Perspectives

• The Changing Role of Developers: The host argues that AI will not replace software engineers but will fundamentally change the role. Developers are becoming "carpenters" who use AI to build bespoke, highly specific tools quickly.
• Technical Debt: AI accelerates development, but it also accelerates the accumulation of technical debt if proper guardrails and code review processes (like Copilot Code Review) are not utilized.
• The Value of "Handmade": While AI is powerful, there remains a high value in maintaining personal, handcrafted projects, even if AI is used to assist with painful implementation tasks.

5. Notable Quotes

• "If AI could do my laundry, then I'm a believer. So far that has not happened, but this [fixing merge conflicts] gets close." — Cassie, regarding the utility of AI in mundane tasks.
• "I like that LLMs are turning us into carpenters, allowing us to build our own bespoke little tools." — A sentiment shared by the host regarding the democratization of custom software creation.

6. Synthesis & Conclusion

The stream highlighted a shift toward agentic development, where the GitHub Copilot CLI acts as an autonomous partner capable of planning, installing, and executing complex tasks. The key takeaway is the importance of mastering the Plan Mode and Model Selection (e.g., using Haiku for speed, Opus for complex logic) to optimize the development lifecycle. The host emphasized that while these tools are powerful, they are not "silver bullets" and require human oversight to manage technical debt and ensure quality.