DEV Community

Shivay Lamba
Shivay Lamba

Posted on

How I built a Content Curator CLI using Github Copilot CLI SDK

#devchallenge #githubchallenge #cli #githubcopilot

GitHub Copilot CLI Challenge Submission

This is a submission for the GitHub Copilot CLI Challenge

What I Built

Building Content Curator: CLI + Copilot SDK

I wanted a simple but practical project to test out the Copilot SDK. I ended up creating an AI-powered CLI that uses Copilot’s agentic core along with real-time web search (via Exa AI) to generate short-form video ideas, hooks, and full scripts for Reels, YouTube Shorts, and TikTok.

Repo: github-copilot-cli-sdk-content-curator

How Content Curator Uses GitHub Copilot SDK & Exa AI

1. Initializing the Copilot Client

At the heart of the app is the CopilotService class, which manages the AI client lifecycle:

this.client = new CopilotClient(clientOptions);
await this.client.start();
await this.createSession();
Enter fullscreen mode Exit fullscreen mode
  • CopilotClient connects directly to Copilot models installed locally via the CLI.
  • client.start() initializes the client, preparing it for session creation.
  • createSession() sets up a streaming session with the chosen model and system prompt.

2. Session Management

Each AI session represents a continuous conversation with the Copilot model:

this.session = await this.client!.createSession({
  model: this.currentModel,
  streaming: true,
  systemMessage: { mode: "append", content: SYSTEM_PROMPT_BASE },
});
Enter fullscreen mode Exit fullscreen mode
  • model – Determines which Copilot model (GPT-4o, GPT-5, Claude, etc.) powers content generation.
  • streaming: true – Enables partial outputs to be sent in real time, allowing the CLI to render content as it’s being generated.
  • systemMessage – Provides base instructions for the agent, ensuring that generated content matches the expected format for short-form videos.

3. Prompt Workflow

Content Curator sends structured prompts to Copilot for:

  • Content generation (/script, /ideas, /hooks)
  • Content refinement (/refine)
  • Generating variations (/more-variations)

Example:

const promptWithSearch = `${searchResults}

---

${basePrompt}`;

await this.sendPrompt(promptWithSearch, contentType);
Enter fullscreen mode Exit fullscreen mode
  • Incorporates real-time search results from Exa AI to ensure generated content is relevant and up-to-date.
  • Supports topic- and platform-specific instructions, so scripts are optimized for Instagram, TikTok, or YouTube Shorts.

Using CopilotClient directly allows Content Curator to:

  • Maintain persistent AI sessions with context across multiple commands.
  • Stream outputs in real time, providing an interactive user experience.
  • Dynamically switch models without restarting the application.

Why Choose Copilot SDK Over Generic LLM Clients?

There are several strong reasons to prefer the Copilot SDK over generic LLM clients:

  • GitHub-native workflows – Direct access to repos, files, and workflows.
  • Built-in agentic behavior – Conversations, tools, reasoning, and MCP support.

The kinds of apps you can build with the Copilot SDK go well beyond content curation:

  • Summarizing PRs and generating release notes
  • Creating learning-path threads from a repository
  • Querying personal knowledge stores
  • Building CLIs and custom GUIs for AI agents

Conclusion

The Copilot SDK exposes the agentic capabilities of the Copilot CLI—including planning, tool execution, and multi-turn workflows—directly in your preferred programming language. This allows you to integrate Copilot into any environment, whether you’re building:

  • GUIs with AI workflows
  • Personal productivity tools
  • Custom internal agents for enterprise processes

Essentially, the SDK acts as an execution platform. It provides the same agentic loop that powers the Copilot CLI, while GitHub manages authentication, models, MCP servers, and streaming sessions.

You remain fully in control of what you build on top of these core capabilities.

Demo

Here's the project link - https://github.com/shivaylamba/github-copilot-cli-sdk-content-curator

Video Walkthrough - https://www.youtube.com/watch?v=znmkkjpntKc

My Experience with GitHub Copilot CLI

I used GitHub Copilot CLI extensively while building this project, primarily as a thinking partner rather than just a code generator. It helped me move faster from intent to implementation by translating natural-language prompts into concrete shell commands, code snippets, and workflow suggestions directly in the terminal.

Copilot CLI was especially useful during early development and iteration. Whether it was scaffolding project structure, generating boilerplate code, debugging errors, or suggesting optimized commands, it reduced context switching between the editor, browser, and documentation. I could stay focused on the problem I was solving instead of searching for syntax or command references.

One of the biggest impacts was how it accelerated experimentation. I could quickly try alternative approaches, validate ideas, and refine implementations by asking follow-up questions in the CLI itself. This made the development process more interactive and exploratory, particularly when working with unfamiliar tools or configurations.

Overall, GitHub Copilot CLI significantly improved my productivity and flow. It didn’t replace my understanding or decision-making, but it acted as a powerful assistant that helped me write, test, and iterate faster—especially in moments where friction would normally slow development down.

Top comments (0)