The latest articles on Forem by Rhythm Dubey (@rhythm_dubey_92c43317b6b3). https://forem.com/rhythm_dubey_92c43317b6b3 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3744806%2F70ac0df4-7a3c-47b0-a582-aa5a67a3e740.jpg https://forem.com/rhythm_dubey_92c43317b6b3 en Rhythm Dubey Fri, 13 Feb 2026 16:10:34 +0000 https://forem.com/rhythm_dubey_92c43317b6b3/building-a-terminal-based-ai-pr-reviewer-with-github-copilot-cli-36fe https://forem.com/rhythm_dubey_92c43317b6b3/building-a-terminal-based-ai-pr-reviewer-with-github-copilot-cli-36fe <p><em>This is a submission for the <a href="https://dev.to/challenges/github-2026-01-21">GitHub Copilot CLI Challenge</a></em></p> <p>Code reviews usually happen after a pull request — I wanted to bring them directly into the terminal before pushing code.</p> <h2> What I Built </h2> <p>I built a terminal-based AI PR Reviewer that uses GitHub Copilot to analyze Python files before they’re committed or pushed.</p> <p>The idea was simple: instead of waiting for a GitHub PR review, what if I could run an AI-powered review directly in my terminal?</p> <p>The tool:</p> <ul> <li>Detects staged or modified Python files</li> <li>Sends them to Copilot for structured review</li> <li>Enforces a strict JSON output format</li> <li>Displays issues in a clean, color-coded terminal table</li> <li>Optionally generates a PR-ready markdown comment</li> </ul> <p>It’s lightweight, focused, and designed to fit naturally into a developer’s workflow.</p> <p>I wanted something practical — not just a demo, but something I could actually use before pushing code.</p> <h2> Demo </h2> <p><a href="https://github.com/rhythm-dubey/ai-pr-reviewer" rel="noopener noreferrer">https://github.com/rhythm-dubey/ai-pr-reviewer</a></p> <p>You can review staged files with:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code>python reviewer.py <span class="nt">--staged</span> </code></pre> </div> <p>The tool analyzes changed Python files and categorizes issues into:</p> <ul> <li>🔴 Critical</li> <li>🟡 Warning</li> <li>⚪ Suggestion</li> </ul> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F3ye0wv0hn4h9ca5id0dw.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F3ye0wv0hn4h9ca5id0dw.png" alt=" " width="800" height="491"></a></p> <h2> My Experience with GitHub Copilot CLI </h2> <p>This project pushed me to explore Copilot beyond inline code suggestions.</p> <p>Instead of using Copilot just inside an editor, I used it programmatically to:</p> <ul> <li>Create structured review prompts</li> <li>Enforce JSON output formats</li> <li>Handle inconsistent model responses</li> <li>Debug parsing failures</li> </ul> <p>One of the biggest challenges was ensuring the AI returned strict JSON consistently. I had to carefully design the prompt to strongly enforce JSON output, and even then I implemented fallback parsing and error handling to avoid breaking the CLI experience.</p> <p>That forced me to think more deeply about prompt design, validation, and defensive programming around unpredictable model outputs.</p> <p>I also learned that AI integration isn’t just about calling a model — it’s about building reliable guardrails around unpredictable outputs.</p> <p>Overall, working with Copilot CLI made experimentation fast. It allowed me to prototype quickly, iterate on prompts, and test different review strategies directly from the terminal.</p> <p>This project started as a challenge entry, but it gave me practical insight into building AI-assisted developer tooling.</p> devchallenge githubchallenge cli githubcopilot