Automating PR Descriptions with an AI Agent: A Two-Step Workflow

Writing clear PR descriptions is tedious, but AI can do it in seconds. In this tutorial, you'll build a simple, repeatable workflow that extracts the diff, sends it to an LLM with a focused prompt, and post the result as a PR description. You'll use GitHub CLI and a Node.js script — no complex frameworks required.

Step 1: Collect Context with GitHub CLI

First, get the diff and commit list. The GitHub CLI (gh) makes this trivial. Create a file called collect-context.sh:

#!/bin/bash
BASE_BRANCH="main"
DIFF=$(gh pr diff --repo YOUR_OWNER/YOUR_REPO)
COMMITS=$(gh pr view --json commits --jq '.commits[].messageHeadline')

{
  echo "=== DIFF ==="
  echo "$DIFF" | head -c 12000   # trim to avoid token limits
  echo ""
  echo "=== COMMITS ==="
  echo "$COMMITS"
} > pr-context.txt

Make it executable (chmod +x collect-context.sh) and run it against an open PR. This gives you a clean file with the diff and commit subjects — exactly what the AI needs.

Step 2: Craft the Prompt

Now, the magic. Create generate-description.mjs that reads that file and calls your model. Here's a Node.js version using fetch (Node 18+) so you don't need extra libraries:

import fs from 'fs';

const context = fs.readFileSync('pr-context.txt', 'utf8');

const prompt = `You are a senior engineer. Write a concise PR description from the diff and commits.
Use this format:
## Summary
## Changes
- bullet list
## Test Plan
(if obvious from diff, otherwise say "Manual testing")

Context:
${context}
`;

const res = await fetch("https://api.openai.com/v1/chat/completions", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "Authorization": `Bearer ${process.env.OPENAI_API_KEY}`
  },
  body: JSON.stringify({
    model: "gpt-4o-mini",
    messages: [
      { role: "system", content: "You write clear, technical PR descriptions." },
      { role: "user", content: prompt }
    ],
    temperature: 0.3
  })
});

const data = await res.json();
const description = data.choices[0].message.content;
fs.writeFileSync('pr-description.md', description);
console.log('Done! Saved to pr-description.md');

Step 3: Punch It Into the PR

Finally, take the generated description and update the PR body:

gh pr edit --body-file pr-description.md

That's it. Run the three steps in sequence and you'll have a polished description.

Make It One Command

Wrap it in a single command using &&:

./collect-context.sh && node generate-description.mjs && gh pr edit --body-file pr-description.md

Add that to your shell aliases or npm scripts for daily use.

Taking It Further

This workflow is deliberately simple, but you can extend it:

• Use structured outputs to force the model to return JSON (summary, changes, test_plan) for easier post-processing.
• Add a step to check for merge conflicts and include that in the PR description.
• Trigger it automatically with a GitHub Action on every new PR.