Jules Actions

Invoke a powerful remote coding agent using GitHub Actions

Get Started • Examples • API Docs

What is Jules?

Jules is a remote AI coding agent from Google Labs powered by Gemini 3 Pro. It works autonomously in a cloud VM—analyzing your codebase, implementing features, fixing bugs, and creating pull requests while you focus on what matters.

This action lets you trigger Jules from any GitHub event: issues, pull requests, schedules, or workflow dispatches.

Check out our example workflows for ideas on using Jules workflows to automatically fix bugs, improve performance, and more!

✨ Quick Start

1. Get a Jules API Key

1. Authenticate with GitHub at jules.google.com
2. Generate an API key from your account settings
3. See API docs for details

2. Add to GitHub Secrets

1. Go to Settings → Secrets and variables → Actions
2. Click New repository secret
3. Name: JULES_API_KEY, Value: your key

3. Create a Workflow

Create .github/workflows/security-agent.yml:

name: Daily Security Scan

on:
  schedule:
    - cron: '0 6 * * *'  # Every day at 6 AM
  workflow_dispatch:

jobs:
  scan:
    runs-on: ubuntu-latest
    permissions:
      contents: read
    steps:
      - uses: google-labs-code/jules-invoke@v1
        with:
          prompt: |
            You are a security agent. Scan for vulnerabilities and fix them:

            CRITICAL (fix immediately):
            - Hardcoded secrets, API keys, passwords
            - SQL injection, command injection
            - Missing auth on sensitive endpoints

            HIGH PRIORITY:
            - XSS, CSRF vulnerabilities
            - Insecure direct object references
            - Missing input validation

            Rules:
            - Fix highest severity first
            - Keep changes under 100 lines
            - Run tests before creating PR
          jules_api_key: ${{ secrets.JULES_API_KEY }}

Your repo now has a security agent hunting vulnerabilities daily! 🛡️

---

📋 Inputs

Input	Description	Default
prompt	Required. The task for Jules to perform.	—
jules_api_key	Required. Your Jules API key (use secrets).	—
starting_branch	Branch for Jules to start from.	main
include_last_commit	Include the last commit's diff in context.	false
include_commit_log	Include recent commit history in context.	false

---

🚀 Example Workflows

Copy these into .github/workflows/ and customize:

Example	Trigger	Description
weekly-cleanup	Scheduled (cron)	Automated code maintenance and refactoring
performance-improver	Scheduled (daily)	⚡ Hunt and fix performance bottlenecks
bug-fixer	Issue labeled bug	Diagnose and fix bugs (or implement features) with structured prompts
ci-failure-fix	CI workflow fails	Automatically fix failed builds
unblocked-issues	Issue closed	Work on issues that were blocked by the closed issue

See examples/README.md for detailed setup instructions.

---

💡 Writing Good Prompts

Jules excels when you give it clear specifications or measurable targets it can verify. This is especially powerful for scheduled workflows that run continuously.

Example: Weekly Performance Agent

name: Performance Improvement Agent
on:
  schedule:
    - cron: '0 3 * * 1'  # Every Monday at 3 AM

jobs:
  optimize:
    runs-on: ubuntu-latest
    steps:
      - uses: google-labs-code/jules-invoke@v1
        with:
          prompt: |
            You are a performance improvement agent. Run our benchmark suite
            and look for optimization opportunities:

            1. Run `npm run bench` to get current metrics
            2. Identify the slowest endpoints or functions
            3. Implement optimizations (caching, query improvements, algorithm changes)
            4. Re-run benchmarks to verify improvement
            5. Only open a PR if you achieve measurable gains

            Constraints:
            - Don't change public API contracts
            - All existing tests must pass
          jules_api_key: ${{ secrets.JULES_API_KEY }}

Why this works: Jules can run benchmarks, iterate on solutions, and verify its own success. Give Jules a target it can measure, and it becomes a continuous improvement loop.

More ideas for scheduled agents:

• Security audit agent (run npm audit, fix vulnerabilities)
• Dependency updater (update deps, run tests, PR if green)
• Docs updater agent (update docs based on recent changes)
• Test coverage improver (find gaps, add tests, target 90%)

---

🔐 Security

⚠️ Important: Issue-triggered workflows can be exploited by untrusted users opening issues. Always restrict who can trigger Jules for sources like GitHub issues. For things like issues, oftentimes anyone can create an issue so it can be handy to have an allowlist condition for Jules triggering.

Add an allowlist condition to your step:

- name: Invoke Jules
  # Only runs if user is in the allowlist
  if: ${{ contains(fromJSON('["trusted-user", "another-user"]'), github.event.issue.user.login) }}
  uses: google-labs-code/jules-invoke@v1
  with:
    prompt: ...

Best practices:

• Never commit JULES_API_KEY—always use GitHub Actions secrets
• Treat Jules like any team member: review its PRs before merging

---

📚 Resources

• Jules Homepage: jules.google
• Jules Web App: jules.google.com
• API Documentation: jules.google/docs/api/reference
• GitHub Actions: docs.github.com/actions

---

Attribution

If you find this useful, add the badge to your README:

[![Built with Jules](https://img.shields.io/badge/Built%20with-Jules-715cd7?link=https://jules.google)](https://jules.google)

---

Built with ❤️ by Google

Jules • API Docs • Report Issues