Jules Test

This repository serves as a testing ground for the Antigravity (Planner) x Jules (Coder) collaboration model.

🤝 Antigravity x Jules Collaboration Workflow

This project established a standardized protocol for integrating AI Agents into the development lifecycle.

Antigravity (The Planner):
- Analyzes user requirements.
- Drafts structured Task Specifications (JULES_TASK_*.md).
- Dispatches tasks to Jules via CLI.
- Verifies final artifacts.
Jules (The Coder):
- Receives the Spec.
- Executes coding tasks in an asynchronous session.
- Generates a Pull Request (PR) for review.
Human (The Gatekeeper):
- Critical Action: Must manually "Publish PR" in the Jules Console.
- Authorizes Git credentials.
- Approves and Merges the final PR.

Spec Generation: Antigravity creates a task-specific markdown file based on JULES_PROTOCOL.md.
Dispatch: Task is sent using gemini -p "/jules ... $(cat SPEC.md)".
Execution: Jules works in the background (visible in Jules Console).
Publishing (Human Intervention): Jules will stop at "Ready for review". User MUST click "Publish PR" to push code to GitHub.
Verification: Antigravity pulls the branch, verifies functionality, and merges.

⚠️ Empty Repository: Jules cannot initialize on an empty repo. Always ensure an initial commit exists before dispatching tasks.
⚠️ Explicit Authorization: New repositories must be manually checked in Jules Console > Settings > Repository Access.
⚠️ The "Publish" Step: This is the most common blocker. Jules does not auto-push to GitHub. If you don't see the PR, check the Console and hit Publish.
✅ Structured Specs: Using a defined template (JULES_PROTOCOL.md) yielded 100% success rate, whereas loose natural language prompts failed or had ambiguity. Independent, self-contained specs are best.
🔒 Security First: Before making any repo Public, always run a grep scan for keywords like key, token, secret to ensure no AI-generated credentials are leaked.