using-jules

Jules is Google's asynchronous coding agent that clones your repo into a secure VM, executes tasks, and returns diffs or PRs. It works in the background while you continue other work.

CLI Quick Reference

# Launch TUI
jules

# Create session for current repo
jules new "write unit tests for auth module"

# Create session for specific repo
jules new --repo owner/repo "add input validation"

# Parallel sessions (1-5) for same task
jules new --parallel 3 "fix linting errors"

# List sessions and repos
jules remote list --session
jules remote list --repo

# Pull session results
jules remote pull --session 123456
jules remote pull --session 123456 --apply  # Apply patch locally

# Teleport: clone + checkout + apply patch
jules teleport 123456

# Auth
jules login
jules logout

Writing Effective Prompts

Be specific, not vague. Treat it like delegating to a developer.

Bad	Good
"Add authentication"	"Add JWT auth to Express app with login/logout endpoints and middleware for protected routes"
"Fix the bug"	"Fix TypeError in components/LoginForm.js when submitting empty email field - add validation"
"Write tests"	"Write pytest tests for the parse_config function covering edge cases: empty input, malformed JSON, missing required fields"

Include context:

• Specify files/modules to modify
• Mention frameworks and patterns to follow
• Reference existing code patterns when applicable

For complex tasks: Break into smaller, sequential sessions rather than one large request.

Prompt Categories

Debugging: "Help me fix {specific error}", "Trace why this value is undefined", "Add logging to debug {issue}"

Refactoring: "Refactor {file} from {pattern X} to {pattern Y}", "Convert callbacks to async/await", "Extract {logic} into a utility function"

Testing: "Add integration tests for {endpoint}", "Write pytest fixtures mocking {external API}", "Generate property-based tests for {function}"

Documentation: "Add docstrings to {module}", "Write API docs for {endpoint}"

Maintenance: "Upgrade {dependency} and fix breaking changes", "Remove unused imports across the codebase"

Workflow Patterns

Batch processing from file:

cat TODO.md | while IFS= read -r line; do
  jules new "$line"
done

From GitHub issues:

gh issue list --assignee @me --limit 1 --json title | jq -r '.[0].title' | jules new

With Gemini for triage:

gemini -p "find the most tedious issue:\n$(gh issue list --assignee @me)" | jules new

AGENTS.md Integration

Jules reads AGENTS.md from the repo root for context. Use /add-agents-md to generate one.

Keep it updated with:

• Build/test commands (exact, copy-pasteable)
• Architecture overview (one paragraph)
• Coding conventions and lint rules
• Important constraints (what NOT to do)

Session Workflow

1. Create session with specific prompt
2. Jules generates plan - review for understanding
3. Approve or refine the plan
4. Jules executes in isolated VM
5. Review diff/PR when complete
6. Apply locally with jules remote pull --apply or jules teleport