Prerequisites
Before starting, ensure you’ve completed:Step 1: Access AutoDev Dashboard
- Go to app.delino.io
- Select your team from the team switcher
- Navigate to the AutoDev section
- You’ll see the task creation form
Step 2: Select Your Repository
- Click the Repository dropdown
- Select the repository where you want to create a PR
- Only repositories with the AutoDev GitHub App installed will appear
- Verify GitHub App installation
- Check repository permissions
- Refresh the page
Step 3: Write Your Task Prompt
Start with a simple, well-defined task for your first PR:Example 1: Bug Fix
Example 2: Small Feature
Example 3: Refactoring
- Be specific about what to fix or build
- Include file paths using
@
notation - Mention technical requirements
- Specify expected behavior
Step 4: Configure Task Settings
Task Type
For your first PR, leave “Create as Composite Task” unchecked. This creates a simple unit task, perfect for getting started.Advanced Settings (Optional)
AI Agent:- Leave blank to use your team’s default
- Or select a specific agent (e.g., “Claude Code”)
- Leave blank to use the agent’s default
- Or specify a model (e.g., “sonnet”, “gpt-4”)
- Leave blank to use the repository’s default branch
- Or specify a branch (e.g., “develop”, “staging”)
Step 5: Create the Task
- Review your prompt
- Check repository selection
- Click Create Task
- You’ll be redirected to the task details page
Step 6: Monitor Progress
The task details page shows:Initial Status: Pending
- Task is created and queued
- GitHub Actions workflow is being triggered
- Usually completes in seconds
Status: In Progress
- AI agent is analyzing your request
- Code is being written
- Pull requests are being created
- This can take 2-10 minutes depending on complexity
Status: Completed
- All work is done
- Pull requests have been created
- Ready for review
Step 7: Review the Generated PR
Once completed:- Click on the PR in the task details page
- Or navigate to GitHub and find the PR
- Review the changes:
- Check code quality
- Verify it meets requirements
- Look for any issues
What to expect
On GitHub PR page:- AutoDev task link comment
- Task prompt comment (shows what AI was asked to do)
- Code changes in files
- Automated tests running (CI checks)
- Code follows your project’s style
- Tests are passing
- Changes are focused and relevant
- Documentation is updated if needed
- Code doesn’t compile
- Tests are failing
- Changes are unrelated to the task
- Security issues
Step 8: Provide Feedback (Optional)
If changes are needed:Option 1: GitHub Review Comments
- Leave review comments on the PR
- Click “Request changes” or “Comment”
- AutoDev automatically detects the review
- AI agent makes updates based on your feedback
- PR is updated with new changes
Option 2: Manual Update Request
- Go to task details page
- Click Update PR
- Enter update instructions
- AutoDev triggers update workflow
- PR is updated with changes
Step 9: Merge the PR
When you’re satisfied:- Ensure all CI checks pass
- Get required approvals from team
- Merge the PR on GitHub
- Delete the branch (optional)
- Task status updates to “Completed”
Common First-Time Issues
AutoDev not responding to reviews
Causes:autodev.yml
file missing or incorrect- Agent not configured or API key missing
- Verify workflow file exists at
.github/workflows/autodev.yml
- Check file contents match the setup guide
- Ensure workflow has correct permissions
- Verify AI agent is set up correctly
- Check API key in repository secrets
- Review agent configuration guide
PR not created
Cause: Task completed but no branches were needed Solution:- Check task prompt - was it clear what to create?
- Review workflow execution logs
- Task may have determined no changes were needed
CI checks failing
Cause: Generated code has issues Solution:- AutoDev automatically tries to fix CI failures
- Wait for auto-fix workflow to complete
- If it fails again, add review comments with guidance
Next Steps
Now that you’ve created your first PR:Try more complex tasks
Learn about PR management
Optimize your workflow
Advanced features
- Try creating a composite task for complex projects
- Enable auto-approval for trusted workflows
- Configure custom CI fix strategies
Tips for Success
Start simple
- Begin with small, well-defined tasks
- Learn how the AI interprets your prompts
- Gradually increase complexity
Iterate on prompts
- Refine your prompt writing based on results
- Be more specific if output is too broad
- Add constraints for better control
Review carefully
- Always review AI-generated code
- Don’t merge without understanding changes
- Treat it like code from a junior developer
Collaborate with AI
- Use AutoDev for repetitive tasks
- Let it handle boilerplate code
- Focus your time on complex logic and architecture
Provide feedback
- Review comments help the AI improve
- Be specific about what’s wrong
- Guide the AI toward better solutions
Example: Complete First Task
Let’s walk through a complete example: Prompt:- Task created with auto-generated title: “Add input validation to user registration form”
- GitHub Actions workflow triggered
- AI agent analyzes the codebase
- Finds the registration form component
- Adds validation logic
- Adds error message display
- Writes tests for validation
- Creates PR with all changes
- Frontend: Validation logic in registration component
- Frontend: Error message UI components
- Tests: Validation test cases
- Total: ~100-200 lines across 3-4 files
- Check validation logic is correct
- Verify error messages are user-friendly
- Ensure tests cover edge cases
- Confirm no breaking changes
- PR merged
- Feature deployed
- Task completed