Getting Started with DevLoop Runner - Auto-generate PRs from GitHub Issues

What is DevLoop Runner?

DevLoop Runner is an AI-powered development automation service that automatically converts GitHub Issues into Pull Requests. It enables a semi-automated workflow for developers where humans set direction through Issues while AI handles the implementation.

Setup

1. Register GitHub Token

First, register your GitHub Personal Access Token in the Settings page.

1. Go to Settings > GitHub Token
2. Enter your Personal Access Token
3. Set a name and email address for commits

Note: For private repositories, the token needs repo scope access.

2. Configure AI Credentials

Configure at least one of the following AI credentials:

• OpenAI API Key or Codex auth.json
• Claude Credentials or Anthropic API Key

Tip: Configuring both OpenAI/Codex and Claude/Anthropic enables "Optimized Mode" with automatic failover.

Job Execution

Dev Run (Main Feature)

1. Navigate to Run Job page
2. Enter a GitHub Issue URL
3. Select execution mode:
   • All Phases: Planning + Implementation + PR creation
   • Planning Only: Generate plan and design
   • Implementation Only: Generate code based on existing plan
4. Click "Start Dev Run"

Issue Run (Repository Analysis)

1. Select the Issue Run tab
2. Enter the GitHub repository (owner/repo format)
3. Choose a category:
   • Bug Detection: Find bugs and potential issues
   • Refactoring: Identify refactoring opportunities
   • Enhancement: Suggest new features
4. Set the maximum number of issues to generate (1-5)

Monitoring Jobs

• View all your jobs on the Jobs dashboard
• Click any job to see detailed progress
• Phase timeline shows completion status of each step
• Cost display tracks AI token usage

Finalizing PRs

When a job succeeds:

1. Review the draft PR on GitHub
2. Click Finalize on the job detail page
3. This will:
   • Clean up intermediate files
   • Squash commit history
   • Convert draft PR to ready for review

Rollback

If something goes wrong:

1. Click Rollback on the job detail page
2. Choose between:
   • Auto Mode: AI determines optimal rollback point
   • Manual Mode: Specify phase and reason

Best Practices

Writing Effective Issues

• Write clear acceptance criteria
• Include specific requirements
• Provide context and constraints
• Reference related documentation

Execution Strategy

1. Start with "Planning Only" to review the AI plan
2. If the plan looks good, proceed with "Implementation Only"
3. For straightforward tasks, use "All Phases"

---

Have questions? Check the documentation or contact support.