Back to tips
    Your First Dev Run Tutorial: From Issue to PR

    Your First Dev Run Tutorial: From Issue to PR

    Introduction

    With DevLoop Runner, you enter a GitHub Issue and AI automatically implements the code and creates a PR. But if you're using it for the first time, it might not be clear where to begin.

    This article walks you through the entire flow step by step: connecting your repository, running your first Dev Run, reviewing the PR, and executing the Finalize process.

    Follow along, and you'll have your first PR ready in 20-30 minutes.

    Prerequisites

    Before starting this tutorial, make sure you have:

    • A DevLoop Runner account created
    • A GitHub account
    • Push access to the target repository
    • AI credentials configured (Claude API key or OpenAI API key)

    Step 1: Connect Your Repository

    First, connect a GitHub repository to DevLoop Runner.

    1. Open the DevLoop Runner dashboard
    2. Click "Add Repository"
    3. Select the target repository on the GitHub integration screen
    4. Confirm access permissions and click "Connect"

    Once connected, the repository appears in your repository list.

    Tip: The initial connection requires permissions for repository reading, branch creation, and PR creation.

    Step 2: Create Your First Issue

    Dev Run uses a GitHub Issue as its input. Start small.

    What Makes a Good First Issue

    For your first Dev Run, choose an Issue that meets these criteria:

    CriteriaReason
    Small scope of changeEasy to verify results
    Clear requirementsAI can implement without ambiguity
    Easy to testYou can check the testing phase results
    Low riskYou want to experiment with confidence

    Example Issues

    These types of Issues work well for a first Dev Run.

    Example 1: Adding a Utility Function

    ## Overview
    Add a utility function that formats dates as "YYYY/MM/DD".
    
    ## Requirements
    - Create the function in src/utils/dateFormat.ts
    - Accept a Date object and return a string
    - Return an empty string when null or undefined is passed
    
    ## Tests
    - Normal date conversion
    - null / undefined handling
    - Single-digit padding (e.g., 2024/1/5 -> 2024/01/05)
    

    Example 2: Minor Component Update

    ## Overview
    Add a "Blog" link to the header component's navigation.
    
    ## Target File
    - src/components/Header.tsx
    
    ## Requirements
    - Add to the end of existing navigation links
    - Link destination: /blog
    - Display text: "Blog"
    - Responsive support (add to mobile menu as well)
    

    Create the Issue on GitHub, then copy its URL.

    Step 3: Run Dev Run

    Now it's time to run Dev Run.

    1. Open your repository page in DevLoop Runner
    2. Click "New Job"
    3. Select Dev Run
    4. Paste the Issue URL
    5. Choose an execution mode ("All Phases" recommended for first-timers)
    6. Click "Run"

    About Execution Modes

    ModeDescriptionFirst-Time Recommendation
    All PhasesRuns all 10 phasesRecommended
    Planning OnlyPlanning, Requirements, Design onlyWhen you want to review design first
    Implementation OnlyPlanning, Implementation, Documentation, ReportSmall changes that don't need tests

    For your first run, choose "All Phases" to experience the complete workflow.

    Step 4: Monitor Progress

    Once Dev Run starts, you can monitor progress in real-time on the job detail screen.

    The 10-Phase Flow

    Loading diagram...

    During each phase, you'll see the assigned AI persona's icon and what's being processed.

    PhaseAssigned ToEstimated Time
    PlanningAoi (PM)1-2 min
    RequirementsAoi (PM)1-2 min
    DesignRiku (Tech Lead)2-3 min
    Test ScenarioSumire (QA)1-2 min
    ImplementationRiku (Tech Lead)3-5 min
    Test ImplementationSumire (QA)2-3 min
    TestingSumire (QA)1-2 min
    DocumentationKohaku (Tech Writer)1-2 min
    ReportRiku (Tech Lead)1-2 min
    EvaluationAoi (PM)1-2 min

    Total: approximately 15-25 minutes (varies by Issue complexity)

    What to Watch During Execution

    Artifacts are displayed when each phase completes. Pay special attention to:

    • Planning Phase: Has AI correctly understood the Issue's intent?
    • Design Phase: Are the target files for changes appropriate?
    • Testing Phase: Did all tests pass?

    Step 5: Review the Results

    When all phases complete, the project evaluation result is displayed.

    Understanding Evaluation Results

    VerdictMeaningNext Action
    PASSAll requirements metReview and Finalize
    PASS WITH ISSUESRequirements met with minor concernsCheck issues, then decide
    FAILRequirements not metRevise Issue and re-run

    What to Check

    1. Report: A summary of what changed and how
    2. Test Results: Whether all tests passed
    3. Draft PR: The draft PR created on GitHub

    Step 6: Review the PR

    A draft PR has been created on GitHub. Review it just like any normal PR.

    Review Points

    Code quality:

    • Does the change meet the Issue requirements?
    • Does it follow coding conventions?
    • Is there any unnecessary code?

    Test quality:

    • Are test cases sufficient?
    • Are error cases tested, not just happy paths?
    • Do test names clearly describe what's being tested?

    Documentation:

    • Has documentation been updated?
    • Does the content match the code?

    When Fixes Are Needed

    If you find issues during review, you can:

    • Rollback: Roll back to a specific phase and re-run
    • New Issue: Document the fixes needed in a new Issue and run Dev Run again

    Step 7: Execute Finalize

    When the review is complete and everything looks good, execute Finalize.

    1. Click the "Finalize" button on the job detail screen
    2. Finalize executes
    3. The draft PR is converted to a public PR
    4. Commits are cleaned up

    For more on Finalize, see How to Use Finalize.

    Important: Finalize is a human approval process. AI-generated code is never automatically merged. Always review before running Finalize.

    Common First-Time Issues and Solutions

    1. Vague Issue Description

    Symptom: AI implements something different from your intent

    Solution:

    • Specify target files explicitly
    • Write concrete input/output examples
    • State constraints (compatibility, performance, etc.)

    Before:

    Improve the login feature.
    

    After:

    ## Overview
    Add validation to the login form.
    
    ## Target File
    - src/components/LoginForm.tsx
    
    ## Requirements
    - Email format validation
    - Minimum password length check (8+ characters)
    - Display error messages below the form
    - Enable submit button only after validation passes
    

    2. Test Failures

    Symptom: Some tests fail during the testing phase

    Solution:

    • Check the test result report for failure causes
    • Roll back to the implementation phase and re-run
    • Add testing framework or test style information to the Issue

    3. Scope Too Large

    Symptom: Unexpected files are modified, PR becomes huge

    Solution:

    • Limit the Issue scope specifically
    • Use a "Target Files" section to be explicit
    • Split large tasks into multiple Issues

    4. Doesn't Match Coding Conventions

    Symptom: Generated code style differs from the project

    Solution:

    • State "follow existing coding conventions" in the Issue
    • Reference specific files (e.g., "match the style of src/components/Button.tsx")

    Next Steps

    After your first successful Dev Run, explore these articles:

    Summary

    • Start with a small Issue: Utility function additions or minor fixes are ideal
    • Issue quality is the key to success: Be specific about target files, requirements, and test conditions
    • Use All Phases mode to see the full picture: Experience all 10 phases on your first run
    • Track progress in real-time: Review each phase's artifacts as they complete
    • Review like any normal PR: Check code quality, tests, and documentation
    • Finalize is a human approval step: Always review before executing

    The more you use DevLoop Runner, the better you'll get at writing Issues that produce great results. Follow this tutorial to take your first step.

    Get Started with DevLoop Runner

    Auto-generate PRs from GitHub Issues. Let AI accelerate your development.