Software development is a collaborative effort, and managing pull requests and commits efficiently is crucial for maintaining the health of a project. GitLoop's Pull Requests and Commits Review feature is designed to streamline this aspect of development, enhancing both the speed and quality of code reviews. GitLoop provides detailed insights into each pull request and commit, highlighting changes and potential impacts on the rest of the codebase. This enhanced visibility helps reviewers understand the context of changes better, which is especially useful in large projects where understanding the implications of new code can be challenging.

‍

‍

Customizing Code Reviews

A key feature of GitLoop's review agent is the ability to customize it to fit the company's and project's needs, from review process automation to AI prompting and token limitations. Proper configuration ensures better results. To start configuring, simply click on the "Configure" button in the top bar of the review page list, or click on the workspace button in the bottom left, then go to "Settings," and click on "Review Configs" to create a new config for your workspace repository.

Types of Reviews

GitLoop supports two types of reviews:

1. Pull Request Review: If your company works with pull requests, you can configure review settings specific to pull requests.
2. Commit Review: If you prefer working with commits, this option is also available, allowing you to specify the branch or leave it empty to use the default branch.

‍

Automating the Review Process

One of the standout benefits of this feature is the automation of reviews. GitLoop receives real-time data from GitHub, including code changes, where configured AI agents read these changes and perform the required review. To enable automatic review, go to your review config and click on "Auto Review" in the "Incoming/Current Changes" step.

Controlling Limits

As changes vary in size, from small tweaks to large updates, the cost to review can also vary. GitLoop offers the ability to limit the incoming changes' length to manage costs effectively. In the future, the team plans to introduce smarter ways to analyze and review larger changes. Additionally, if your review config requires only short, informative answers, you can configure answer limits accordingly.

Rules Injection

Each project has its own rules, and each company/developer prefers working in a way that ensures the highest quality of project outcomes. While the AI reviews the global aspects of changes, it may sometimes lack the ability to fit the review to the specific needs of the project. GitLoop addresses this by allowing the specification of review prompts and guiding the agent to give better answers. Here are some examples:

Example Prompts

1. Example 1:
   • Prompt: "You will be provided with pull request changes. Your role is to review the pull request and extract the potential issues and problems. At the end of the review, provide a severity score of the changes from 0% to 100%."
   • Purpose: This prompt guides the AI to review the pull request for potential issues and give a score at the end.
2. Example 2:
   • Prompt: "Please review the following commit for adherence to our coding standards, including naming conventions, code comments, and error handling. Highlight any deviations and suggest corrections."
   • Purpose: This prompt ensures the AI checks for compliance with the company’s coding standards and provides specific feedback on any issues.
3. Example 3:
   • Prompt: "Review this pull request to ensure that no critical files such as config.yaml, secrets.json, or any files in the src/main/security/ directory are modified. Flag any changes to these files."
   • Purpose: This prompt helps the AI to focus on verifying that no critical or sensitive files are being altered, which is crucial for maintaining security and stability.

Choosing the AI Model

GitLoop is fully based on OpenAI models, which are proven to handle coding standards effectively, especially the latest GPT-4 models.

Integrations

Soon, GitLoop will offer the ability to integrate with your preferred medium to get notified about changes and filter the desired review, such as Slack, Discord, emails, and GitHub to leave comments directly in the PRs and commits.