GitHub has introduced a new agent based on generative AI that automates task execution in repositories, such as bug fixing, adding functionality, improving documentation, and code refactoring. The agent is integrated into the GitHub Copilot service and starts working after the user assigns it a task through GitHub, the mobile app, or the command interface. For this, the agent creates a secure development environment, analyzes the codebase, considers the context of discussions, and follows the repository’s individual instructions.
During task execution, the agent automatically creates a virtual machine, copies the repository, makes changes, and commits them in a draft pull request. All work steps are accompanied by logs explaining the logic of the decisions made, and upon completion, the user receives a notification to review the result. The user can leave comments on the changes, which the agent will consider and make the necessary adjustments.
To ensure project security, the agent has limitations — it can only make changes to branches it has created, without affecting the main branch or those created by the team. Any changes are subject to mandatory review by other team members, and CI/CD processes are triggered only after approval. The agent’s internet access is restricted to a trusted list of addresses, and all existing repository rules remain in effect.
The new feature is already available for Copilot Enterprise and Copilot Plus users. The agent can be activated in the respective repositories, as well as in development environments like Xcode, Eclipse, JetBrains, and Visual Studio. According to GitHub, the agent is best suited for automating low to medium complexity tasks in well-tested projects, allowing developers to focus on more complex or creative tasks.
