Building an AI bug-fixing agent that runs inside your GitHub Actions runner

Описание и применение

Система автоматического исправления ошибок в репозиториях GitHub использует агент, работающий внутри собственного раннера Actions, сохраняя код приватным и ограничивая возможности агента чтением файлов и созданием PR.

Задача

Production errors need automated fixes, but existing SaaS tools expose code to external services.

Результат

The system achieves automated bug fixes while keeping code private and limiting agent blast radius, but cannot run tests and requires PR permissions to be enabled.

Рабочий процесс

• Receive error from tracker
• Control plane starts workflow via workflow_dispatch
• Runner checks out repo and runs agent
• Agent explores code, writes fix to new branch
• Agent opens PR
• Runner reports status, PR link, summary, token counts back to control plane

Доказательства

The service that coordinates the agent should never be able to read the code.

Источник

The agent has four tools: get_file_contents, list_directory, search_code, create_pull_request.

Источник

The agent cannot run your tests. It can read a test file and mention it, but it cannot execute anything.

Источник