github/gh-aw

What it solves

It enables developers to automate complex repository tasks that require reasoning and interpretation—such as issue triage, pull-request reviews, and CI failure investigation—without replacing existing deterministic CI/CD pipelines.

How it works

Developers define agentic workflows using Markdown files with YAML frontmatter (for configuration like triggers, permissions, and AI engines) and a Markdown body (for instructions to the agent). A GitHub CLI extension (gh-aw) compiles these definitions into standard GitHub Actions workflows (.lock.yml) for execution. The system uses a sandboxed, read-only environment by default, applying writes through validated "safe-outputs" jobs with scoped permissions.

Who it’s for

Developers and repository maintainers who want to integrate AI agents into their GitHub automation to handle non-deterministic tasks.

Highlights

• Supports multiple AI engines including GitHub Copilot, Claude Code, OpenAI Codex, Google Gemini, and Pi.
• Integrates directly into GitHub Actions via a CLI extension.
• Focuses on security with sandboxed execution and controlled write permissions.
• Complements standard GitHub Actions for deterministic builds and tests.