GitHub Agentic Workflows (gh-aw) is GitHub’s open-source tool for defining AI-powered repository automation in natural language: you describe what an agent should do in Markdown, and it runs securely through GitHub Actions. The project has earned more than 5,400 stars on GitHub and is actively developed, with its latest updates in 2026.
An agentic workflow has two parts: YAML frontmatter configuring triggers, permissions, tools and the AI engine, and a Markdown body telling the agent what to accomplish — for example triaging new issues, reviewing pull requests, investigating CI failures, maintaining documentation or reporting on repository health. The gh-aw GitHub CLI extension compiles this source into a standard Actions workflow (.lock.yml) that GitHub executes. Multiple AI engines are built in, including GitHub Copilot, Claude Code, OpenAI Codex, Google Gemini and Pi. Safety is a core design concern: agent jobs are read-only and sandboxed by default, and any GitHub writes are applied through separate “safe-outputs” jobs with scoped permissions that validate the agent’s proposed changes first. The maintainers treat security seriously — a vulnerability found in some mid-2026 releases was disclosed publicly and the affected versions retired as a precaution.
GitHub Agentic Workflows is free and open source under the MIT license. Running workflows consumes your GitHub Actions minutes and whatever AI engine usage you configure.
GitHub Agentic Workflows is for maintainers and engineering teams who want AI agents to handle the reasoning-heavy chores of repository upkeep — triage, review and investigation — with strong guardrails.