As reported on November 18, 2026, GitHub Agentic Workflows is generally available. The open-source project runs coding agents in GitHub Actions for repository tasks such as issue triage and continuous integration (CI) failure analysis. It entered public preview in June 2026.
Developers describe a task in natural-language Markdown, configure it with YAML front matter, and compile the definition into a standard Actions workflow. Teams can use Copilot or other coding agents while reusing existing runner groups and policy restrictions.
Write access is separated from the agent by default. An agent starts with read-only access; separate jobs, using their own write tokens, apply any changes it proposes. GitHub says agents run in sandboxed containers protected by Agent Workflow Firewall, and a threat-detection job scans proposed changes before they are applied. These safeguards do not automatically cover directly granted write permissions or custom jobs added by a workflow author. Authors can also require reviewer approval before a write job runs through GitHub Environment protection rules.
Since the public preview, GitHub has added support for Actions Runner Controller (ARC), its Kubernetes operator for self-hosted runners. ARC can use Docker-in-Docker to let an agent launch a sandbox container. The project’s guide notes a configuration trade-off: ARC runs the Docker-in-Docker sidecar as a privileged container by default, placing a privileged component in the same environment as an agent designed for minimal permissions.
GitHub’s documented price is $0.01 per AI Credit (AIC), with a default limit of 1,000 AIC, or $10, per run. Enterprise control documentation also lists defaults of 5,000 AIC per day, 400 AIC per threat-detection run and a 20-minute maximum for agent steps; organizations can override them. The gh aw forecast command, no longer experimental, estimates monthly AI Credit spending, including threat-detection costs.
In GitHub’s announcement, Ben Alton, a senior manager in Accenture’s strategy, architecture and innovation team, said one of his team’s leads estimated a saving of 29 hours a week on development work alone. The announcement did not explain how that figure was measured.
A tester took a narrower approach. In an account of an issue-triage workflow, freenet middleware team lead Sascha Mirow-Stanitzok said the default budget was too generous for that task and set limits of 100 AIC, 15 interaction rounds and 10 minutes per run. He also warned that telling an agent to ignore instructions in issue text does not itself protect a workflow; he would not use the same pattern for workflows that change code or create pull requests.
GitHub’s FAQ recommends conventional Actions for reproducible builds and tests, reserving agentic workflows for tasks that require understanding or generating content. The project is available to GitHub Free, Team and Enterprise users on github.com. Its default Copilot CLI engine requires a Copilot license.