GitHub Agentic Workflows hits GA, and the agent now lives inside your Actions permission model

GitHub has launched GitHub Agentic Workflows in general availability, bringing its open source project, developed by GitHub Next, directly into the Actions permission model. This allows users to run agents against their repositories from within GitHub, rather than from a laptop with their own token. The agent, which compiles natural-language instructions from a Markdown file into standard Actions YAML, can now be used for tasks like issue triage, CI failure analysis, documentation updates, and test writing.

GitHub Copilot or other coding agents perform the actual work, while the agent executes in a sandboxed container behind the Agent Workflow Firewall. New features at GA include ARC self-hosted runners with Docker-in-Docker, a non-experimental gh aw forecast for threat detection costs and pricing, per-workflow model selection, templates via gh aw add-wizard, and support for Cloud Hypervisor or NVX micro-VMs and enclaves against data leaks in cross-repository workflows.

GitHub Agentic Workflows is available on Free, Team, and Enterprise plans, and requires a Copilot license for the default GitHub Copilot CLI engine. The release emphasizes the importance of security, with measures like read-only permissions by default, validated outputs, and proposed changes scanned before application. The changelog provides various examples and tools for testing and estimating costs before implementing the new workflows.