Built on the open-source Gemini CLI released in June, the action installs a lightweight AI agent straight into any repository. By tagging @gemini-cli in issues or pull-request comments, teams can label bugs, draft fixes, suggest tests, or create release notes, all while the job runs asynchronously on GitHub runners. Repository-wide guidance lives in a GEMINI.md file, and the model can draw on up to one million tokens of context, so large monorepos stay within scope. Access is free for developers who bring a Google AI Studio API key.
The workflow ships as a composite GitHub Action that bundles the NPM version of gemini-cli. Authentication flows through a repository secret or Workload Identity Federation; tasks run in Docker, Podman, or macOS Seatbelt sandboxes, and optional OpenTelemetry hooks let operators export traces and metrics. Because the agent can be triggered on a schedule or from chat-style prompts, routine triage and ad-hoc code reviews share the same configuration.
Gemini CLI can now manage issues and labels on Github Actions as well as take over some of the development tasks!
— TestingCatalog News 🗞 (@testingcatalog) August 6, 2025
Your new SWE coworker 👀 https://t.co/koKYASh7wj pic.twitter.com/9Q5atfKTOn
Positioned between local chat assistants and GitHub’s Copilot, the launch moves Gemini into the continuous-integration layer, aiming to trim review cycles while letting maintainers pin exactly which commands the model may execute.
Google has spent the summer scattering Gemini across its platform: six new data-focused agents, a Conversational Analytics API, and now a DevOps integration illustrate a push to seat Gemini at every stage of the developer workflow.
Gemini CLI GitHub Actions is available today under an Apache-2.0 licence; Google says a stable v1 and managed enterprise runners will follow before the end of the year.
