agents-cli is a CLI and skill package provided by Google for agent development on the Gemini platform. Its positioning is "providing tools for coding agents," rather than being a coding agent itself. It does not replace Antigravity CLI, Claude Code, or Codex; instead, it makes your coding agent better at building, evaluating, and deploying ADK agents on Google Cloud.
Main capabilities It provides CLI commands and skills covering the agent building lifecycle, including scaffold, ADK code patterns, eval, deploy, publish, and observability. It is built on top of Google Cloud's agent technology stack, helping users avoid having to learn each CLI and service on their own.
How to use Prerequisites: Python 3.11+, uv, and Node.js. After installation, in Antigravity CLI, Claude Code, Codex, or any coding agent, directly ask the agent to use agents-cli for the build task, for example, "Use agents-cli to build a caveman-style agent that compresses verbose text into terse, technical grunts." The CLI can also be used directly in the terminal independently of a coding agent, and can run all commands such as agents-cli scaffold, eval, and deploy. For existing agent projects, agents-cli scaffold enhance can add deployment and CI/CD to them.
Local development and deployment Local development (create, run, eval) does not require Google Cloud; you can use an AI Studio API key to run Gemini locally through ADK. Deployment and cloud features require Google Cloud. When deploying an agent, resources are deployed in the user's own Google Cloud project; users are responsible for these resources and should review the Google Cloud Terms of Service.
Extensions agents-cli skills cover the agent building lifecycle. For adjacent areas, other skill suites can be installed alongside it, for example, agent-skills covers general software engineering workflows (ideation, spec gates, planning, code review), and google/skills covers Google Cloud fundamentals (BigQuery, Cloud Run, Firebase, GKE).
Telemetry and privacy By default, each agents-cli command invocation sends one usage signal to the user's own Google Cloud project, writing a structured log entry (logName=projects/<project>/logs/agents-cli) that records the command name (such as deploy, scaffold.create), exit code, error category, duration, CLI/Python version, and operating system. It never records file contents, arguments, or other inputs. Commands added by extensions do not record their names. The requested User-Agent and X-Goog-Api-Client headers carry usage metadata for Google's usage analysis, including CLI and Python versions, operating system and version, CPU architecture, command name, and a random invocation ID; the log entry content remains in the user's project. Telemetry can be opted out of via an environment variable; use agents-cli info to view the current telemetry status.
Contributions and support You can submit bug reports, feature requests, and ideas through issues to influence the roadmap. You can also share projects or feedback via agents-cli@google.com.
