Agents CLI is a command-line tool and a set of agent skills for building, evaluating, and deploying agents on Google Cloud's Gemini Enterprise Agent Platform. It is a tool for a coding agent, not a coding agent itself: it installs skills and commands that let an assistant such as Claude Code, Codex, or Antigravity CLI scaffold, evaluate, and ship agents without a person learning each underlying Google Cloud service by hand. The CLI also runs standalone from a terminal.
The skills cover the development lifecycle: the ADK Python API for agents, tools, orchestration, callbacks, and state, project scaffolding, an evaluation methodology with metrics, datasets, and LLM-as-judge grading, deployment to Agent Runtime, Cloud Run, or GKE, registration with Gemini Enterprise, and observability through Cloud Trace and logging.
Commands span the same stages. You create an agent project, run it from a prompt, run an eval dataset and grade the traces, compare or cluster eval results, auto-tune prompts from eval data, provision infrastructure and CI/CD, deploy to Google Cloud, and publish to Gemini Enterprise. Local development with create, run, and eval works with a Gemini key from AI Studio, while deployment and cloud features need a Google Cloud project.
Features
- For coding agents: installs skills and commands so an assistant becomes capable at building agents, rather than being an agent itself
- Lifecycle skills: workflow rules, the ADK Python API, scaffolding, evaluation, deployment, Gemini Enterprise publishing, and observability
- Scaffolding: creates projects and adds deployment or CI/CD to existing ones
- Evaluation: runs agents over datasets, grades traces against metrics, synthesizes scenarios, compares runs, clusters failure modes, and auto-tunes prompts
- Deployment targets: Agent Runtime, Cloud Run, and GKE, with infrastructure and CI/CD provisioning
- Local development: create, run, and eval work with a Gemini key from AI Studio without a cloud project
- Standalone or agent-driven: every command runs directly in a terminal, and the skills let a coding agent run them for you
- Observability: Cloud Trace, logging, and third-party integrations
