Google Agents CLI
CLI commands and coding-agent skills for building, evaluating, and deploying ADK agents on Google Cloud.
Per-dimension scores and reasoning
Evidence shows: CLI commands like deploy and publish call Google Cloud APIs, but README does not explicitly mention least privilege or user confirmation mechanisms. Scaffold and eval commands exist, but no mention of sensitive data handling or data flow transparency. Dependency management uses uv, but no dependency security audit is provided. External effects (deploying resources) are mentioned in README, but rollback mechanisms are not described. Source attribution is clear (official Google organization), but no finer-grained attribution is provided. Deductions: lack of specific evidence for least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, and rollback.
Evidence shows: README and code examples (e.g., eval_config.yaml, response_quality.py) are largely consistent in description, but no complete error handling or failure messages are provided. Dependency availability: README lists prerequisites like Python, uv, Node.js, but no guarantee of dependency availability. Deductions: insufficient evidence for failure messages and dependency availability.
Evidence shows: README clearly targets coding assistant users, providing multiple usage scenarios (local development, deployment, evaluation). Capability boundaries are explained in FAQ (e.g., need Google Cloud for deployment). Trigger precision: command list is clear, but no detailed trigger conditions are provided. Environment fit: prerequisites are listed, but no detailed configuration for different environments. Deductions: insufficient evidence for trigger precision and environment fit.
Evidence shows: README provides clear information architecture, including installation instructions, command list, FAQ, and documentation links. Naming stability: command names are consistent. Examples and FAQ provide basic usage examples. Known limitations are partially mentioned in FAQ (e.g., need Google Cloud for deployment), but not comprehensive. License is Apache-2.0, version changelog exists (RELEASE_NOTES.md). Maintenance responsibility is clear (official Google). Deductions: known limitations not comprehensive.
Evidence shows: Output usability: CLI commands and skills are designed to improve coding assistant efficiency, but no actual output examples are provided. Marginal value: README explains difference from ADK, but no comparison with other tools. Cost-benefit: no performance or cost data provided. Deductions: lack of actual output examples and cost-benefit data.
Evidence shows: Claims in README (e.g., support for multiple coding assistants) are supported by documentation links, but no independent verification. Cross-source corroboration: relies only on README and a few code files, lacking external verification. Fact-inference separation: README distinguishes facts (e.g., command list) from inferences (e.g., 'make your coding assistant an expert'), but inferences are not explicitly labeled. Deductions: insufficient cross-source corroboration.
- Static review cannot verify actual execution behavior; all trust-related scores are inferred from documentation and code.
- README does not explicitly mention least privilege, user confirmation, or data flow transparency; deployment commands may incur cloud resource costs.
- Dependency security lacks audit evidence; recommend checking for dependency vulnerabilities before use.
What does this agent do, and when should you use it?
Google Agents CLI is a command-line tool and skill suite for building agents on the Gemini Enterprise Agent Platform. It uses ADK as the agent framework and covers scaffolding, local runs, linting, evaluation, deployment, publishing, and observability. The CLI can be run directly in a terminal or installed as skills for coding assistants including Codex, Claude Code, and Antigravity CLI. Its evaluation workflow can generate inference traces, grade them against metrics, compare result files, analyze failure modes, synthesize scenarios, and optimize prompts. Its cloud delivery boundary is the Google ecosystem, including Agent Runtime, Cloud Run, GKE, and Gemini Enterprise registration.
agents-cli setup installs the CLI and skills for coding agents, while agents-cli scaffold <name> creates an agent project; agents-cli scaffold enhance can add deployment, CI/CD, or RAG to an existing one. Development commands include agents-cli install, agents-cli lint, and agents-cli run "prompt". For evaluation, agents-cli eval generate runs an agent over evaluation cases and produces traces, then agents-cli eval grade grades those traces against metrics; related commands compare results, analyze failures, synthesize multi-turn scenarios, and optimize prompts. agents-cli deploy deploys an agent to Google Cloud, and agents-cli publish gemini-enterprise registers it with Gemini Enterprise. Infrastructure commands can provision single-project infrastructure, CI/CD environments, a datastore for RAG, and data-ingestion infrastructure.
- A Codex user wants to start a new ADK agent project and use one CLI workflow through development and Google Cloud deployment.
- A team with an existing local agent project needs to add deployment, CI/CD, or RAG through
agents-cli scaffold enhance. - An evaluation engineer needs to run an agent across an evaluation dataset, produce traces, and grade them using available metrics.
- A delivery team needs to deploy an agent on Google Cloud or register it with Gemini Enterprise.
- A developer investigating weak evaluation results needs to compare result files, cluster failure modes, or tune prompts using evaluation data.
What are this agent's strengths and limitations?
- Combines ADK development, evaluation, deployment, and publishing into a single CLI and coding-agent skill set.
- Its evaluation workflow explicitly covers trace generation, metric grading, comparisons, failure-mode analysis, scenario synthesis, and prompt optimization.
- Works both as a standalone terminal CLI and as skills for documented coding assistants including Codex, Claude Code, and Antigravity CLI.
- Includes a documented Google Cloud delivery path spanning Agent Runtime, Cloud Run, GKE, CI/CD, datastores, and data ingestion.
- Its core scope depends on Gemini Enterprise Agent Platform, ADK, and Google Cloud; the supplied material does not document a cross-cloud or cross-model migration path.
- The local environment requires Python 3.11+, uv, and Node.js.
- Local Gemini execution needs an AI Studio API key, while deployment and cloud features require Google Cloud.
- The supplied material does not detail integration mechanics or compatibility constraints for coding assistants beyond the named examples.
How do you install or deploy this agent?
Prerequisites are Python 3.11+, uv, and Node.js. Install the CLI and skills with uvx google-agents-cli setup. To install only the skills, run npx skills add google/agents-cli. Local Gemini and ADK development can use an AI Studio API key; deployment and cloud features require Google Cloud. Use agents-cli login to authenticate with Google Cloud or AI Studio.
How do you use this agent?
Create a project with agents-cli scaffold <name>. From the project, run one prompt with agents-cli run "prompt", install project dependencies with agents-cli install, and run Ruff checks with agents-cli lint. For evaluation, run agents-cli eval generate followed by agents-cli eval grade. Deploy with agents-cli deploy; register the result with Gemini Enterprise using agents-cli publish gemini-enterprise.
How does this agent compare with similar options?
It is not an alternative to Antigravity CLI, Claude Code, or Codex; it supplies CLI commands and skills that help those coding assistants build, evaluate, and deploy ADK agents. Compared with using ADK directly, ADK is the agent framework, while Google Agents CLI adds end-to-end scaffolding, evaluation, deployment, and publishing workflows.
FAQ
Can I use it without Google Cloud?
create, run, and eval: the supplied material says Gemini can run locally with ADK using an AI Studio API key. Deployment and cloud features require Google Cloud.Do I need a coding assistant?
How does it evaluate an agent?
agents-cli eval generate to generate traces over evaluation cases, then agents-cli eval grade to grade those traces against metrics. Result comparison, failure analysis, and prompt optimization are also documented.Can it be added to an existing project?
agents-cli scaffold enhance is documented for adding deployment, CI/CD, or RAG to an existing project.