Google Agents CLI Skills

google/agents-cli packages a CLI and seven Agent Skills for coding agents that build, evaluate, deploy, publish, and observe Google ADK agents. It is designed for Gemini CLI, Claude Code, Codex, Antigravity, and other coding agents that need source-backed Google Cloud and Gemini Enterprise Agent Platform workflows instead of generic agent advice.

The project is the active successor named by Google Cloud Agent Starter Pack's README for new projects. It gives agents a lifecycle path from scaffolded ADK code to evals, deployment targets, Gemini Enterprise registration, and observability.

Knowledge Freshness

The repository is current and the README describes agents-cli as the CLI and skills for building agents on Gemini Enterprise Agent Platform. The skills target Google ADK, Agent Runtime, Cloud Run, GKE, CI/CD, evals, datastore infrastructure, observability, and Gemini Enterprise publishing.

Google Cloud product names, Agent Platform launch stages, CLI commands, deployment defaults, IAM requirements, eval metrics, and ADK APIs can change. Use the skills for lifecycle structure, then verify flags, cloud terms, service availability, pricing, and permissions against the current docs and target project before running infrastructure or deployment commands.

Retrieval Sources

This listing is grounded in:

• The google/agents-cli repository README.
• The public Agents CLI documentation and skills reference.
• The gemini-extension.json package metadata.
• Representative SKILL.md files for workflow, scaffolding, evaluation, and deployment.
• The Apache-2.0 license text.
• Current GitHub repository metadata.

Core Workflow

Install only the skills:

npx skills add google/agents-cli

Install the CLI plus skills:

uvx google-agents-cli setup

Create a prototype ADK agent:

agents-cli scaffold create my-agent --agent adk --prototype

Run the evaluation loop:

agents-cli eval run

Capability Scope

The README lists these seven Agent Skills:

The CLI commands cover setup, login, scaffold, run, install, lint, eval generate/grade/compare/analyze/optimize, deploy, publish, infrastructure, datastore provisioning, data ingestion, project info, and skill updates.

Production Rules

Use these skills as an agent-building lifecycle, not a shortcut around cloud review:

• Clarify the agent purpose, tools, APIs, data sources, safety constraints, and deployment preference before scaffolding.
• Prefer prototype-first development until behavior and eval cases are clear.
• Preserve scaffolded manifests, deployment structure, and guidance files unless the user explicitly approves structural changes.
• Do not deploy, publish, provision infrastructure, set up CI/CD, or run datastore ingestion without explicit approval and a known Google Cloud target.
• Treat evals as the validation path for agent behavior; do not write brittle tests that assert exact LLM output text.
• Review IAM, service accounts, Secret Manager access, Terraform state, regions, billing, and rollback before production deployment.
• Summarize eval traces, deployment output, and observability data without leaking prompts, credentials, user data, project IDs, or trace details.

Use Cases

• Ask Codex to create a prototype ADK agent with the right scaffolding and guidance files.
• Use Claude Code to enhance an existing agent project with Agent Runtime, Cloud Run, or GKE deployment support.
• Have Gemini CLI build an eval dataset, run traces, grade results, and iterate through a quality flywheel.
• Use an agent to deploy only after explicit approval, with service account, secret, region, and target checks.
• Register a deployed agent with Gemini Enterprise and then inspect Cloud Trace, logs, analytics, and observability output.

Source Review

• The README says agents-cli provides CLI commands and skills for building agents on Gemini Enterprise Agent Platform.
• The README explicitly lists Gemini CLI, Claude Code, Codex, Antigravity, and other coding agents as supported coding assistants.
• The setup section documents uvx google-agents-cli setup and a skills-only install path with npx skills add google/agents-cli.
• The Agent Skills table lists seven skills covering workflow, ADK code, scaffolding, evaluation, deployment, publishing, and observability.
• The workflow skill requires understanding the agent requirements before code or scaffolding, then routes through scaffold, build, eval, deploy, publish, and observe phases.
• The deploy skill covers Agent Runtime, Cloud Run, GKE, CI/CD, secrets, service accounts, rollback, and production infrastructure.
• The evaluation skill documents eval generation, grading, comparison, failure analysis, and prompt optimization.
• The license file identifies the repository as Apache-2.0.

Duplicate Review

Checked current content/skills/, content/tools/, content/mcp/, open pull requests, and repository-wide content for Google Agents CLI, google/agents-cli, google-agents-cli, agents-cli, ADK agent skills, Agent Runtime deployment skills, and Gemini Enterprise Agent Platform skills. The existing Google Agent Development Kit entry covers google/adk-python as a tool/framework; no dedicated Google Agents CLI skills entry, exact source URL duplicate, target file, or open duplicate PR was found.

Disclosure

Editorial listing. No paid placement or affiliate link is used. The upstream repository is Apache-2.0 licensed. Google Cloud, Gemini Enterprise Agent Platform, Agent Runtime, Cloud Run, GKE, AI Studio, model providers, and third-party observability or CI/CD services may have separate billing, terms, privacy, and access-control requirements.