AI Agents

Google Agents CLI: One-Command Skill Injection for Enterprise RAG Agents

The open-source CLI launched April 22, 2026, as part of Gemini Enterprise Agent Platform supplies seven skills for scaffolding, LLM-as-judge evaluation, and deployment to Agent Runtime, extending Karpathy-inspired workflows to production RAG systems.

By Diane Okafor July 8, 2026 4 MIN READ

In a sleek modern corporate technology workspace inside a large enterprise office building an anonymous software developer wearing professional business casual attire of a solid colored button down shirt and dark slacks sits at a wide wooden desk covered with computer hardware including a high end laptop connected via multiple cables to two external monitors a mechanical keyboard ergonomic mouse and bundles of ethernet and power cables running toward visible server racks in an adjacent climate controlled room separated by a glass wall the server room contains orderly rows of tall black server cabinets with small status indicator lights in green and blue patterns representing Google Cloud infrastructure supporting the Gemini Enterprise Agent Platform the developer faces away from the camera slightly to the side with hands positioned over the keyboard illustrating active use of Google Agents CLI for scaffolding LLM as judge evaluation and one command skill injection into production RAG agents the desk surface also holds generic hardware security keys coiled network cables a plain notebook and a ceramic coffee mug while the background shows additional anonymous colleagues at similar workstations engaged with the Agent Development Kit ADK and deployment workflows to Agent Runtime large windows along one wall reveal an urban city skyline with daylight mixing evenly with overhead office lighting potted plants in the corners abstract diagrams on wall mounted whiteboards and neatly arranged storage cabinets containing spare hardware components the overall environment conveys a realistic enterprise setting for extending Karpathy inspired agent workflows into scaled operational systems with every element grounded in tangible objects such as racks of computing equipment networking gear and standard office furniture arranged to highlight the practical deployment of Google Agents CLI within the Gemini Enterprise Agent Platform and Agent Runtime without any visible markings or text on any surfaces — Illustration: AI Intel Report

Google Agents CLI is an open-source tool that injects seven specialized skills into coding agents via a single setup command to enable full-lifecycle enterprise RAG agent builds with LLM-as-judge evaluations and deployment to Gemini Enterprise Agent Platform.

Enterprise teams building retrieval-augmented generation agents must coordinate code generation, evaluation, and production rollout across multiple tools and environments. The Agents CLI addresses these coordination challenges by supplying a standardized command set that aligns development steps with established agentic engineering practices.

Organizations increasingly depend on RAG agents for data-intensive tasks, which raises the priority of repeatable processes for scaffolding, testing, and deployment.

What Background Does Agent Development Present for Enterprises?

Andrej Karpathy has promoted agentic engineering methods that position AI systems as autonomous agents handling multi-step workflows. Enterprises have adopted similar methods for RAG agents that must retrieve information and execute reasoning sequences.

Without unified tooling, teams often assemble separate scripts for each lifecycle stage, increasing maintenance overhead and risk of inconsistent results.

How Does the Agents CLI Address These Issues?

Google introduced the Agents CLI on April 22, 2026, within the Gemini Enterprise Agent Platform on Google Cloud. The tool functions as the programmatic backbone for the agent development lifecycle.

It supplies a machine-readable interface that connects directly to the Agent Development Kit and supports end-to-end management from initial code creation through evaluation and cloud deployment.

What Are the Technical Capabilities of the Agents CLI?

A single setup command adds seven skills to the coding agent environment. The skills include workflow management, ADK code patterns, scaffolding from templates such as agentic_rag, evaluation with LLM-as-judge and adaptive rubrics, deployment to Agent Runtime or Cloud Run or GKE, Gemini Enterprise registration and publishing, and observability configuration.

The CLI accepts commands such as scaffold, eval generate, eval grade, eval optimize, deploy, and publish gemini-enterprise. It also supports infrastructure tasks for RAG datastores and CI-CD pipelines.

Users can execute the setup through uvx google-agents-cli setup or npx skills add google/agents-cli depending on their runtime preference.

Execute the setup command to inject the seven skills into the coding agent.
Run the scaffold command to generate agent code from the agentic_rag template.
Use eval generate to create test scenarios.
Apply eval grade to score scenarios with LLM-as-judge methods.
Execute deploy to target Agent Runtime, Cloud Run, or GKE.
Run publish gemini-enterprise to register the agent with the enterprise platform.

Selected Agents CLI Commands and Their Primary Functions
Command	Function
scaffold	Creates agent code from ADK templates including agentic_rag
eval generate	Produces evaluation scenarios for assessment
eval grade	Scores scenarios using LLM-as-judge and adaptive rubrics
deploy	Deploys to Agent Runtime, Cloud Run, or GKE
publish	Registers agent with Gemini Enterprise

The evaluation workflow produces structured outputs that teams can review before proceeding to deployment stages.

How Does the Tool Support Different Development Modes?

Agents CLI functions in Agent Mode when paired with coding agents including Claude Code, Gemini CLI, Codex, and Cursor. The injected skills guide the coding agent through each lifecycle step.

Standalone Human Mode allows direct command execution by developers without an intervening coding agent. Local development proceeds through AI Studio before cloud deployment.

What Are the Market and Stakeholder Implications?

Chief AI officers and implementation teams gain a consistent interface that reduces the need for bespoke scripts across RAG projects. The quantitative scorecard output supplies measurable performance data that supports ROI assessments.

Direct deployment paths to Google Cloud services allow teams to maintain control over infrastructure choices while accelerating movement from prototype to production.

What Reactions Have Emerged from Early Users?

Today, we are thrilled to introduce Agents CLI in Agent Platform, the unified programmatic backbone for the Agent Development Lifecycle (ADLC) on Google Cloud.Ivan Cheung, Pier Paolo Ippolito, Elia Secchi, Google authors

Community feedback has noted the alignment between the CLI capabilities and established agentic engineering principles.

What Comes Next for Agent Development on Google Cloud?

Documentation and codelabs accompany the release, enabling teams to replicate the demonstrated RAG workflow. Continued integration with additional Google Cloud services is expected to expand template options and evaluation criteria.

Early adopters can apply the current feature set to standardize agent builds while monitoring platform updates for further lifecycle automation.

Frequently asked

How does the Agents CLI integrate with existing coding agents?

The Agents CLI supports Agent Mode when used with tools such as Claude Code, Gemini CLI, and Cursor, allowing the seven skills to guide automated development steps.

What evaluation approach does the tool apply to RAG agents?

It generates scenarios and scores them through LLM-as-judge methods with adaptive rubrics, producing quantitative scorecards for performance measurement.

Can enterprises use the Agents CLI without an AI coding agent?

Yes, the tool operates in standalone Human Mode for direct command-line use by development teams.