agents-cli

About agents-cli

agents-cli is a command-line tool that scaffolds, evaluates, and deploys AI agents on Google Cloud. It is built to be driven by a coding agent-such as Claude Code or Antigravity-and handles the lifecycle from project creation to production deployment. The CLI uses Google's open-source Agent Development Kit (ADK) for multi-agent orchestration and targets Vertex AI Agent Engine, Cloud Run, and GKE.

Review

The tool addresses the distance between a demo agent and a production system. Instead of weeks of setup, a developer can use a coding agent to run agents-cli commands that scaffold a full project, run evaluations, and deploy with a human approval gate. The CLI operates headless, so a coding agent can iterate on agent behavior against a defined success signal without a person watching the output.

Key Features

• Scaffold - agents-cli create generates a complete project with agent code, tools, tests, Dockerfile, observability configuration, Terraform, and CI/CD pipelines.
• Evaluate - agents-cli eval run creates synthetic conversations from success criteria, tests the agent, and computes metrics. The coding agent can then fix code and prompts, iterating until the criteria are met.
• Deploy - agents-cli deploy ships the agent to Vertex AI Agent Engine, Cloud Run, or GKE, using Secret Manager for credentials and requiring explicit human approval before proceeding.
• Headless operation - The CLI is built for agent-driven workflows; it can run unattended, allowing the coding agent to self-optimize based on evaluation scores.
• ADK integration - The tool uses Google's Agent Development Kit, an open-source, model-agnostic framework that handles agent plumbing like state management, tool calling, and orchestration.

Pricing and Value

agents-cli is currently listed as free. The reference page does not detail any future pricing tiers or models. Underlying Google Cloud resources (Vertex AI, Cloud Run, GKE) incur their own costs.

Pros

• Reduces the repetitive setup work for production agent deployments by scaffolding a full project in one command.
• Evaluation loop automatically generates test data and metrics, letting a coding agent iterate against a concrete signal.
• Deployment integrates with Google Cloud services and enforces a human approval step, preventing unattended production pushes.
• Headless mode allows a coding agent to drive the entire cycle without a human watching the output.
• Built on open-source ADK, so the agent framework is model-agnostic and customizable.

Cons

• Not well suited for teams that do not use Google Cloud or who want a provider-agnostic deploy pipeline out of the box; the default deploy targets are Google-specific.
• The evaluation loop's guardrails against overfitting or metric gaming are implemented through the coding agent's prompt/skill layer, not enforced by the CLI itself.
• CLI output is primarily human-readable prose; the tool is still being improved to produce structured, agent-friendly output, which can cause a coding agent to misinterpret results.

agents-cli fits a workflow where a developer works alongside a coding agent to build and ship agents on Google Cloud. It is most useful when the agent's behavior can be evaluated quantitatively and the team is comfortable with the ADK ecosystem. The human-in-the-loop deploy gate and the open-source foundation make it an option for those who want to accelerate the production path without handing over full autonomy.

Open 'agents-cli' Website