Skip to content
/

Google Agents CLI

google-agents-cli · google/agents-cli · ★ 2.6k · last commit 2026-05-19

Primitive shape 7 total
Skills 7
00

Summary

google/agents-cli — Summary

Google Agents CLI is Google's official cross-IDE skill pack and CLI tool for building and deploying AI agents on the Gemini Enterprise Agent Platform (formerly Vertex AI). It ships 7 agent skills covering the full development lifecycle — workflow, ADK code patterns, scaffolding, evaluation, deployment, publishing, and observability — plus a Python CLI (agents-cli) that wraps Google Cloud infrastructure operations.

Problem it solves: Building agents for Google Cloud requires learning Google ADK Python API, Cloud Run, GKE, Vertex AI Agent Runtime, Terraform, and CI/CD pipelines. Agents CLI packages all this expertise as skills that any coding agent can load, reducing the expertise barrier to production deployment.

Distinctive trait: The first official vendor CLI + skill pack designed to be cross-IDE from the ground up — explicitly targets Gemini CLI, Claude Code, Codex, Antigravity, and any compliant coding agent via the agentskills.io standard. Each skill ships explicit STOP — Do NOT write code yet guardrails and cross-skill references ("Read skill X before phase Y").

Target audience: Developers building agents for Google Cloud's Gemini Enterprise Agent Platform, regardless of which coding assistant they use.

Scope: 2,589 GitHub stars, Apache-2.0, 2 contributors, active (last push May 2026). PyPI package: google-agents-cli.

Differs from seeds: Unlike superpowers (Archetype 1 — behavioral enforcement without a CLI), agents-cli ships both a dedicated Python CLI binary (agents-cli) AND 7 skills — the hybrid CLI+skills pattern is unique. The hard "STOP — Do NOT write code yet. Scaffold first" guardrail in the workflow skill resembles superpowers' Iron Law approach but is goal-directed (scaffold first to get eval boilerplate) rather than process-directed.

01

Overview

google/agents-cli — Overview

Origin

Published by Google LLC (google GitHub organization) as part of the Google Cloud Gemini Enterprise Agent Platform ecosystem. The agents-cli is the developer-facing tooling for Google ADK (Agent Development Kit). Companion tools include: Google ADK Python library (adk.dev), Gemini CLI, Antigravity.

Philosophy

"Turn your favorite coding assistant into an expert at building and deploying agents on Google Cloud."

Key principles:

  1. Coding-agent first: The CLI teaches coding agents how to use it — skills ARE the documentation
  2. Cross-IDE portability: Works with Gemini CLI, Claude Code, Codex, Antigravity, or any agent
  3. Guard-railed development: Skills enforce "scaffold before code" and "eval before deploy" discipline
  4. Eval-driven iteration: The eval skill explicitly mandates 5-10+ iterations

The STOP Guardrail

The workflow skill opens with:

STOP — Do NOT write code yet. If no project exists, scaffold first with agents-cli scaffold create <name>. If the user already has code, use agents-cli scaffold enhance . to add the agents-cli structure. Run agents-cli info to check if a project already exists. Skipping this leads to missing eval boilerplate, CI/CD config, and project conventions.

This is a directive enforced through skill activation — whenever the agent loads the google-agents-cli-workflow skill, it is instructed to check for scaffolding before coding.

ADK Focus

Google ADK (Agent Development Kit) is the Python framework that agents-cli wraps:

  • Agents, tools, orchestration, callbacks, state
  • Multi-agent coordination (sequential, parallel, LLM-as-router, loop)
  • Built-in evals, deployment targets

Platform Naming

"The platform formerly known as 'Vertex AI' is now Gemini Enterprise Agent Platform. Users may refer to products by different names."

The skill includes an explicit product name mapping table to handle user confusion.

02

Architecture

google/agents-cli — Architecture

Distribution

  • Type: Claude Code plugin + skill-pack + CLI tool
  • CLI Install: uvx google-agents-cli setup (installs CLI + skills to all detected IDEs)
  • Skills only: npx skills add google/agents-cli
  • PyPI: uv tool install google-agents-cli
  • License: Apache-2.0

Repository Structure

google/agents-cli/
├── .claude-plugin/
│   └── plugin.json               # Claude Code plugin manifest
├── skills/                        # 7 Agent Skills
│   ├── google-agents-cli-workflow/
│   │   └── SKILL.md              # Master lifecycle skill
│   ├── google-agents-cli-adk-code/
│   │   └── SKILL.md              # ADK Python API patterns
│   ├── google-agents-cli-scaffold/
│   │   └── SKILL.md              # Project scaffolding
│   ├── google-agents-cli-eval/
│   │   ├── SKILL.md              # Evaluation methodology
│   │   └── references/           # criteria-guide, user-simulation, etc.
│   ├── google-agents-cli-deploy/
│   │   ├── SKILL.md              # Deployment to Cloud Run / Agent Runtime / GKE
│   │   └── references/           # cloud-run, agent-runtime, gke, cicd
│   ├── google-agents-cli-publish/
│   │   └── SKILL.md              # Gemini Enterprise registration
│   └── google-agents-cli-observability/
│       └── SKILL.md              # Cloud Trace, logging, monitoring
├── gemini-extension.json         # Gemini CLI plugin manifest
├── docs/                         # Documentation site
├── CONTRIBUTING.md
├── LICENSE
├── RELEASE_NOTES.md
└── README.md

Plugin Manifests

// .claude-plugin/plugin.json
{
  "name": "google-agents-cli",
  "version": "0.2.0",
  "description": "Scaffold, develop, evaluate, and deploy AI agents with Google ADK. Bundles 7 skills for the agent development lifecycle.",
  "author": { "name": "Google LLC" },
  "homepage": "https://github.com/google/agents-cli",
  "license": "Apache-2.0"
}

Required Runtime

  • python >= 3.11
  • uv (Python package manager)
  • node >= LTS (for skills install via npx)
  • Google Cloud SDK + credentials (for deployment)

Target AI Tools

  • Gemini CLI (primary)
  • Claude Code (.claude-plugin/)
  • Codex
  • Antigravity (Google's agent platform)
  • Any agentskills.io-compatible agent
03

Components

google/agents-cli — Components

Agent Skills (7)

Skill Description Key Feature
google-agents-cli-workflow Full ADK development lifecycle and coding guidelines. Always active — entrypoint for all agent development STOP guardrail + phase-to-skill cross-reference table
google-agents-cli-adk-code ADK Python API — agents, tools, orchestration, callbacks, state API patterns, multi-agent types, tool decoration
google-agents-cli-scaffold Project scaffolding — create, enhance, upgrade Scaffold-first enforcement
google-agents-cli-eval Evaluation methodology — metrics, evalsets, LLM-as-judge, trajectory scoring Eval-fix loop (5-10+ iterations expected)
google-agents-cli-deploy Deployment to Agent Runtime, Cloud Run, GKE, CI/CD Deployment target decision matrix
google-agents-cli-publish Gemini Enterprise registration Register agent in Gemini Enterprise catalog
google-agents-cli-observability Cloud Trace, logging, third-party integrations Observability setup post-deploy

CLI Commands (agents-cli)

Command Purpose
agents-cli setup Install CLI + skills to all detected coding agents
agents-cli scaffold create <name> Create new agent project with eval boilerplate + CI/CD
agents-cli scaffold enhance . Add agents-cli structure to existing project
agents-cli scaffold upgrade Upgrade project to newer agents-cli version
agents-cli run "prompt" Run agent with single prompt
agents-cli install Install project dependencies
agents-cli lint Code quality checks (Ruff)
agents-cli eval run Run agent evaluations
agents-cli eval compare Compare two eval result files
agents-cli deploy Deploy to Google Cloud
agents-cli publish gemini-enterprise Register with Gemini Enterprise
agents-cli infra single-project Provision single-project infrastructure
agents-cli infra cicd Set up CI/CD pipeline + staging/prod infra
agents-cli infra datastore Provision datastore for RAG
agents-cli data-ingestion Run data ingestion pipeline
agents-cli info Show project config + CLI version
agents-cli update Force reinstall skills to all IDEs
agents-cli login Authenticate with Google Cloud or AI Studio
agents-cli login --status Show authentication status

Skill Cross-Reference Table (from workflow skill)

Phase Skill to load When
0 — Understand Read .agents-cli-spec.md if present
1 — Study samples Notable samples table
2 — Scaffold /google-agents-cli-scaffold Before creating project
3 — Build /google-agents-cli-adk-code Before writing agent code
4 — Evaluate /google-agents-cli-eval Before running any eval
5 — Deploy /google-agents-cli-deploy Before deploying
6 — Publish /google-agents-cli-publish After deploying
7 — Observe /google-agents-cli-observability After deploying

Reference Files (in deploy and eval skills)

The google-agents-cli-deploy skill bundles reference files:

  • references/cloud-run.md — scaling, Dockerfile, session types
  • references/agent-runtime.md — deploy.py CLI, AdkApp pattern, Terraform
  • references/gke.md — GKE Autopilot, Kubernetes manifests
  • references/terraform-patterns.md — custom infrastructure, IAM
  • references/cicd-pipeline.md — full CI/CD pipeline setup
  • references/testing-deployed-agents.md — testing per deployment target

The google-agents-cli-eval skill bundles:

  • references/criteria-guide.md — 8 evaluation criteria, match types
  • references/user-simulation.md — ConversationScenario, user simulator
  • references/builtin-tools-eval.md — google_search tool eval
  • references/multimodal-eval.md — multimodal inputs
05

Prompts

google/agents-cli — Prompts

Verbatim Excerpt 1: Workflow Skill — STOP Guardrail

File: skills/google-agents-cli-workflow/SKILL.md Technique: Hard-stop directive + prerequisite enforcement

---
name: google-agents-cli-workflow
description: >
  This skill should be used when the user wants to "develop an agent",
  "build an agent using ADK", "run the agent locally", "debug agent code",
  "test an agent", "deploy an agent", "publish an agent", "monitor an agent",
  or needs the ADK (Agent Development Kit) development lifecycle and coding
  guidelines. Entrypoint for building ADK agents.
  Always active — provides the full workflow (scaffold, build, evaluate,
  deploy, publish, observe), code preservation rules, model selection
  guidance, and troubleshooting steps for ADK or any agent development.
metadata:
  author: Google
  license: Apache-2.0
  version: 0.2.0
  requires:
    bins:
      - agents-cli
    install: "uv tool install google-agents-cli"
---

# ADK Development Workflow & Guidelines

> **STOP — Do NOT write code yet.** If no project exists, scaffold first with
`agents-cli scaffold create <name>`. If the user already has code, use
`agents-cli scaffold enhance .` to add the agents-cli structure. Run
`agents-cli info` to check if a project already exists. Skipping this leads to
missing eval boilerplate, CI/CD config, and project conventions.

Prompting technique: Hard-stop command at the top of the skill with explicit consequence ("Skipping this leads to missing..."). The description field is designed as an activation trigger — listing all phrases that should activate this skill (very broad, catches any agent-related intent).

Verbatim Excerpt 2: Eval Skill — Eval-Fix Loop Instructions

File: skills/google-agents-cli-eval/SKILL.md Technique: Iterative loop enforcement with explicit counter

## The Eval-Fix Loop

Evaluation is iterative. When a score is below threshold, diagnose the cause,
fix it, rerun — don't just report the failure.

### How to iterate

1. **Start small**: Begin with 1-2 eval cases, not the full suite
2. **Run eval**: `agents-cli eval run`
3. **Read the scores** — identify what failed and why
4. **Fix the code** — adjust prompts, tool logic, instructions, or the evalset
5. **Rerun eval** — verify the fix worked
6. **Repeat steps 3-5** until the case passes
7. **Only then** add more eval cases and expand coverage

**Expect 5-10+ iterations.** This is normal — each iteration makes the agent better.

### Shortcuts That Waste Time

Recognize these rationalizations and push back — they always cost more time
than they save:

Prompting technique: Step-numbered loop with exit condition ("until the case passes"). Counter injection ("Expect 5-10+ iterations") normalizes iteration and prevents premature stopping. Anti-pattern list ("Shortcuts That Waste Time") preemptively blocks agent rationalization behavior.

Verbatim Excerpt 3: Deploy Skill — Deployment Target Decision Matrix

File: skills/google-agents-cli-deploy/SKILL.md

## Deployment Target Decision Matrix

Choose the right deployment target based on your requirements:

| Criteria | Agent Runtime | Cloud Run | GKE |
|----------|-------------|-----------|-----|
| **Languages** | Python | Python | Python (+ others via custom containers) |
| **Scaling** | Managed auto-scaling | Fully configurable | Full Kubernetes scaling |
| **Session state** | Native VertexAiSessionService | In-memory (dev) / Cloud SQL | In-memory / Cloud SQL |
| **Setup complexity** | Lower (managed, purpose-built) | Medium | Higher (Kubernetes expertise) |

**Ask the user** which deployment target fits their needs. Each is a valid
production choice with different trade-offs.

Prompting technique: Decision matrix forces explicit user choice — the agent cannot pick a default. "Ask the user" instruction prevents autonomous decisions on infrastructure choices.

09

Uniqueness

google/agents-cli — Uniqueness & Positioning

Differs from Seeds

The closest seed is superpowers (Archetype 1 — skills-only enforcement) in the "skills that enforce development discipline" sense. But agents-cli adds a dedicated CLI binary, cloud deployment capabilities, an eval-driven development loop, and multi-IDE support — making it more like a full developer toolkit than a behavioral framework.

Unlike taskmaster-ai (Archetype 3 — MCP-anchored toolserver), agents-cli uses the agentskills.io format (not MCP) for skill delivery, and its CLI is a developer tool (scaffold/eval/deploy) not a task management system.

No seed has the combination of: 7 lifecycle skills + dedicated CLI + cross-IDE plugin manifests + cloud deployment + built-in evaluation loop.

Distinctive Properties

  1. Eval-first philosophy: The google-agents-cli-eval skill is the most detailed skill in the set, with 4 reference files and explicit iteration expectations ("Expect 5-10+ iterations"). Evaluation is treated as a first-class development stage, not an afterthought.

  2. Skills as guard rails, not documentation: The STOP directive and "Don't waste time" anti-pattern lists are behavioral instructions embedded in skills — the skill format is used for enforcement, not just knowledge delivery.

  3. Product name confusion handling: The explicit Vertex AI → Gemini Enterprise Agent Platform naming table acknowledges that Google's product renaming creates user confusion — and addresses it in the skill.

  4. Cross-IDE setup automation: agents-cli setup auto-detects and installs to all IDEs simultaneously — the only CLI in the corpus that handles multi-IDE skill installation.

  5. Reference file bundles: The deploy and eval skills bundle reference files (.md files in references/) that agents load on demand during execution — lazy-loading detailed documentation within skills.

Observable Failure Modes

  • Google Cloud dependency: All deployment targets (Agent Runtime, Cloud Run, GKE) require a Google Cloud account — no local deployment option
  • ADK version drift: Skills reference google-agents-cli ~= 0.2.0; version mismatch causes failures — the skill includes version check instructions
  • Context compaction risk: 7 skills + reference files can fill a context window on long sessions — explicitly acknowledged in the workflow skill
  • Two-contributor repo: With only 2 contributors, long-term maintenance risk is higher than typical vendor repos

Cross-References in Batch

  • Uses agentskills.io format (see asyncope-agentskills-io) for skill delivery
  • Can deploy to AgentCore equivalent (Google's Agent Runtime, analogous to agentcore-samples)
  • Deployed to Google Gemini Enterprise (not AWS/Azure/Anthropic)
04

Workflow

google/agents-cli — Workflow

The 8-Phase Development Lifecycle

The google-agents-cli-workflow skill defines 8 phases:

Phase Skill Artifact
0 — Understand (none) .agents-cli-spec.md (read)
1 — Study samples (none) Cloned reference samples
2 — Scaffold /google-agents-cli-scaffold Project skeleton with eval boilerplate + CI/CD
3 — Build /google-agents-cli-adk-code Agent Python code
4 — Evaluate /google-agents-cli-eval Eval results + passing test suite
5 — Deploy /google-agents-cli-deploy Deployed agent endpoint
6 — Publish /google-agents-cli-publish Registered in Gemini Enterprise
7 — Observe /google-agents-cli-observability Monitoring setup

Scaffold-First Enforcement (Approval Gate)

STOP — Do NOT write code yet. If no project exists, scaffold first with agents-cli scaffold create <name>.

This is the primary approval gate — the agent must check for an existing project before writing any code. Skipping scaffold means missing:

  • Eval boilerplate (tests/eval/evalsets/, tests/eval/eval_config.json)
  • CI/CD config
  • Project conventions

Eval-Fix Loop (Phase 4)

The eval phase explicitly requires iteration:

  1. Run agents-cli eval run
  2. Read scores — identify failures
  3. Fix code (prompt, tool logic, instructions, or evalset)
  4. Rerun eval — verify fix
  5. Repeat 3-5 until case passes
  6. Only then add more eval cases

Expect 5-10+ iterations. This is normal — each iteration makes the agent better.

Deployment Target Decision Gate

The deploy skill includes a decision matrix requiring the developer to choose:

  • Agent Runtime (managed, lower complexity)
  • Cloud Run (custom infra, event-driven)
  • GKE (full Kubernetes control)

This is an explicit decision point, not a default.

Session Continuity Instruction

"Re-read the relevant skill before each phase — not after you've already started and hit a problem. Context compaction may have dropped earlier skill content."

This handles the context compaction problem explicitly — agents are instructed to reload skills at each phase boundary.

06

Memory Context

google/agents-cli — Memory & Context

Context Compaction Handling

The google-agents-cli-workflow skill explicitly addresses context compaction:

"Re-read the relevant skill before each phase — not after you've already started and hit a problem. Context compaction may have dropped earlier skill content. If skills are not available, run uvx google-agents-cli setup to install them."

This is a proactive context management instruction — agents are told to reload skill content at each phase boundary rather than assuming prior content is still in context.

Session State File

The skill checks for .agents-cli-spec.md at Phase 0:

"If .agents-cli-spec.md exists in the current directory, read it — it is your primary source of truth."

This file serves as the project memory — accumulated context about the agent being built. Like a CLAUDE.md but agent-specific.

Progressive Skill Loading

The 7 skills implement a progressive loading model aligned with agentskills.io:

  1. Discovery: All 7 skill descriptions are available to the coding agent
  2. Activation: When user says "deploy my agent", google-agents-cli-deploy activates
  3. Execution: Full deploy skill instructions + reference files loaded on demand

Task Tracking (Eval Phase)

"When doing 5+ eval-fix iterations, use a task list to track which cases you've fixed, which are still failing, and what you've tried."

This recommends creating a local task tracking file during intensive eval loops — an ad-hoc persistence mechanism.

No Persistent Memory Service

Unlike AgentCore (which provides managed memory), agents-cli relies on:

  • Files in the project directory (.agents-cli-spec.md)
  • Task lists for eval tracking
  • Standard git history for code changes
07

Orchestration

google/agents-cli — Orchestration

Multi-Agent

Yes — agents-cli targets multi-agent systems as a first-class use case. The google-agents-cli-adk-code skill covers ADK multi-agent patterns including:

  • Sequential agents (A → B → C)
  • Parallel agents (fan-out + aggregation)
  • LLM-as-router (dynamic routing to specialized subagents)
  • Loop agents (iterative refinement)

Orchestration Pattern

Task-decomposition-tree — the skill itself uses a phase-to-skill decomposition (each phase activates a different skill). The agents being built with agents-cli typically use ADK's multi-agent patterns.

Isolation Mechanism

Container (for deployed agents on Cloud Run / GKE) or none (local development). The agents-cli dev command runs locally without isolation.

Multi-Model

Yes for deployed agents (any Google Cloud model on Agent Runtime). The skills themselves run in whatever model the user's coding agent uses — no model specification at the skill layer.

The deploy skill references:

  • --deployment-target agent_runtime → uses Vertex AI models
  • Cloud Run / GKE → any container-compatible model

Execution Mode

  • Skills: Event-driven (activated by task matching)
  • CLI: One-shot per command
  • Deployed agents: Event-driven (HTTP invocations)
  • Local dev: Interactive-loop (agents-cli dev)

Cross-IDE Portability

High — explicitly designed for Gemini CLI, Claude Code, Codex, Antigravity. The agentskills.io-compatible format ensures the 7 skills work across all compliant agents without modification.

Consensus

None. Multi-agent coordination patterns are ADK framework concerns, not agents-cli concerns.

08

Ui Cli Surface

google/agents-cli — UI/CLI Surface

Dedicated CLI Binary

Yesagents-cli is a full Python CLI installed via uv.

Attribute Value
Binary name agents-cli
Install uv tool install google-agents-cli OR uvx google-agents-cli setup
PyPI package google-agents-cli
Is thin wrapper No — own runtime (wraps Terraform, Docker, GCloud, ADK, but has own orchestration)
Subcommands ~18

CLI Subcommands (complete list from README)

agents-cli setup                 — Install CLI + skills to all IDEs
agents-cli login                 — Auth with Google Cloud or AI Studio
agents-cli login --status        — Auth status
agents-cli scaffold <name>       — Create new agent project
agents-cli scaffold enhance      — Add deployment/CI/CD/RAG to existing project
agents-cli scaffold upgrade      — Upgrade to newer version
agents-cli run "prompt"          — Run agent with single prompt
agents-cli install               — Install project dependencies
agents-cli lint                  — Code quality (Ruff)
agents-cli eval run              — Run evaluations
agents-cli eval compare          — Compare two eval result files
agents-cli deploy                — Deploy to Google Cloud
agents-cli publish gemini-enterprise — Register with Gemini Enterprise
agents-cli infra single-project  — Provision infrastructure
agents-cli infra cicd            — Set up CI/CD pipeline
agents-cli infra datastore       — Provision datastore for RAG
agents-cli data-ingestion        — Run data ingestion
agents-cli info                  — Show project config + CLI version
agents-cli update                — Reinstall skills to all IDEs

Local UI / Dashboard

None — no local web dashboard or TUI. Observability is cloud-based (Cloud Trace, Cloud Logging, third-party integrations).

Documentation Site

https://google.github.io/agents-cli/ — built with standard docs tooling, hosted via GitHub Pages.

IDE Integration

IDE Integration Method
Claude Code .claude-plugin/plugin.json
Gemini CLI gemini-extension.json
Codex Standard skills install
Antigravity Standard skills install

agents-cli setup auto-detects installed IDEs and installs skills to each. agents-cli update reinstalls to all IDEs — useful when skills are updated.

Observability (for deployed agents)

  • Cloud Trace + Cloud Logging (built-in)
  • Third-party: Grafana, Datadog, Dynatrace (via integrations)
  • agents-cli observability skill guides setup

Related frameworks

same archetype · same primary tool · same memory type

Claude-Flow / Ruflo ★ 55k

Eliminates single-agent context limits and sequential bottlenecks by orchestrating fault-tolerant swarms of specialized AI agents…

Hermes Agent (NousResearch) ★ 168k

Self-improving personal AI agent with closed learning loop, 7 terminal backends, and messaging gateway — not tied to any AI…

OpenCode ★ 165k

Terminal-first AI coding agent with multi-model routing, native desktop app, and a typed .opencode/ configuration system for…

OpenHands ★ 75k

Open-source AI software development platform (open-source Devin alternative) with Docker sandbox isolation, 77.6% SWE-bench…

DeerFlow ★ 70k

Long-horizon superagent that researches, codes, and creates by orchestrating parallel sub-agents with isolated contexts in Docker…

oh-my-openagent (omo) ★ 60k

Multi-provider AI agent orchestration for OpenCode: escape vendor lock-in by routing Sisyphus (Claude/Kimi/GLM) and Hephaestus…