agents-cli: Google's 7-skill agent stack (2026)

Install google/agents-cli if your agent work already touches Google ADK (Agent Development Kit), Gemini, or Google Cloud deployment. Skip it as a default skill pack if your team is outside that stack, because the official signal is much stronger than the public user-feedback signal.

This week's pick is Google's own agents-cli, an Apache 2.0 open-source repository with about 5.1K GitHub stars, 538 forks, 31 commits, and 13 releases in the current snapshot. 1 The timing is the reason to look now: v1.0.0 became the first generally available release on June 30, 2026, and v1.1.0 landed on July 10 with guided brainstorming, cleaner eval output, Windows fixes, and refreshed bundled skills. 2

The short verdict: pilot it for Python ADK agents and Google Cloud shipping workflows. For a general Claude Code, Codex, Cursor, or Cline user who only wants better local coding behavior, this is probably too cloud-specific to install everywhere.

What it gives your agent

agents-cli is a tool for coding agents, not a coding agent itself. Google's README describes it as a way to "Turn your favorite coding assistant into an expert at building and deploying agents on Google Cloud." 1

The package installs seven Agent Skills: workflow, ADK code, scaffold, eval, deploy, publish, and observability. 3 Together, those skills cover the lifecycle that usually gets scattered across docs: brainstorming the agent spec, creating a project, writing Agent Development Kit code, evaluating behavior, deploying to Agent Runtime, Cloud Run, or Google Kubernetes Engine, registering in Gemini Enterprise, and wiring observability through Cloud Trace, logging, and analytics. 1 4

The most interesting part is not the scaffold command. It is the eval loop. The eval skill describes a five-stage Quality Flywheel: prepare data, run inference, grade traces, analyze failures, and optimize prompts or behavior. 5 TechTimes reached the same conclusion in its July 1 review, calling the evaluation pipeline "the most technically significant part of agents-cli." 6

That makes agents-cli different from many skill packs in this channel. It is less a style guide and more a workflow layer for shipping an agent.

Install path

The fastest full install is one command:

uvx google-agents-cli setup

That path installs the CLI plus the bundled skills into the coding-agent environment. 1 If you only want the skills and want your agent to handle the rest, use the skills registry path instead:

npx skills add google/agents-cli

The README documents npx skills add google/agents-cli as the skill-only route. 1 The Python package route is uv tool install google-agents-cli, and PyPI lists the package as google-agents-cli with a 1.1 MB wheel in the current release. 7

Check the prerequisites before you run it. The documented setup expects Python 3.11 or newer, uv, and Node.js. 1 Local development tasks such as scaffold, run, and eval can use an AI Studio API key, while deployment and cloud features require a Google Cloud project. 1

If you use...	Start here
You want CLI + skills	Run `uvx google-agents-cli setup`. 1
You only want Agent Skills	Run `npx skills add google/agents-cli`. 1
You manage Python tools directly	Run `uv tool install google-agents-cli`. 7
You need to refresh skills without auth	Run `agents-cli setup --skip-auth`. 4

The supported-agent claim is broad: the project says it works with Antigravity CLI, Claude Code, Codex, and any coding agent that supports the SKILL.md standard. 1 Treat Cline and other compatible tools as a standards question, not a separate Google-maintained integration, unless their own docs explicitly list this package.

First workflows to test

Start with the new Phase 0 flow. v1.1.0 changed the workflow skill so the agent can run an interactive brainstorming dialogue before code is written, and it can surface assumptions for review when it cannot ask follow-up questions. 2

Use the google-agents-cli-workflow skill.
Help me design a small customer-support triage agent before writing code.
Ask only the questions needed to define inputs, tools, success criteria, deployment target, and evaluation cases.

Then test scaffold on a disposable project. The scaffold skill supports create, enhance, and upgrade, with deployment targets including Agent Runtime, Cloud Run, GKE, and none. 8

Use the google-agents-cli-scaffold skill.
Create a minimal ADK Python agent for FAQ retrieval.
Target local development first; do not configure cloud deployment yet.
Return the generated file tree and the command to run the agent locally.

The serious test is evaluation. The workflow skill labels evaluation as "the most important phase" and says not to skip it after building the agent. 4 The eval skill also tells users to expect 5 to 10 or more iterations during optimization. 5

Use the google-agents-cli-eval skill.
Generate a small evaluation dataset for this agent with happy-path, ambiguous, and refusal cases.
Run inference, grade the traces, cluster failures, and suggest one prompt or tool-use change.

Only test deploy after the local loop works. The deploy skill includes a decision matrix across Agent Runtime, Cloud Run, and GKE, and it notes that Agent Runtime deployment can take 5 to 10 minutes. 9

Feedback so far

The official ecosystem signal is strong. Google Cloud's quickstart, last updated July 10, walks through setup, scaffold, build, evaluate, deploy, and observe with Agents CLI and ADK. 10 Product Hunt listed agents-cli this week as a free developer tool and AI infrastructure tool, with 169 followers and the tagline "The CLI your coding agent uses to ship agents." 11

The independent feedback is thinner. Cole Medin's June 11 video said, "Building an AI agent is easy these days. The wild part is that shipping one to production is finally just as easy too," and the available YouTube data put that video at 6,792 views and 215 likes. 12 Google-run videos had a larger footprint: the three official videos in the snapshot totaled 46,561 views and 1,246 likes. 13 14

X/Twitter discussion exists, but it is still early-adopter sized. Kirk Marple shared an agents-cli article on July 11 and drew 88 likes, 15 reposts, 192 bookmarks, 10 replies, and 16,737 views in the captured snapshot. 15 EveryDev.ai framed the pitch plainly: "Building AI agents on Google Cloud still means wiring up scaffolding, evals, CI/CD, and deploys by hand. agents-cli by @GoogleOSS handles all of it from one open-source CLI." 16

The best skeptical signal came from another X/Twitter post, not a long review: Abhishek Shinde wrote, "For some reason Google has 3 different coding CLI agents." 17 That is the caveat to keep in mind. agents-cli is clear once you see it as a skill and lifecycle layer, but Google's agent-tooling names can still blur together.

Do not count Product Hunt as user validation yet. The Product Hunt reviews page had zero reviews in the captured listing. 18

Limitations before adoption

The first limitation is maturity. The README still flags the feature as subject to Google Cloud's Pre-GA Offerings Terms, even though the release notes call v1.0.0 generally available. 1 2 That mixed status is enough reason to pilot before putting it into a production release path.

The second limitation is contribution flow. Google's contributing file says pull requests are not currently accepted, and feedback goes through issues instead. 19 That is fine for a Google-owned tool, but it means outside fixes do not move like a normal community-maintained GitHub project.

The third limitation is stack fit. The ADK code skill says Python is the only supported language for now, with other languages coming later. 20 If your agents are TypeScript-first or framework-agnostic, the workflow concepts may still help, but the code path is not aimed at you yet.

The fourth limitation is cloud dependency. Local scaffold, run, and eval can work with an AI Studio API key, but deployment and cloud features require a Google Cloud project. 1 If your team avoids Google Cloud, the strongest parts of this package become less useful.

The fifth limitation is the RAG path. v1.0.0 removed the previous RAG scaffold template, --datastore flag, infra datastore, and data-ingestion commands, moving RAG users toward a clone-and-study recipe instead. 2 That is workable for experienced ADK users, but it is no longer a one-command RAG starter.

Quick reference

Repository	google/agents-cli 1
Maintainer	Google. 1
License	Apache 2.0. 1
Latest release in snapshot	v1.1.0 on July 10, 2026. 2
Current scale	About 5.1K stars, 538 forks, and 13 releases. 1
Skill count	Seven bundled skills: workflow, ADK code, scaffold, eval, deploy, publish, observability. 3
Best first test	Run Phase 0 design, scaffold a local ADK Python agent, then force one eval loop before deployment. 4 5
Install if	You build ADK agents, want Google Cloud deployment support, or need an eval-first workflow inside Claude Code, Codex, Antigravity CLI, or another `SKILL.md`-aware agent. 1
Skip for now if	Your stack is non-Python, non-Google Cloud, or you need mature community reviews before adding a skill pack to production workflows. 18 20

Install agents-cli as a controlled pilot, not as a blanket default. Give it one small ADK agent, one eval dataset, and one deployment target. If the eval loop changes the way your agent behaves before you ship, the package has earned a permanent slot.

Cover image: Product Hunt cover from agents-cli.

agents-cli: Google's 7-skill agent stack