Google's skill for deploying agents with agents-cli to Agent Runtime, Cloud Run or GKE: choosing a target, CI/CD pipelines, service accounts, secrets, rollback, Agent Gateway and production troubleshooting.
Agents CLI Project Scaffolding Skill
Google's agents-cli skill for creating, enhancing and upgrading ADK agent projects: scaffold create with the adk_python or adk_go template, deployment targets (Agent Runtime, Cloud Run, GKE), CI/CD runners and a prototype-first workflow.
Install
Other ways to install
About
- agents-cli scaffold create, enhance and upgrade commands
- adk_python and adk_go templates with A2A built in; extension templates such as LangChain
- Deployment targets: Agent Runtime, Cloud Run and GKE, plus GitHub Actions or Cloud Build CI/CD
- Prototype-first workflow: --prototype now, scaffold enhance later
- Smart-merge upgrades with dry-run, auto-approve and force fallbacks
- Maps legacy Vertex AI product names to current CLI flags
Agents CLI Project Scaffolding is the skill in Google's google/agents-cli suite that owns project setup. It teaches an agent to use the agents-cli scaffold commands to start a new Agent Development Kit (ADK) project, add deployment or CI/CD to an existing one, or upgrade a project to a newer agents-cli release, and it hands agent code to google-agents-cli-adk-code and deployment operations to google-agents-cli-deploy.
For a new project the agent must first clarify requirements through the google-agents-cli-workflow skill. It then maps the user's choices to flags: --agent adk_python (the default) or adk_go, --deployment-target agent_runtime, cloud_run or gke, --cicd-runner github_actions or google_cloud_build, a session store, and --prototype to skip Terraform and CI/CD. It translates older product names (Vertex AI Agent Engine and similar) into current flag values, notes removed flags, and explains that retrieval, sandboxed execution, memory, OAuth and guardrails come from clone-and-study recipes rather than templates; third-party templates such as LangChain are scaffolded from the google/agents-cli-extensions repository.
It lists the practical constraints: project names of at most 26 lowercase characters, never creating the directory first, passing the right guidance filename (CLAUDE.md, GEMINI.md or AGENTS.md) and --agent-directory when code is not in app/. Upgrades use smart-merge, with --dry-run, --auto-approve and --force as fallbacks. It recommends a prototype-first path, checking an agent with agents-cli run and agents-cli eval run rather than pytest assertions on model output, and scaffolding a throwaway reference project in /tmp to borrow Terraform or CI files. Requires agents-cli (uv tool install google-agents-cli). Apache-2.0.
SKILL.md
The instructions your agent loads, as published by Google.
Related Skills
See all Agents CLI Project Scaffolding alternativesGoogle's entrypoint skill for building ADK agents with agents-cli: a phased lifecycle from spec and recipes through scaffold, build, eval, deploy, publish and observe, with code-preservation rules and a debugging routine.
Google's skill for writing agent code with the Agent Development Kit (ADK) in Python or Go: agent types, tools, callbacks, state, multi-agent orchestration, the graph Workflow API and a catalogue of reference recipes, used with agents-cli.
Skill: Netlify Deploy
by Netlify
Netlify's official deploy skill: netlify.toml deploy contexts and their precedence, CLI and drag-and-drop deploys, Deploy Previews, build hooks, skew protection, and secrets-scanning build failures.
