google/agents-cli
7 skills87k installs19.1k starsGitHub
Install
npx skills add https://github.com/google/agents-cliSkills in this repo
1Google Agents Cli Adk CodeGoogle Agents CLI ADK Code is an agent skill packaged as an ADK 2.0 Workflow API cheatsheet for builders using Google’s Agent Development Kit from the CLI ecosystem. It targets solo developers who already have or are starting Python agent projects and need prerelease-safe upgrade paths, dependency bounds fixes in pyproject.toml, and clarity on experimental limitations such as no Live Streaming and Python 3.11 minimum. The readme emphasizes that scaffolded templates silently stay on 1.x unless the <2.0.0 cap is removed, then documents uv and pip reinstall flows plus a quick import check for google.adk.workflow. Multi-phase usefulness shows up when you validate agent architecture before ship and when you operate agents without corrupting shared storage across major ADK versions. Treat it as focused reference material for Workflow graphs and routes rather than a full product launch or SEO playbook.12.5kinstalls2Google Agents Cli WorkflowGoogle Agents CLI Workflow is the umbrella skill for solo builders shipping agents on Google Cloud with the Agent Development Kit. It tells your coding agent to stop and scaffold first: run `agents-cli scaffold create <name>` for greenfield work or `agents-cli scaffold enhance .` when code already exists, then confirm state with `agents-cli info` so eval boilerplate, CI/CD, and project conventions are not missing. The skill documents the full ADK lifecycle from local run and debug through test, deploy, publish, and monitoring, plus model selection and code preservation rules that keep iterative agent work safe. It is framed as always active whenever someone asks to develop an agent, run locally, debug, test, deploy, or monitor—making it the coordination layer for the rest of the google/agents-cli skill suite. Install agents-cli with `uv tool install google-agents-cli` (or `uvx google-agents-cli setup`) before relying on the workflow steps.12.4kinstalls3Google Agents Cli ScaffoldGoogle Agents CLI Scaffold is the project-bootstrap skill for ADK agents when you need a new repo layout or want to bolt deployment and CI/CD onto existing code. It documents `agents-cli scaffold create` for greenfield agents, `scaffold enhance` to add agents-cli structure to a current directory, and `scaffold upgrade` when templates evolve—along with template choices and deployment targets called out in the scaffold guide. Solo builders should not run create until Phase 0 from the workflow skill clarifies what the agent does, which tools or APIs it needs, and whether they want a prototype or full deployment path. The skill deliberately stops at scaffolding: agent implementation belongs elsewhere in the suite, as do live deploy operations. That separation keeps your agent from generating ADK Python before eval hooks and infrastructure stubs exist, which is the most common failure mode for first-time Cloud agent projects.12.4kinstalls4Google Agents Cli EvalGoogle Agents CLI Eval is procedural knowledge for solo builders shipping Google ADK agents that call google_search or other built-in grounding. The skill corrects a common trap: treating google_search like a custom function tool in evaluation trajectories. In practice, search is injected into the model config and surfaces as grounding_metadata, which still triggers session-level unexpected-tool errors and breaks multi_turn_tool_use_quality. You learn which metrics remain trustworthy—especially adaptive final_response_quality—and how to shape datasets when reference answers cannot be stable across search results. Use it while hardening an agent before launch, when debugging flaky eval pipelines, or when comparing agent variants without rewriting your whole test harness.12.4kinstalls5Google Agents Cli DeployGoogle Agents CLI Deploy guides indie builders through promoting a scaffolded Google ADK agent from repo to Vertex AI Agent Runtime. Unlike typical container-first flows, Agent Runtime accepts source-based bundles: dependencies from uv export, code archived and applied without maintaining a Dockerfile in the default scaffold path. The skill maps AdkApp responsibilities—setup, operation registration, feedback capture, and streaming queries—to what deploy.py actually ships. Terraform owns the reasoning engine resource, including scaling and concurrency knobs in deployment/terraform/service.tf, while CI-friendly lifecycle rules prevent Terraform from fighting live source uploads. Expect explicit prerequisites: scaffold first, then deploy, then lean on observability skills for traces and analytics.12.4kinstalls6Google Agents Cli ObservabilityGoogle Agents CLI Observability teaches solo builders how to make Google ADK agents legible in production GCP. The optional BigQuery Agent Analytics plugin streams structured interaction events through the Storage Write API so you can query sessions, tool lineage, and feedback without bolting on a separate analytics product. Scaffold-time --bq-analytics wires Terraform-backed datasets and GCS offloading; post-scaffold installs follow ADK integration docs. Paired guidance covers always-on Cloud Trace and prompt-response logging assumptions for projects created with google-agents-cli-scaffold. You get SQL examples, schema evolution notes, and dashboard paths suited to indie operators who need eval-ready logs and incident forensics without a dedicated data team.12.4kinstalls7Google Agents Cli Publishgoogle-agents-cli-publish is a Google ADK suite skill for solo builders and small teams who already have a running agent and need it registered with Gemini Enterprise. It walks through prerequisites—a reachable deployment, an existing Gemini Enterprise app in Cloud Console, and for Agent Runtime the deployment_metadata.json produced by agents-cli deploy—and then explains how to run publish with the right flags or interactive prompts. The skill differentiates programmatic registration from interactive usage, compares ADK versus A2A paths, and calls out Cloud Run and GKE setups where you supply the agent card URL directly. Permission gaps for Discovery Engine invokers are called out so publish failures are diagnosable. Install agents-cli via uv tool install google-agents-cli before invoking. Pair this skill after deploy completes and before teammates rely on the agent inside Gemini Enterprise.12.4kinstalls