Agent CLI: AI Builds Agents in Minutes via 7 Skills
Install Agent CLI with one command to give coding agents 7 skills—workflow, scaffold, eval, deploy—for building, testing, and deploying ADK agents from a single English prompt, cutting dev time from days to minutes.
Solves AI Agent Dev Pain Points
Building AI agents wastes tokens as models hunt scattered docs on Agent Development Kit (ADK), Cloud Run integration, and deployment. Agent CLI fixes this by injecting 7 targeted skills into any coding agent (e.g., Cloud Code, Gemini CLI), providing instant context. Result: Agents go from idea to running ADK-based app in minutes, not days, without hallucinated code or manual setup. Trade-off: Requires global PATH setup post-install for seamless access.
7 Skills Enable End-to-End Agent Lifecycle
Agent CLI installs these skills globally via uvx google-agent-cli setup (express mode handles it in seconds):
- Workflow: Forces AI to clarify requirements before coding, preventing unasked-for builds.
- ADK Code: Embeds full ADK API syntax (hundreds of methods), ensuring accurate agent definitions without guesswork.
- Scaffold: Generates project structure, files, folders, dependencies from templates—e.g., agent.py, Dockerfile for Cloud Run.
- Evaluation: Runs unit tests on agent behavior; input sample query + expected output to verify "agent works end-to-end" or flag bugs.
- Deployment: One-command push to Cloud Run, Agent Engine, or custom targets—replaces 2-week DevOps workflows.
- Publish: Registers agent in Gemini Enterprise (org's internal app store) for cross-team use, like sales accessing eng-built agents.
- Observability: Logs production prompts, tool calls, token usage to debug breaks.
These make AI self-sufficient: No more doc-scraping token burn; skills handle complexity.
Demo: Single-Prompt CSV-to-Infographic Agent
In Cloud Code (works with any tool), prompt: "Use Agent CLI to build a simple agent that takes a CSV file and generates an infographic summary."
AI auto-generates:
- design_spec.md (for approval).
- agent.py (FastAPI server with ADK root agent).
- Dockerfile (Cloud Run ready).
- sample_data.csv (for testing).
It smoke-tests ("analyze sample_data.csv"), evaluates, confirms success. Run adk web for live UI: Upload CSV path → agent analyzes → outputs infographic report (overview, viz; upgrade model like Gemini 3.1 Pro for better results). Traces show tool calls; edit code anytime. Deploy next via skill. Full cycle: 1 prompt → scaffolded, tested, runnable agent. Proves portability—not Google-tool locked.