agents-cli
GitHub

Build AI agents that meet your production standards.

Agents CLI teaches your coding agent how to ship them on Google Cloud.

Paste into your coding agent: Follow google.github.io/agents-cli/setup.md to build my first agent

Read the docs

Works with Antigravity CLIClaude CodeCodex& more

The whole lifecycle, one CLI

Your coding agent builds until evals pass, ships to Google Cloud, and learns from production.

  1. A working agent with one toolDevelopment guide
  2. 4.71 of 5, above your 4.5 barEvaluation guide
  3. One command for Agent Runtime, Cloud Run or GKEDeployment guide
  4. One lookup_order call in 125 msCloud Trace guide
You:

Build a support agent that looks up orders.

Using the scaffold skill

Terminal
agents-cli create support-agent --agent adk_python --prototype
✅ Success! Your agent project is ready.
app/agent.py
def lookup_order(order_id: str) -> dict:
    return {"order_id": order_id, "status": "shipped"}

root_agent = Agent(
    name="support_agent",
    instruction="You answer questions about orders.",
    tools=[lookup_order],
)
Coding agent:

I created the project and wrote a lookup_order tool.

You:

Evaluate it before we ship. Aim for 4.5 out of 5.

Using the eval skill

Terminal
agents-cli eval run
Step 1/2: eval generate
Step 2/2: eval grade
RESULT: PASS (custom_response_quality 4.71 >= 4.50)
Eval results
Mean score4.71 / 5Above your 4.5 bar
  1. 5
  2. 5
  3. 4
  4. 5
  5. 5
  6. 4
  7. 5
7 cases, 0 errors
Coding agent:

I added 4 order cases, 7 in total, and ran the evals.

You:

Looks good. Deploy it to Cloud Run.

Using the deploy skill

Terminal
agents-cli scaffold enhance . --deployment-target cloud_run
✅ Enhancement complete!
agents-cli deploy --project my-project
✅ Deployed to Cloud Run.
Deploy target
  • Agent Runtime
  • Cloud Run
  • GKE
support-agentLive

Region us-east1 Revision support-agent-00001-k7d

Traffic100%
support-agent-482910375521.us-east1.run.app
Coding agent:

I added Cloud Run to the project and deployed it.

You:

Is it calling the tool in production?

Using the observability skill

Terminal
agents-cli run --url https://support-agent-482910375521.us-east1.run.app --mode a2a "Where is order 1042?"
Trace explorer
  1. invoke_workflow support_agent2.41 s
  2. invoke_agent support_agent2.40 s
  3. call_llm970 ms
  4. execute_tool lookup_order125 ms
  5. call_llm1.28 s
Coding agent:

I sent a test request and opened its trace.

Your rules, as extensions

An extension adds your checks and skills to agents-cli. Everyone keeps running the same agents-cli deploy.

  • Enforce a policy. Run your checks before every deploy.
  • Teach your coding agent. Add skills that spell out your rules.
How extensions work
  • Team Aagents-cli deploy
  • Team Bagents-cli deploy
  • CIagents-cli deploy
built-in deployCloud Run

Build agents with any framework

Pick your framework. The commands stay the same.

Agents CLI
  • create
  • run
  • eval
  • infra
  • deploy
  • Agent Runtime
  • Cloud Run
  • GKE

Explore the docs