agents-cli — Agent Build/Eval/Deploy Skills for Coders

Practical Notes

• Start local-first: scaffold + run + eval before touching cloud deploy.
• Keep your agent lifecycle reproducible by pinning tool versions in CI.

Mental model: it’s “skills + lifecycle commands”

agents-cli is not a coding agent itself. It’s a set of CLI commands and skills that teach your coding assistant how to:

• scaffold an agent project,
• run and evaluate it,
• deploy and operate it on Google Cloud.

Practical workflow (minimal)

1) Scaffold a project

agents-cli scaffold my-agent
cd my-agent

2) Run the agent locally (one-shot prompt)

agents-cli run "Summarize the latest logs and propose fixes"

3) Add evaluation early

Even if your eval set is small, it prevents regressions:

agents-cli eval run

Team hygiene

• Treat the CLI as the “source of truth” for commands your agent should use.
• Keep a short AGENTS.md in the repo that says which commands are required before PR.
• When the agent changes code, ask it to run agents-cli lint (or your repo’s lint) before it claims done.

FAQ

Q: Is this a replacement for Claude Code/Codex/Gemini CLI? A: No. It augments them with skills and lifecycle commands for agent projects.

Q: Can I use it without a coding agent? A: Yes. The CLI is designed to work standalone from your terminal.

Q: Do I need Google Cloud for local development? A: Not for basic scaffold/run/eval; cloud is mainly for deployment and ops features.