vm0-cli_skill

Overview

This skill provides CLI workflows for building, deploying, and running AI agents inside secure VM0 sandboxes. It covers installing the vm0 CLI, creating agent projects, managing volumes and artifacts, configuring model providers, and running agents both interactively and in CI/CD. Use it to automate agent lifecycles and handle inputs, outputs, secrets, and logs from the command line.

How this skill works

The skill exposes vm0 CLI commands to initialize projects, compose deployment configs (vm0.yaml), and run agents with prompts, env files, variables, and secrets. It also manages storage: volumes for input files and artifacts for outputs, supports provider setup for LLM keys, and surfaces logs, metrics, and usage stats. All commands support non-interactive flags for scripts and CI pipelines.

When to use it

• Install or verify the vm0 CLI on a workstation or CI runner
• Bootstrap a new agent project and generate vm0.yaml
• Deploy agent configurations to the VM0 platform
• Run agents with prompts, environment files, and artifact/volume bindings
• Manage input volumes and agent output artifacts
• Automate agent runs in CI/CD or resume/continue long-running sessions

Best practices

• Authenticate first (vm0 auth login or set VM0_TOKEN in CI)
• Use vm0 init to create a correct project structure and vm0.yaml
• Push volumes before runs and pull artifacts after runs to sync data
• Store secrets via vm0 secret and reference them in vm0.yaml as ${ { secrets.NAME } }
• Use non-interactive flags (-y, --env-file, --name, --credential) in CI to avoid prompts
• Inspect logs (vm0 logs) and usage (vm0 usage) for debugging and cost tracking

Example use cases

• Local development: vm0 init, edit vm0.yaml, vm0 compose vm0.yaml -y, then vm0 cook "Analyze code"
• Batch CI run: export VM0_TOKEN, vm0 compose vm0.yaml -y, vm0 run my-agent --artifact-name report --env-file=.env
• Data-processing: vm0 volume init && vm0 volume push to provide datasets, then vm0 run my-agent "Process files" --volume-version input-data=latest
• Model provider setup: vm0 model-provider setup --type openrouter-api-key --credential "sk-or-xxx" and set-default
• Debugging: vm0 run my-agent "Task" -v then vm0 logs <run-id> --metrics to inspect issues