vm0_skill

Overview

This skill provides shell-based access to the VM0 API for executing AI agents in secure sandboxed environments. Use it to start runs, monitor progress, retrieve logs and metrics, and download inputs (volumes) and outputs (artifacts). It is focused on command-line curl workflows that integrate into scripts and CI pipelines. Prefer the vm0 CLI skill when available for higher-level convenience.

How this skill works

The skill issues authenticated HTTP requests to the VM0 REST API using your VM0_TOKEN environment variable. It supports listing agents, creating and cancelling runs, fetching run status and logs, and downloading volumes and artifacts as tar.gz archives. Commands are optimized for piping to jq and for handling 302 redirects (curl -L) when downloading files.

When to use it

• Run agents in isolated, reproducible sandboxes from scripts or CI.
• Automate large batches of agent executions and capture logs and metrics.
• Inspect and download inputs provided to runs (volumes) for debugging or reproducibility.
• Collect and download run outputs (artifacts) for downstream processing.
• Monitor run health and cancel problematic executions programmatically.

Best practices

• Store VM0_TOKEN in a secure environment variable and avoid committing it to source control.
• Wrap curl calls in bash -c '...' when piping to other commands to preserve environment variables.
• Use cursor-based pagination for list endpoints to handle large result sets gracefully.
• Download volumes and artifacts with curl -L and extract locally with tar -xzf for inspection.
• Include run-specific metadata or variables in the run payload for traceability.

Example use cases

• Start an agent run that processes an uploaded dataset and then download the resulting artifact for analysis.
• Stream and filter run logs to detect failures early and trigger automated rollbacks.
• List available agent versions to select a reproducible environment for production runs.
• Download an input volume to reproduce a run locally for debugging.
• Cancel long-running or stuck runs from an automation script.