Skip to content

AGENTS.md

Latest commit

 

History

History
55 lines (44 loc) · 3.62 KB

AGENTS.md

File metadata and controls

55 lines (44 loc) · 3.62 KB
Raw

Isaac Automator - Agent Instructions

This repository is Isaac Automator: a command-line tool that deploys Isaac Sim, Isaac Lab, and Isaac Lab Arena to public clouds (AWS, GCP, Azure, Alibaba Cloud) as ready-to-use GPU Isaac Workstations (a remote desktop cloud VM with the Isaac apps installed, plus start/stop/destroy lifecycle and data transfer).

If you are an AI agent working with this repository, first decide your role:

  • Operating Isaac Automator - you want to use the tool: deploy a workstation, connect to it, run a demo, stop/start it, and tear it down. Read ai/automator.agent.md and follow the procedures in ai/skills/. That is the entry point for operator agents. Start there.
  • Developing Isaac Automator - you want to change the tool itself (its Python CLI, Terraform, or Ansible). See README.md and CONTRIBUTING.md. The operator docs are not for editing source.

Operator quickstart (the short version)

The full procedures are in ai/skills/; this is the orientation.

  1. Prerequisite: Docker running, cloud credentials available, then ./build once.
  2. Everything runs in a container. The top-level scripts (./deploy-aws, ./start, ./novnc, ...) wrap commands that run inside the isaac_automator Docker image. Run them from the repo root.
  3. Deploy: ./deploy-<cloud> with options; e.g. install Isaac Lab and a demo in one shot. See ai/skills/deploy-workstation.skill.md.
  4. Connect: ./novnc <name> (2D desktop), NoMachine (live 3D viewport), or ./ssh <name> (shell). See ai/skills/connect-workstation.skill.md.
  5. Control cost / clean up: ./stop <name>, ./start <name>, and ./destroy <name> --yes. See ai/skills/manage-lifecycle.skill.md.

Rules for agents (read before acting)

  • It provisions real, paid cloud infrastructure. Treat every deploy/start as live-fire. Use the cheapest viable GPU instance unless told otherwise.
  • Always clean up what you create. Stop when idle; destroy (./destroy <name> --yes) when done - stopped instances still bill for storage, only destroy stops all cost.
  • Run non-interactively. The deploy commands prompt for any required option you omit; an unanswered prompt hangs an agent forever. Pass every required option on the command line, and use --existing replace (not ask). For a headless invocation that does not need a TTY, call the container directly: docker run --rm --network host -v "$(pwd)":/app isaac_automator "<command>".
  • Never print, paste, or commit credentials. Pass them via environment variables (e.g. AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY) and reference them by name only.
  • Viewport over remote desktop: noVNC shows the 2D desktop but generally not the live Isaac Sim 3D viewport (Omniverse Kit renders via a Vulkan surface VNC does not capture). Use NoMachine for live 3D, or capture rendered output headlessly with the app's own recorder. Details in ai/skills/connect-workstation.skill.md and ai/skills/troubleshoot.skill.md.

Map

  • ai/automator.agent.md - operator agent entry point (mental model, golden rules, skills index).
  • ai/skills/ - step-by-step operator procedures (deploy, run-demos, connect, lifecycle, transfer-data, troubleshoot).
  • README.md - full human documentation for using and understanding the tool.