The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches public signing keys and configuration from verified vendor infrastructure.
scripts/prepare_cli_repository.py downloads the Isaac ROS APT key from https://isaac.download.nvidia.com/isaac-ros/repos.key.
The script verifies the download using a hardcoded SHA-256 digest (d45fbd909e89914960a66dba17d4c659adf908ab4d6fe2cf55381e9ffe5ce59d) to ensure integrity.
SKILL.md and references/release-4.6.md reference a production USD scene asset hosted on an official NVIDIA S3 bucket: https://omniverse-content-production.s3-us-west-2.amazonaws.com/Assets/Isaac/6.0/Isaac/Samples/ROS2/Scenario/g1_ros_navigation.usd.
[REMOTE_CODE_EXECUTION]: The skill manages and executes code from pinned external repositories belonging to the vendor's official GitHub organizations.
scripts/doctor.py prepares checkouts from github.com/nvidia-isaac, github.com/isaac-sim, and github.com/NVIDIA-ISAAC-ROS.
These dependencies are pinned to specific immutable git commit hashes in upstream-versions.lock.json.
The skill uses adapter scripts (e.g., references/change-map/scripts/run.py) to execute entrypoints from these checkouts after verifying the commit hash and worktree state (via shared/dependency_manifest.py).
[COMMAND_EXECUTION]: The skill makes extensive use of system commands for orchestration and environment verification.
scripts/preflight.py executes nvidia-smi, docker info, and ldconfig to discover host capabilities.
scripts/check_runtime.py and scripts/guarded_gantry_release.py use ros2 CLI tools and ROS 2 services to monitor and control the simulation environment.
scripts/play_viewer.py launches a headless Chromium process to interact with the WebRTC stream via the Chrome DevTools Protocol.
[PRIVILEGE_ESCALATION]: Documentation within the skill provides instructions for system-level configuration using sudo.
references/isaac-ros-setup.md provides commands for installing APT packages (sudo apt-get install) and initializing Docker configuration (sudo isaac-ros init docker).
These actions are scoped to the initial bootstrap and prerequisite setup of the host machine.
[SAFE]: The skill implements significant defensive measures to maintain safety boundaries.
shared/simulation_gate.py enforces a "fail-closed" policy that rejects control commands unless it can verify the target is a simulation (checking the USD stage identifier, ROS_DOMAIN_ID isolation, and specific simulation hardware plugins).
scripts/http_transport.py implements a restricted HTTP client that explicitly rejects redirects to prevent server-side request forgery or protocol smuggling.

isaac-humanoid-cloud-control