The Agent Skills Directory

[SAFE]: The skill implements a 'Non-negotiable hardware boundary,' explicitly instructing the agent never to command physical equipment automatically. It mandates a multi-step human review process for all physical actions, including reconciling deck layouts and verifying calibrations.
[SAFE]: Bundled utility scripts (_common.py, validate_manifest.py, etc.) include security-hardened file handling. The safe_input_path function effectively prevents path traversal and symlink attacks by resolving paths relative to the current working directory and checking for '..' components.
[SAFE]: Dependency management is handled via pinned versions from a well-known registry (PyPI), specifically PyLabRobot==0.2.1. The skill provides clear instructions for isolated environment setup using uv.
[SAFE]: The Visualizer component uses localhost-only networking (127.0.0.1) for browser-based rendering of robot states, which is a standard and acceptable practice for development tools.
[SAFE]: The skill's architecture separates protocol planning (data) from execution (code). The simulation plans generated are non-executable JSON data, minimizing the risk of accidental or malicious code execution during the review phase.

pylabrobot