XSafeClaw is an agent safety platform for local and multi-runtime agent systems. It connects runtime trajectory collection, tool-call inspection, human approval, and risk auditing into a unified safety access layer.

My work focused on nanobot-style runtime integration:

  • local agent session discovery;
  • controlled tool invocation;
  • unified platform display;
  • risk-control workflow validation;
  • documentation and engineering support.

The project is useful for presenting a trustworthy AI direction because it combines system building with concrete safety governance.