ClawHub
Back to skill

Security audit

Changde AI Physical Alignment Architecture (CPAA)

Security checks across malware telemetry and agentic risk

Overview

The skill only runs a local JSON compliance check, but its documentation recommends immediately cutting actuator power in real hardware without safe-stop safeguards.

Review carefully before using this with robots, motors, or other physical equipment. Treat it as a configuration checker only unless you add a hardware-specific safe-stop procedure, operator controls, and validated interlocks before any actuator power changes.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (1)

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill explicitly recommends that a failed compliance check should immediately cut actuator power at the hardware layer, but it does not require any warning, safe-stop sequence, or hazard analysis. In a physical/embodied AI context, abrupt power loss can create unsafe states such as dropped loads, loss of braking, unstable motion, or interruption of critical processes, making this materially dangerous.

VirusTotal

61/61 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.