ClawHub
Back to skill

Security audit

Xiaomi

Security checks across malware telemetry and agentic risk

Overview

This Xiaomi smart-home skill is purpose-related, but it needs review because it handles reusable device-control tokens and can change real appliance states with limited safeguards.

Install only if you are comfortable letting an agent control real Xiaomi devices. Treat Xiaomi account credentials, device IPs, and especially tokens as secrets: do not commit them, paste them into shared chats, or store real values in ordinary markdown unless the files are private and access-controlled. Review any token_extractor.py before running it, because it is referenced but was not included in the inspected artifact.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (4)

Description-Behavior Mismatch

Medium
Confidence
89% confidence
Finding
The skill is presented as local LAN control, but it also documents pulling device IPs and long-lived tokens from Xiaomi Cloud, which materially expands the trust boundary and sensitive-data exposure. That mismatch can mislead users and reviewers into granting or storing credentials they did not expect to be necessary for a supposedly local-only skill.

Context-Inappropriate Capability

Medium
Confidence
92% confidence
Finding
Including Xiaomi Cloud token extraction in a skill whose stated purpose is LAN device control introduces access to sensitive account-linked device credentials that can enable broad control over household devices. Because these tokens are reusable secrets, exposing or mishandling them can let an attacker enumerate devices and issue commands beyond the narrow user action originally intended.

Vague Triggers

Medium
Confidence
84% confidence
Finding
The natural-language mappings are broad enough that an agent could trigger real-world device actions from loosely phrased user text without strong confirmation or device scoping. In a home-automation context, ambiguous activation can cause unauthorized or accidental control of appliances such as heaters, humidifiers, or cookers, creating both safety and privacy risks.

Missing User Warnings

High
Confidence
95% confidence
Finding
The documentation instructs users to extract and store device IPs and 32-byte tokens, but provides no warning that these are sensitive credentials granting direct control over physical devices. Storing such secrets in references files or handling them casually increases the chance of accidental disclosure through logs, repositories, screenshots, or other agent-accessible context.

VirusTotal

56/56 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.