The Semantic Handshake Engine

• Outcome: Review
• Latest audit: May 29, 2026 at 9:26 PM UTC
• Version:
  1.0.0

Overview

This skill is not clearly malicious, but it creates stored IoT control templates from LLM output with limited safety, validation, and privacy guidance.

Install only if you understand that this is for IoT adapter generation, not a harmless document-only skill. Do not paste secrets or proprietary API details unless you trust the local model server, and manually inspect any generated method, endpoint, payload, and target device before allowing another system to use the saved template on real hardware.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The document promotes autonomous AI execution over physical home systems while providing no concrete safety, privacy, fallback, or human-override requirements. In the context of HVAC, power, and other environmental controls, omission of these constraints can normalize unsafe deployments that may cause physical harm, property damage, or loss of operator control.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The whitepaper explicitly includes infrared thermal imaging, mmWave radar point clouds, and device/network telemetry, all of which can reveal occupancy, behavior, and sensitive in-home activity. Because the document lacks privacy boundaries, consent, retention, minimization, or access-control guidance, it encourages surveillance-capable designs without adequate protections.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The skill explicitly promotes dynamic generation of executable control templates for physical IoT devices, but it provides no warning about real-world consequences such as unintended device actuation, unsafe state changes, or network-side effects. Because the generated output is intended to be executed later against physical systems, omission of safety constraints and operator warnings increases the likelihood of unsafe or unintended actions.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The usage section instructs users to run the script and exercise device-control scenarios, including triggering smart-device interactions, without any warning about network access, authorization requirements, or physical effects. In a skill whose purpose is to control IoT devices, encouraging execution of test cases without safety guidance materially increases the risk of unauthorized device interaction or accidental actuation.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The skill transmits user-supplied device descriptions and API/protocol documentation to a local HTTP LLM endpoint without any privacy notice, consent flow, or transport security. Even though the destination is localhost, this can expose sensitive internal API details, credentials embedded in docs, or proprietary device information to another service on the host, and plaintext HTTP also weakens trust boundaries if the local service is replaced, proxied, or logged.

VirusTotal

66/66 vendors flagged this skill as clean.