Thinking Protocol

Security checks across malware telemetry and agentic risk

Overview

This appears to be a broad reasoning-helper skill, with minor scope and language notes but no evidence of unsafe commands, credential use, persistence, or data movement.

Install only if you want a general reasoning-style helper that may influence how the agent structures analysis across a wide range of tasks. Review its activation wording and language behavior if you prefer skills that apply only on explicit request.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (2)

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The skill presents itself as a broadly applicable 'comprehensive thinking protocol' and injects persistent behavioral instructions without a narrowly scoped activation condition. In agent systems, overly broad activation can cause this skill to apply to unrelated tasks, overriding local task requirements, increasing prompt-surface area, and creating instruction conflicts that may affect safety, reliability, or policy compliance.

Natural-Language Policy Violations

Low

Confidence: 78% confidence
Finding: The instruction to 'Think in natural language (Chinese/English based on query)' imposes a language-selection policy without explicit user consent. While not directly dangerous, it can cause unintended behavior, mismatch user expectations, and interfere with higher-priority instructions or platform language policies in multilingual contexts.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal