费曼.skill - 理查德·费曼思维操作系统

Security checks across malware telemetry and agentic risk

Overview

This is an instruction-only Feynman-style roleplay skill with no code, credential access, persistence, or hidden system changes, though its real-person impersonation style needs user awareness.

Install only if you want a Feynman-inspired roleplay advisor. In serious, educational, or public-facing contexts, ask the agent to keep saying it is using a Feynman-inspired perspective rather than speaking as Richard Feynman himself.

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 (2)

Natural-Language Policy Violations

Medium

Confidence: 87% confidence
Finding: The skill hard-codes a specific output style and Chinese phrasing rules without requiring current-user consent, which can override user preferences and reduce transparency about why the assistant is responding in that style. In isolation this is not a high-severity security flaw, but it can degrade user agency and make downstream responses less aligned with explicit instructions, especially when combined with strong persona constraints.

Ssd 1

Medium

Confidence: 95% confidence
Finding: The skill instructs the model to respond directly 'as Richard Feynman,' use first-person identity, and avoid meta-level clarification after the first disclaimer. That creates impersonation and disclosure risks: later turns may present synthetic statements as if they came from the real person, making it easier for users to misattribute authority or miss that this is a simulation.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal