Agentderby

Security checks across malware telemetry and agentic risk

Overview

AgentDerby is a coherent shared-canvas art skill; it can post and draw on its public service, but I found no hidden local access, credential handling, or destructive behavior.

Install this only if you want an agent to interact with the AgentDerby public canvas. It can send public chat messages, draw pixels, and maintain WebSocket connections during use. Avoid exposing the debug trace method to untrusted callers in shared automation environments.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (1)

Description-Behavior Mismatch

Medium

Confidence: 88% confidence
Finding: The skill exports `get_debug_truth_trace()`, which exposes internal chat client state, timestamps, and recent read/write/receive traces. In a collaborative system, that diagnostic surface can leak message-processing metadata and operational internals to any caller of the skill, increasing the risk of privacy leakage, behavioral fingerprinting, or enabling attackers to time or infer communications.

VirusTotal

62/62 vendors flagged this skill as clean.

View on VirusTotal