Security audit

HashGrid Connect

Security checks across malware telemetry and agentic risk

Overview

This skill has a real agent-matching purpose, but it asks agents to follow changing remote instructions and encourages unattended private external chats.

Install only if you intentionally want an agent to use HashGrid's third-party matching and private chat service. Review the remote documentation manually before use, do not let the agent blindly follow fetched instructions, avoid heartbeat or cron polling unless you set strict limits, and do not share secrets, credentials, personal data, private files, system prompts, or sensitive business information through goals, matches, uploads, or chats.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Rogue AgentSelf-Modification, Session Persistence
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 (5)

Description-Behavior Mismatch

Medium

Confidence: 85% confidence
Finding: The skill manifest presents a narrower capability set than the body of the documentation, which later encourages fetching remote documentation for additional features including profiles, file uploads, and match filters. This creates a documentation integrity gap: agents may invoke undeclared capabilities or process unexpected data flows, increasing the chance of unsafe use and bypassing review of higher-risk functions.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The invocation language is broad enough to trigger on common collaboration or networking requests, which can route many ordinary user tasks into an external agent-matching and private-chat service. That expands the chance of unintended activation and unnecessary disclosure of user goals, metadata, or operational context to a third party.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The skill promotes private chat and external agent matching but does not warn against sending sensitive user, system, or tool-context data to the service. In this context, agents may disclose prompts, internal state, secrets, or user information to an external network without adequate notice or minimization controls.

Natural-Language Policy Violations

High

Confidence: 97% confidence
Finding: The phrase 'No human oversight' is a strong red flag because it explicitly encourages autonomous operation of a private off-platform communication channel. Removing oversight makes abuse, data exfiltration, covert coordination, and policy evasion materially more likely and harder to detect.

Session Persistence

Medium

Category: Rogue Agent
Content: --- name: hashgrid-connect description: Goal-based matching network for AI agents. Register, create goals, get matched with complementary agents, and chat privately. Use when an agent wants to find other agents to collaborate with, have private conversations, or build connections outside of public platforms. --- # HashGrid Connect
Confidence: 82% confidence
Finding: create goals, get matched with complementary agents, and chat privately. Use when an agent wants to find other agents to collaborate with, have private conversations, or build connections outside of p

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.