ClawHub
Back to skill

Security audit

Neural Memory Enhanc

Security checks across malware telemetry and agentic risk

Overview

This memory skill is mostly coherent, but it asks agents to automatically retain conversation-derived information without clear user consent or sensitivity limits.

Install only if you are comfortable with an agent keeping durable local memories from your conversations. Treat it as a persistent memory system: avoid using it with secrets, credentials, regulated personal data, or confidential project details unless you have reviewed where the database is stored and how to delete or isolate memories. Also verify the neural-memory package and any optional embedding or LLM configuration before enabling it.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The documentation makes materially inconsistent privacy and dependency claims: it says there is 'Zero LLM dependency' and that no external data transfer occurs, while later requiring LLM support/API and referencing network-related operation. Users may rely on the stronger privacy claims and unknowingly expose conversation data or deploy the skill in environments that do not match its actual trust assumptions.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill instructs the agent to automatically process and persist important conversation segments without requiring explicit user consent or warning about local retention. Because the content includes decisions, errors, preferences, and arbitrary chat-derived facts, the skill can capture sensitive personal, business, or credential-adjacent information in a durable store.

Ssd 3

Medium
Confidence
97% confidence
Finding
The workflow explicitly tells the agent to persist user preferences, errors, decisions, and important conversation segments across sessions. In a memory skill, this materially increases danger because the entire purpose is durable recall, so over-collection and long-lived retention of sensitive natural-language data is not incidental but central to operation.

Ssd 3

Medium
Confidence
95% confidence
Finding
The auto-extraction feature broadly encourages storing arbitrary facts, decisions, and task-like content from free-form conversation without any sensitivity classifier or exclusion policy. In practice, this can vacuum up confidential project details, personal data, internal incidents, or embedded secrets and retain them in searchable memory for future disclosure or misuse.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.