ClawHub

OpenClaw Memory Qdrant

Security checks across malware telemetry and agentic risk

Overview

This is a transparent semantic memory skill, with privacy-sensitive memory behavior that is disclosed and mostly user-controlled.

Install this only if you want agent memory. Keep autoCapture off unless you explicitly want conversations saved, avoid storing secrets, and use only in-memory mode or a trusted Qdrant server for sensitive data. Expect the first run to download an embedding model from Hugging Face.

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 (3)

Description-Behavior Mismatch

Medium
Confidence
93% confidence
Finding
The plugin advertises itself as fully local, but `pipeline('feature-extraction', 'Xenova/all-MiniLM-L6-v2')` may fetch model artifacts from an external source on first use. That creates an undeclared network dependency and potential privacy/supply-chain exposure, especially in environments that expect strict offline behavior.

Missing User Warnings

High
Confidence
97% confidence
Finding
When `autoCapture` is enabled, the plugin iterates over user messages and persists selected content to long-term memory automatically, with no explicit consent, notice, or per-item confirmation. This can silently retain sensitive personal data such as contact details, preferences, or decisions beyond the current session, increasing privacy and data-retention risk.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
With `autoRecall` enabled, stored memories are injected into the agent prompt context without explicit disclosure to the user at the time of use. Even though the code attempts prompt-injection hardening, hidden context augmentation can expose prior sensitive information unexpectedly and influence model behavior in ways the user cannot observe or control.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal