ClawHub
Back to skill

Security audit

memory-distiller-free

Security checks across malware telemetry and agentic risk

Overview

The skill mostly looks like a local log summarizer, but its documentation mixes in unclear callback, external API, credential, and missing-script instructions that users should review before installing.

Install only if you are comfortable with a skill that can read local memory logs, run a Node.js command, and write summaries back into persistent memory. Before use, require the publisher to clarify whether the skill is local-only, remove or explain callback/API-key instructions, and provide the referenced memory-compress.js script; do not provide API keys or send logs to any callback unless you explicitly intend that data flow.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (4)

Description-Behavior Mismatch

Medium
Confidence
91% confidence
Finding
The skill is presented as a local file-processing tool, but the interface and documentation also introduce a callback URL and imply network interaction. This creates a trust-boundary mismatch: users may authorize a seemingly local log distiller while the skill design permits outbound data flow, which is especially risky because logs may contain sensitive conversation history or operational details.

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The documentation first states the tool is purely local and requires no API key, then later describes API integration and API-key configuration. Such contradictory security claims can mislead operators into exposing credentials or permitting egress they would otherwise forbid, undermining informed consent and safe deployment.

Context-Inappropriate Capability

Medium
Confidence
88% confidence
Finding
A basic local log-distillation skill has no clear need for API-integration and API-key handling guidance, yet the documentation advertises those capabilities. Unjustified credential and external-service features enlarge the attack surface and may normalize secret entry or data exfiltration in a context that is likely to process sensitive logs and memory data.

Missing User Warnings

Medium
Confidence
90% confidence
Finding
The instructions recommend appending generated content directly into MEMORY.md without a prominent warning or confirmation step. In an agent context, this can lead to unintended modification or poisoning of persistent user memory, causing corruption of future context, propagation of bad summaries, or storage of sensitive distilled content in long-lived files.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.