Security audit

memclaw

Security checks across malware telemetry and agentic risk

Overview

This appears to be a legitimate memory skill, but it needs Review because it can persist, migrate, index, and prune sensitive conversation memories with under-specified user consent and control.

Install only if you are comfortable giving this skill control over long-term memory. Before use, require explicit approval for migration, commits that extract durable memories, and non-dry-run maintenance; avoid storing secrets, regulated data, credentials, or sensitive personal details unless there is a clear retention and deletion plan.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (8)

Vague Triggers

Medium

Confidence: 94% confidence
Finding: The skill instructs agents to use MemClaw for all memory operations and to replace the built-in memory system, which is overly broad delegation to a third-party plugin. This can cause unnecessary exposure of conversation data to the plugin, bypass safer defaults, and pressure agents to use the tool even when memory use is not needed or when native mechanisms would be more appropriate.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The migration command is presented as a normal maintenance action without a prominent warning that it may copy, transform, or otherwise affect existing memory data. In a memory plugin context, this is dangerous because an agent could trigger migration without informed user consent, leading to unintended data movement, duplication, privacy issues, or hard-to-reverse state changes.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The guidance explicitly recommends storing user preferences and other potentially sensitive profile information in long-term memory, but it provides no consent, minimization, retention, or sensitivity-screening guidance. In a memory plugin whose purpose is to persist and retrieve conversational context, this omission increases the chance that agents will collect personal data by default and retain it longer than users expect.

Missing User Warnings

Medium

Confidence: 85% confidence
Finding: The document explicitly defines storage locations for user preferences and personal information but provides no guidance on consent, minimization, access controls, retention, or handling of sensitive data. In a memory plugin whose purpose is to persist and retrieve user context across sessions, this omission can lead to over-collection or unsafe retention of personal data, increasing privacy and compliance risk.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The `cortex_add_memory` documentation encourages storing arbitrary message content, including `system` and `assistant` roles, but provides no warning about sensitive data retention, consent, or data minimization. In a memory plugin explicitly positioned as replacing built-in memory, this omission can lead agents or operators to persist secrets, personal data, or other confidential conversation content by default.

Missing User Warnings

Medium

Confidence: 82% confidence
Finding: The `cortex_commit_session` entry says it closes the session and triggers memory extraction, but the warning is minimal and does not clearly communicate the privacy and state-change consequences of doing so. An agent may invoke it routinely without understanding that it finalizes a thread and may transform ephemeral conversation data into durable extracted memory.

Missing User Warnings

Medium

Confidence: 80% confidence
Finding: The `cortex_maintenance` tool can run commands like `prune` and `reindex`, which may alter or remove stored data when `dryRun` is false, yet the documentation lacks a prominent safety warning. This creates risk of unintended destructive actions by automated agents or users who interpret maintenance as harmless housekeeping.

Missing User Warnings

Medium

Confidence: 84% confidence
Finding: The `cortex_migrate` tool imports data from native memory into MemClaw but omits warnings about duplication, unintended retention expansion, privacy boundary changes, or irreversible changes to the memory corpus. In a plugin marketed to replace built-in memory, migration expands the blast radius because historical data may be brought under a new indexing and retrieval model without adequate operator awareness.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.