Security audit

记忆编排器

Security checks across malware telemetry and agentic risk

Overview

This is a disclosed memory-management skill, but users should be careful because it can persist conversation memory, including personal preferences, to disk.

Install only if you want the agent to manage durable memory. Treat memory files and cleanup logs as sensitive, choose a specific storage path, avoid storing secrets or highly sensitive personal data, and review retention/deletion settings before enabling persistence or automated cleanup.

SkillSpector

By NVIDIA

Vulnerability Patterns

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The skill prominently advertises persistent storage, layered retention, and automatic cleanup, but its top-level description does not clearly warn users that conversation-derived data may be written to disk, retained long-term, or deleted automatically. In a memory-management skill, omission of these warnings increases the chance that operators store sensitive user data without informed consent or lose important data through cleanup policies they did not notice.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The cleanup section describes automatic promotion, archival, and deletion behavior, including FIFO eviction and expiry-driven cleanup, without a strong user-facing warning about irreversible data loss. Because this skill is specifically designed to manage user memory and may handle preferences, identity, and project context, silent or poorly signposted cleanup can cause confidentiality and integrity issues through unintended retention changes or destructive deletion.

Ssd 3

Medium

Confidence: 96% confidence
Finding: The examples encourage storing user preferences, identity-related attributes, and decision authority in long-term or important memory with persistence enabled, but provide no minimization, redaction, consent, or policy guardrails. This is dangerous because it normalizes retention of potentially sensitive personal data indefinitely, especially the 'important' tier described as never cleaned and manually modified only.

Ssd 3

Medium

Confidence: 94% confidence
Finding: The save/load workflow explicitly demonstrates saving all memory to a local file and reloading it later, which enables broad retention of conversation-derived user data without any safeguards around scope limitation, encryption, access control, or consent. In the context of an agent memory skill, this materially increases the risk of privacy leakage, excessive retention, and exposure of sensitive data from disk.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.