MindVault - 思维永生

• Outcome: Review
• Latest audit: Aug 5, 2026 · 7:53 AM UTC
• Version: 1.0.5

Overview

MindVault is a chat-archiving memory skill, but users should review it because its local-only privacy claims conflict with documented cloud platform support and it persistently stores sensitive conversation history.

Install only if you are comfortable with your conversation history being copied into persistent local archive files and derived memory files. Use it in a scoped project directory, avoid chats containing credentials or regulated data, review archives before reuse or sharing, and do not rely on the local-only claim for Coze or any paid automation/reporting workflow unless its data flow is separately documented.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Behavioral AST: exec() Call, eval() Call, Dynamic Import
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (8)

Description-Behavior Mismatch

Medium

• Confidence: 86% confidence
• Finding: The manifest promises that all data remains local and is never uploaded to the cloud, yet the platform support section explicitly includes Coze via API retrieval and a cloud knowledge base. This mismatch can cause users to expose sensitive conversation archives under a false privacy assumption, which is a meaningful security and privacy risk for a memory/archiving skill.

Intent-Code Divergence

Medium

• Confidence: 80% confidence
• Finding: The documentation states backup files are stored in a local archive directory, but other sections describe alternate storage backends such as Marvis data.db and Coze API/cloud knowledge bases. For a tool handling conversation history, inconsistent storage descriptions undermine users' ability to assess where sensitive data resides and increase the risk of accidental disclosure or improper retention assumptions.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The README promotes permanent conversation archiving, rule extraction, and detailed recovery of prior context, but does not warn that this can capture secrets, personal data, credentials, internal code, or regulated information from prior chats. In a skill with read/write/exec permissions and an explicit memory system, the lack of privacy boundaries, retention guidance, and redaction controls materially increases the risk of sensitive data persistence and later unintended reuse or exposure.

Vague Triggers

Medium

• Confidence: 83% confidence
• Finding: The manifest declares very broad activation keywords such as long-term memory, thinking protocol, self-check loop, search-first, and data assets, which can match many ordinary user requests unrelated to this skill’s specific function. In an agent ecosystem, overbroad triggers can cause the skill to be invoked unexpectedly, increasing exposure to its read/write/exec permissions and expanding the attack surface for prompt steering or unintended file operations.

Missing User Warnings

Medium

• Confidence: 89% confidence
• Finding: The README advertises automatic archival and weekly report sending, but does not clearly warn users that conversation contents may be persistently stored and potentially transmitted or delivered elsewhere. In a skill with read/write/exec permissions and explicit memory/archive behavior, this omission can lead to unintentional retention or disclosure of sensitive conversations because users may not realize these actions are happening automatically.

Missing User Warnings

Medium

• Confidence: 89% confidence
• Finding: The script exports full conversation content, tool outputs, attachment paths, and indexes to disk with no explicit consent prompt, redaction step, or sensitivity warning. In a memory/archive skill with read/write/exec permissions, this increases the risk of unintentionally persisting secrets, local file paths, internal prompts, and other sensitive material into broadly accessible archive files.

Ssd 3

High

• Confidence: 92% confidence
• Finding: The script is explicitly designed to archive complete user conversation content, including attachment paths and system-style messages, into export files. In the context of a long-term memory skill with local read/write/exec permissions, this creates a real data exposure risk because sensitive prompts, credentials, internal instructions, or private filesystem locations may be retained longer than intended and later accessed by other tools, users, or processes.

Ssd 3

Medium

• Confidence: 84% confidence
• Finding: Incremental archiving preserves a cumulative historical record across runs, which increases the blast radius of any accidental retention or later compromise of the archive directory. While persistence is part of the feature, in this skill context it materially raises privacy and sensitive-data retention risk because previous conversation data remains indexed and easy to enumerate.

VirusTotal

VirusTotal findings are pending for this skill version.

Static analysis

No suspicious patterns detected.