Security audit

无限记忆库(专业版)

Security checks across malware telemetry and agentic risk

Overview

The skill is a coherent local memory vault, but it asks the agent to automatically capture, copy, index, and retain user and built-in memory data with unclear consent and one inconsistent sync instruction.

Review this carefully before installing. It is not showing exfiltration or destructive behavior, but it is designed to make user conversations and built-in memory more durable and searchable on disk. Install only if you want automatic local memory capture, configure narrow sync rules, avoid storing sensitive personal or business data unless necessary, and confirm how to stop the daemon and delete or redact saved entries.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (8)

Description-Behavior Mismatch

Medium

Confidence: 89% confidence
Finding: The description advertises unrelated SEO and ranking-improvement use cases even though the rest of the skill implements a memory vault. This mismatch can cause the skill to activate in unrelated contexts and mislead users or routing logic about what data the skill will process, increasing the chance of unnecessary retention of sensitive conversation content.

Intent-Code Divergence

High

Confidence: 95% confidence
Finding: The file claims the system never modifies built-in memory, but the configuration example earlier includes instructions to '同步到内置记忆（单向)', which contradicts the safety model. Contradictory persistence rules are dangerous because an agent or implementer may enable bidirectional sync and overwrite, leak, or duplicate sensitive memory across stores.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The activation language is broad and includes generic marketing terms that are not tightly bound to a concrete user request. Overbroad triggering is risky for a memory skill because it may engage on unrelated conversations and start storing or indexing data users did not intend to persist.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The skill instructs automatic capture and storage of user information before responding, but it does not provide a prominent upfront privacy notice or require consent at the point of collection. In a memory tool, this is especially dangerous because conversations may contain personal, confidential, or regulated data that will be persisted and indexed without clear user awareness.

Missing User Warnings

High

Confidence: 96% confidence
Finding: The skill describes automatic synchronization from built-in memory into local structured storage without a clear warning that sensitive data may be copied, duplicated, and become more discoverable through indexes and semantic search. Copying memory into another store expands the attack surface and retention footprint, especially for preferences, decisions, or personal details.

Ssd 3

Medium

Confidence: 94% confidence
Finding: These instructions direct the agent to automatically write project, people, and decision data into persistent storage and maintain indexes, creating a built-in data retention path from ordinary conversation to disk. The context makes this more dangerous because the skill is explicitly designed for long-term structured storage, so any accidental capture becomes durable, searchable, and easier to exfiltrate later.

Ssd 3

Medium

Confidence: 97% confidence
Finding: The skill explicitly says to auto-capture important information, store it, update indexes, and only then reply. That ordering removes the opportunity for informed consent and can silently persist sensitive content before the user understands that retention is happening.

Ssd 3

Medium

Confidence: 93% confidence
Finding: The FAQ normalizes automatic identification and storage of conversational information such as projects, contacts, decisions, and preferences. This reinforces a design that encourages broad persistent collection of potentially sensitive user data, which is risky in a skill whose main function is long-term retention and retrieval.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.