Security audit

EO Ability: Memory

Security checks for vulnerabilities and agentic risk

Overview

This is an instruction-only memory skill with no installer code, but it asks the agent to automatically save and reload cross-session memories without clear consent, scope, or deletion details.

Install only if you intentionally want cross-session memory. Before using it, confirm where memories are stored, whether automatic saving/loading can be disabled, how to inspect and delete memories, and avoid storing secrets, credentials, health, financial, legal, personal, or confidential business information unless you have explicit controls for that data.

SkillSpector

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 (6)

Vague Triggers

Medium

Confidence: 87% confidence
Finding: The skill advertises automatic memory behavior in broad terms without clear gating conditions, which can cause the agent to save or restore information on common conversational cues rather than explicit user consent. In a cross-session memory skill, overbroad activation materially increases the chance of unintended persistence or retrieval of sensitive context.

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The example trigger phrase "继续上次的研究" is generic and likely to occur in normal conversation, yet it is shown as causing automatic recovery of prior context. This can lead to silent recall of historical data that the user did not intend to resurface in the current session or environment.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The description prominently promises cross-session persistence of user preferences and project context but does not warn that potentially sensitive data may be retained beyond the current conversation. Users may disclose academic, personal, or business information without understanding retention or retrieval implications.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The skill says key decisions are automatically saved, but gives no user-facing warning about retention, review, or deletion. Automatic saving without transparency is especially risky for a memory feature because users may unintentionally cause confidential preferences, project details, or sensitive conversation content to be stored long term.

Ssd 3

Medium

Confidence: 95% confidence
Finding: The skill is explicitly designed to persist user preferences, mentor requirements, and project context across sessions, yet it provides no visible data classification rules, storage boundaries, or exclusions for sensitive information. In this context, the absence of safeguards makes inadvertent collection and later disclosure of private or confidential data substantially more likely.

Ssd 3

Medium

Confidence: 94% confidence
Finding: The examples encourage storing and searching broad prior discussion details such as negotiation topics, research progress, and adviser preferences, then automatically restoring them later. This creates a realistic risk of exposing historical user data in the wrong context, especially if search/load is broad or triggered without precise scoping.

Static analysis

No suspicious patterns detected.