Security audit

主动式代理(免费版)

Security checks across malware telemetry and agentic risk

Overview

This skill openly aims to make an agent proactive, but it asks the agent to persist user context broadly without clear consent, retention, or sensitive-data limits.

Review this skill before installing if you work with secrets, personal data, regulated data, or confidential business context. Use it only where persistent memory is acceptable, and require explicit user confirmation plus clear rules for what may be stored, how long it is retained, and how it can be inspected or deleted.

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 (6)

Vague Triggers

Medium

Confidence: 86% confidence
Finding: The skill description claims broad applicability to general development scenarios and emphasizes proactive behavior without defining tight activation boundaries or explicit opt-in conditions. In an agent skill with read/exec/write capabilities, this can cause the skill to activate in situations where users did not intend persistent memory, proactive actions, or autonomous behavior, increasing the chance of overreach and unsafe execution.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The quick-start instructions tell users to call the skill and provide inputs, but they do not specify precise conditions for when it should or should not be used. Because the skill promotes proactive behavior and state persistence, vague invocation guidance can lead to unintended use in sensitive contexts, including projects where persistence or autonomous follow-up is inappropriate.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The WAL section instructs the agent to write user preferences, decisions, deadlines, and corrections into memory before responding, but it does not warn users that this data may persist beyond the current interaction. This creates a consent and privacy risk because users may disclose sensitive information without realizing it will be stored and potentially reused across sessions.

Ssd 3

Medium

Confidence: 96% confidence
Finding: The skill explicitly requires persisting broad categories of user-provided information into memory before responding, including preferences, decisions, deadlines, and corrections. Without data minimization, sensitivity filtering, retention limits, or access controls, this creates a substantial risk of storing secrets, personal data, or business-sensitive context that may later leak through prompts, memory retrieval, logs, or other agent actions.

Ssd 3

Medium

Confidence: 93% confidence
Finding: The memory architecture promotes cross-session persistence of user context as a core feature, but it does not define limits for sensitive data classes, tenant boundaries, retention duration, or retrieval authorization. In a proactive agent context, persistent cross-session memory increases the blast radius of any prompt leakage, overbroad retrieval, or compromise of backing storage.

Ssd 3

Low

Confidence: 82% confidence
Finding: The example normalizes writing a user's project preference into both a working buffer and long-term memory, reinforcing broad retention as routine behavior. While the example uses relatively low-sensitivity data, it operationalizes a pattern that can easily be applied to more sensitive user inputs without safeguards.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.