Back to skill

Security audit

Skill 开发

Security checks across malware telemetry and agentic risk

Overview

The skill is not malware, but it can automatically make persistent changes to other skills and memory notes with broad activation triggers, so it needs review before installation.

Install only if you intentionally want an agent rule that can edit skills and update memory records. Before using it, narrow the trigger conditions, require explicit confirmation or a dry-run diff before edits, and clarify whether MEMORY.md writes are allowed. I found no executable payloads, credential handling, exfiltration, or destructive behavior in the artifacts, so this is a Review concern rather than a malicious finding.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The file defines a boundary of 'only modify skill files' while also requiring updates to MEMORY.md after every change, creating an internal policy contradiction. This ambiguity can cause the agent to write outside the declared scope, weakening user expectations and making unintended file modifications more likely.

Vague Triggers

Medium
Confidence
91% confidence
Finding
The trigger phrases are broad, generic, and likely to appear in normal discussion about skills, which can cause this skill to activate unintentionally. Because the skill is designed to modify other skills and update workspace memory, accidental activation could lead to unauthorized or unnecessary changes to prompts, rules, or related files without a clearly intentional user request.

Vague Triggers

Medium
Confidence
89% confidence
Finding
The skill metadata description includes broad activation language such as '发现技能有漏洞/流程缺失/逻辑错误...时触发', which can match ordinary troubleshooting or discussion contexts rather than an explicit request to invoke this skill. In an agent system, ambiguous triggers can cause unintended activation of a skill that edits other skills, increasing the risk of unauthorized or accidental modifications.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The trigger list contains very broad phrases like '优化技能', '修改技能', '技能有问题', and '技能漏洞' without any gating conditions, confirmation requirements, or scope checks. Because this skill is authorized to modify SKILL.md and related files and even encourages '主动修复,不等指令', accidental invocation could lead to unsolicited changes to system behavior or prompt assets.

Missing User Warnings

Medium
Confidence
97% confidence
Finding
The instruction to 'directly use SkillManage to edit' and 'don't ask whether to change' authorizes automatic file modification without clear upfront user warning or confirmation. In an agentic environment, this can lead to unauthorized or surprising persistent changes, especially when the trigger for activation is broad and includes vague phrases like 'improve skill' or 'skill has issues'.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.