Security audit

Remove-AI sence（去除AI味）

Security checks across malware telemetry and agentic risk

Overview

This skill performs disclosed local UI cleanup scans and edits, with no evidence of hidden data theft, persistence, or destructive behavior.

Install this only for projects where you are comfortable with a tool scanning frontend source files and creating local report/spec files. Review the generated report before approving replacements, and check version control afterward because the skill is designed to make project-wide UI edits.

SkillSpector

By NVIDIA

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (6)

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The trigger phrases “帮我去掉这个项目的AI味道” and especially “/de-ai” are broad enough to plausibly match ordinary user requests about UI cleanup or branding changes, causing the skill to activate in contexts where the user may not expect automated scanning and modification behavior. Because the skill is described as scanning all frontend files and performing one-click replacements, accidental invocation could lead to unintended project-wide changes.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The README states that the skill will scan all frontend files, generate an HTML report, modify files, and write a UI spec file, but it does not prominently warn that these actions create local artifacts and alter project contents. This is dangerous because users may invoke the skill expecting advice only, while the skill performs broad filesystem reads and writes that can expose code structure in reports or cause unwanted modifications.

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The trigger phrases are broad enough to activate on common design requests like UI beautification or color unification, which can cause the skill to run in situations where the user did not explicitly request scanning, file creation, or code modification. In this skill, accidental activation is more dangerous because later steps instruct filesystem writes and opening a browser automatically.

Vague Triggers

Medium

Confidence: 87% confidence
Finding: The usage section repeats ambiguous activation criteria tied to subjective judgments like 'looks AI-generated' and generic requests such as 'professionalize' or 'unify colors'. This increases the chance that the skill will take action without clear user consent, especially since its workflow includes scanning source files and preparing modifications.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill mandates generating an HTML report file in the project root and opening it in a browser, but does not require explicit user consent or warning about these side effects. Automatic file creation and browser launching can violate user expectations, expose local project details in rendered output, and create unwanted persistent artifacts.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The instruction to create `.claude/ui-spec.md` introduces an additional filesystem write without an explicit warning or consent step. This can silently alter the repository, pollute version control, or create policy/configuration files the user did not ask for.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.