客厅装修避坑指南

• Outcome: Review
• Latest audit: Aug 28, 2026 · 10:03 AM UTC
• Version: 1.0.0

Overview

This is a documentation-only renovation skill, but its narrow living-room label is materially broader in practice and includes legal, real-estate, contractor, and promotional source material.

Review this skill carefully before installing if you expect a tightly scoped living-room assistant. It should be treated as broad general renovation reference material, and users should independently verify contractor, contract, legal, and property-purchase advice rather than relying on it as authoritative guidance.

SkillSpector

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Memory Poisoning: Persistent Context Injection, Context Window Stuffing, Memory Manipulation
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration

Findings (9)

Description-Behavior Mismatch

Medium

• Confidence: 95% confidence
• Finding: The reference file massively exceeds the declared skill purpose of a living-room decoration guide and includes broad advice on pets, contracts, home buying, water/electric work, and contractor selection. Over-broad instructions increase the chance the agent will answer outside its intended safety and product boundaries, creating prompt-scope drift and making it easier for unrelated or risky guidance to be surfaced under a misleading skill identity.

Context-Inappropriate Capability

Low

• Confidence: 88% confidence
• Finding: Including home-purchase and contractor-selection advice in a living-room-focused skill creates unnecessary capability expansion beyond user expectations. While not directly an exploit primitive, it can mislead users, bypass product scoping, and cause the agent to provide consequential advice in areas that may require stronger review or disclaimers.

Context-Inappropriate Capability

Low

• Confidence: 90% confidence
• Finding: Contract, pricing, and renovation-company screening guidance is outside the declared living-room guide scope and can cause the agent to drift into legal/financial or vendor-selection advice. This broadens the skill’s effective authority and may expose users to unreviewed high-stakes recommendations not justified by the manifest.

Intent-Code Divergence

Medium

• Confidence: 97% confidence
• Finding: The file contains embedded instructions forbidding platform/contacting language while simultaneously including repeated WeChat/contact promotions and platform references. This contradiction is risky because hidden promotional content can steer outputs toward off-platform contact, advertising, or lead-generation behavior, undermining trust and policy controls.

Description-Behavior Mismatch

Medium

• Confidence: 95% confidence
• Finding: The skill is advertised as a 客厅装修避坑指南, but the referenced content includes broad unrelated domains such as legal disputes, whole-home customization, kitchen/bathroom/bedroom renovation, and old-house remodeling. This overbroad scope increases the chance the agent will answer outside its intended domain, making unsafe or low-quality guidance more likely and weakening policy/control boundaries around the skill.

Context-Inappropriate Capability

Medium

• Confidence: 96% confidence
• Finding: Including legal-rights and litigation guidance inside a living-room renovation skill creates an unjustified authority expansion into legal advice. If surfaced to users, the agent may provide jurisdiction-sensitive or incomplete legal guidance under the guise of a decoration skill, which can mislead users into taking harmful real-world actions.

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: The skill contains property-purchase and location-investment guidance such as avoiding cemeteries, hospitals, high-voltage lines, and choosing school/metro/commercial areas, which is unrelated to living-room renovation. This broadens the skill into quasi-financial and real-estate decision support without matching the declared purpose, increasing the risk of misleading recommendations and user overreliance.

Context Window Stuffing

Medium

• Category: Memory Poisoning

Content

## 第八、磁吸广告框做装饰画。


  A3放正本 A4放副本\n营业执照\n营业执照证书\nTHORIZATION\n专利证书\n新城\n城\n授权\n工长.com(贵阳)\n伙伴\n李wang\n新汇\n105日\n107805858058405405500540550054050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050050\n\n


记得买哑光面的效果更好，大的尺寸是a3，小的是a4，加起来才六块钱，还有好多颜色的边框可以选。搜营业执照磁吸展示相框，找一家便宜的就行。

• Confidence: 91% confidence
• Finding: The corpus contains obvious context-window stuffing/noise, including a long meaningless numeric sequence embedded in otherwise normal content. Large irrelevant payloads can dilute higher-priority instructions, reduce model attention to safety-critical guidance, and make prompt-injection or behavior drift more effective when combined with other embedded instruction blocks.

Context Window Stuffing

Medium

• Category: Memory Poisoning

Content

▲原始户型图

客房 340 $$\frac{1}{100}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$$$\frac{1}{120}$$ (尺寸标注，原文为公式形式，此处保留)

▲设计布局图

• Confidence: 93% confidence
• Finding: The file contains large volumes of noisy, repetitive, and low-value text such as long formula-like dimension strings, which can waste context budget and dilute relevant safeguards or instructions. In an agent setting, context stuffing can reduce retrieval quality, increase instruction collisions, and make it easier for important policy constraints to be ignored or crowded out.

Static analysis

No suspicious patterns detected.