Back to skill

Security audit

橱柜定制避坑指南

Security checks for vulnerabilities and agentic risk

Overview

This skill does not run code, but its advice sources contain under-disclosed ads, personal contact prompts, unrelated renovation guidance, and some safety-sensitive DIY guidance.

Review before installing. This skill appears unlikely to affect your computer or private data, but do not treat its renovation advice as neutral or fully curated; it may surface ads, personal contact details, unrelated home-buying or renovation advice, and unsafe DIY suggestions unless your agent filters them carefully.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

other

Warning
Location
references/ref-259.md:20
Finding
Undisclosed Commercial Promotion and Personal Contact Redirection in Trusted Reference Content## Vulnerability Details **File Location**: `references/ref-259.md:20-22` **Related Locations**: `SKILL.md:22`, `references/ref-385.md:12-15`, `references/ref-386.md:15-16`, and `references/ref-387.md:30-31` **Vulnerability Type**: Undisclosed promotional redirection in Agent-consumed reference material **Risk Level**: Medium ### Vulnerable Content The following is an English translation of the complete promotional segment at `references/ref-259.md:20-22`: ```text Click the image above to buy it. Now there is an unprecedented manufacturer promotion. All products are offered to homeowners at a 30% discount. This handmade sink is available in stainless steel and nano-black, with various sizes and specifications. No matter where you are, it will be shipped free even at a loss! Click the image above to enter the Classroom Marketplace and make a purchase. The promotion may end at any time. ``` The package also contains personal contact redirection at `references/ref-385.md:12-15`, translated as: ```text We have worked with integrity in renovation for ten years and continue to make an effort. Public platform account: x Personal WeChat and QQ: 493078178 Follow quickly: ``` This contradicts the representation in `SKILL.md:22` that referral advertisements were removed from the corpus. ### Technical Analysis The reference files are trusted knowledge sources that the Skill directs the Agent to consult when answering renovation and cabinetry questions. They are therefore capable of influencing generated answers even though they contain no executable code. The affected references contain direct purchasing prompts, urgency-based sales language, marketplace redirection, subscription requests, and a personal WeChat/QQ identifier. Because the primary Skill documentation states that promotional material was removed, neither the Agent nor the user is adequately warned that retrieved content may serve a commercial purpose. T ...[truncated 2140 chars]
Remediation
## Remediation Suggestions 1. Remove all purchasing prompts, discounts, marketplace directions, urgency language, subscription requests, QR-code directions, and personal contact identifiers from every reference file. 2. Sanitize at least the confirmed related locations: - `references/ref-259.md:20-22` - `references/ref-385.md:12-15` - `references/ref-386.md:15-16` - `references/ref-387.md:30-31` 3. Search the complete corpus for promotional indicators such as “buy,” “discount,” “marketplace,” “promotion,” “follow,” “subscribe,” “backend reply,” “WeChat,” “QQ,” “contact,” and QR-code instructions. 4. Add an explicit rule to `SKILL.md` prohibiting the Agent from reproducing source advertisements, personal contact details, marketplace directions, and subscription prompts. 5. Correct or temporarily remove the claim in `SKILL.md:22` that referral advertising has been removed until a complete sanitization pass has been performed. 6. Separate factual product-selection guidance from source boilerplate during content ingestion. Retain only technically relevant renovation information. 7. Add automated content checks to the release process so future reference updates fail validation when promotional language, personal identifiers, or commercial redirection is detected. 8. Where a commercial relationship is intentionally retained, disclose it prominently and require the Agent to distinguish sponsored material from neutral guidance.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (26)

Description-Behavior Mismatch

Medium
Confidence
94% confidence
Finding
The file content does not match the declared purpose of a cabinet-customization pitfall-avoidance skill. This can mislead users and downstream agents into relying on irrelevant guidance, reducing trustworthiness and potentially causing incorrect advice in a domain where users expect targeted renovation recommendations.

Description-Behavior Mismatch

High
Confidence
95% confidence
Finding
The file content is materially unrelated to the skill’s declared purpose of cabinet customization guidance, instead providing broad balcony renovation advice. This mismatch can mislead downstream agents or users, weaken trust boundaries, and cause the skill to surface irrelevant or unintended guidance in contexts where users expect specialized cabinet-related recommendations.

Description-Behavior Mismatch

High
Confidence
97% confidence
Finding
The reference content is materially misaligned with the skill’s declared purpose: the skill claims to provide cabinet customization guidance, but this file contains bathroom renovation advice. This can cause the agent to retrieve and present irrelevant guidance, undermining user trust and potentially leading to poor renovation decisions if users rely on the skill for domain-specific recommendations.

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The referenced content is materially unrelated to the skill’s declared purpose of cabinet customization pitfall guidance, instead discussing general appliance purchasing tips. This creates a scope-integrity issue: the agent may retrieve and present irrelevant advice, degrading reliability and potentially misleading users who expect specialized cabinet guidance.

Description-Behavior Mismatch

Medium
Confidence
90% confidence
Finding
The file content is materially unrelated to the declared skill purpose of cabinet-customization pitfall guidance and instead contains balcony cat-climbing/storage design inspiration. This mismatch can mislead downstream users or agents, degrade trust in the skill, and cause incorrect retrieval or recommendations, though it does not by itself introduce direct code-execution or data-exfiltration risk.

Description-Behavior Mismatch

High
Confidence
95% confidence
Finding
The reference file is materially misaligned with the skill’s stated purpose: it contains balcony renovation content instead of cabinet customization guidance. In a user-facing advisory skill, this can cause the agent to retrieve and present irrelevant or misleading renovation advice, reducing reliability and potentially leading users to make inappropriate design or construction decisions.

Description-Behavior Mismatch

High
Confidence
97% confidence
Finding
The referenced file is materially unrelated to the skill’s declared purpose of cabinet-customization pitfall guidance and instead contains bedroom renovation and micro-remodeling content. This creates a skill-data integrity issue: users may receive irrelevant or misleading advice, and downstream agents that trust the skill metadata could surface incorrect recommendations, reducing reliability and potentially causing poor renovation decisions.

Description-Behavior Mismatch

Medium
Confidence
93% confidence
Finding
The reference content is clearly about dishwasher selection, while the skill is declared as a cabinet customization pitfall guide. This scope mismatch can cause the agent to retrieve and present irrelevant guidance, reducing reliability and potentially misleading users making renovation decisions based on the skill’s stated purpose.

Description-Behavior Mismatch

Medium
Confidence
90% confidence
Finding
The file content is materially misaligned with the stated skill purpose: instead of focused guidance on custom-cabinet pitfalls, it provides a general home-renovation case study plus product-promotion content. This can degrade the skill’s reliability, cause irrelevant or misleading responses, and introduce undisclosed commercial bias when the agent retrieves and summarizes this reference.

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The referenced file content is about window storage and design ideas, while the skill is advertised as a cabinet customization pitfall guide. This mismatch can cause the agent to retrieve and present irrelevant guidance, reducing reliability and potentially misleading users making purchase or renovation decisions.

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The file content does not match the skill's stated purpose of providing cabinet-customization pitfall guidance; instead it is a general home-decoration showcase with shopping-style recommendations. This can mislead downstream users or agents into relying on irrelevant material, degrading decision quality and potentially causing unsafe or incorrect renovation advice in a domain where accurate dimensions, materials, and installation guidance matter.

Description-Behavior Mismatch

Medium
Confidence
92% confidence
Finding
The reference content does not match the skill’s declared purpose of providing cabinet customization pitfall-avoidance guidance; it is mostly broad interior-decoration commentary plus product promotion. This mismatch can cause the agent to return irrelevant or commercially biased advice, undermining user trust and potentially leading users to make decisions without the promised domain-specific safety guidance.

Description-Behavior Mismatch

Low
Confidence
97% confidence
Finding
The file content is largely unrelated to the stated skill purpose of cabinet-customization pitfall guidance. It consists mainly of a home-tour narrative plus product-promotion sections ('点击立即购买'), which can degrade skill integrity, mislead users about the skill’s expertise boundaries, and introduce undisclosed commercial influence. In this skill context, the mismatch increases risk because users expect focused renovation guidance and may instead receive irrelevant or biased recommendations.

Description-Behavior Mismatch

Medium
Confidence
94% confidence
Finding
The file content is largely a general home-decoration showcase and shopping-oriented article, not a focused 橱柜定制避坑指南 knowledge resource as declared in the skill metadata. This mismatch can cause the agent to surface irrelevant or commercially biased guidance, weakening trust boundaries and increasing the chance of misleading users who expect specialized cabinet customization advice.

Description-Behavior Mismatch

Medium
Confidence
96% confidence
Finding
The embedded purchase prompts and mini-program shopping calls-to-action introduce undisclosed commercial steering into what is presented as an advisory knowledge-base skill. This is dangerous because the agent may promote products or purchases instead of giving neutral recommendations, creating manipulation, conflict-of-interest, and possible policy-compliance issues.

Description-Behavior Mismatch

Medium
Confidence
90% confidence
Finding
The skill is scoped as a cabinet/custom-furniture avoidance guide, but this section expands into broad home-buying and investment-style advice. That scope drift is risky because users may rely on unsupported real-estate guidance presented under an unrelated skill, increasing the chance of harmful financial decisions and eroding trust boundaries for what the skill is supposed to cover.

Description-Behavior Mismatch

Medium
Confidence
87% confidence
Finding
Large portions of the file contain general renovation, installation, lighting, bathroom, flooring, and even broader home-selection advice that are not clearly tied to the declared cabinet-customization purpose. This overbroad content increases the chance that the agent will answer outside its competence and provide authoritative-seeming recommendations on unrelated topics, including safety-relevant installation decisions.

Description-Behavior Mismatch

Medium
Confidence
94% confidence
Finding
The file content is materially unrelated to the declared skill purpose of cabinet customization pitfalls and instead provides children's room layout advice. This creates a scope-integrity problem: the agent may retrieve and present irrelevant guidance, reducing trustworthiness and potentially causing incorrect recommendations in a domain where users expect specialized renovation advice.

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The reference content is materially out of scope for a cabinet-customization skill and instead provides children's room environmental/decoration guidance. This can cause the agent to answer unrelated questions with irrelevant safety-sensitive advice, increasing the risk of misleading recommendations and weakening the integrity of the skill's intended behavior.

Missing User Warnings

High
Confidence
94% confidence
Finding
The content states that if a load-bearing wall cannot be pre-embedded, 'forcing' the embedding is still possible but will damage the load-bearing structure, framing it as an available option rather than clearly prohibiting it. In a home renovation guidance skill, this can encourage unsafe structural modification by non-experts, creating real risks of structural weakening, code violations, and injury.

Vague Triggers

Medium
Confidence
87% confidence
Finding
The file instructs users to reply with very broad single-word triggers like “衣柜” and “定制” to retrieve content. Such generic triggers can easily collide with unrelated user messages, causing the skill to activate or route content outside the user’s precise intent, which weakens scope control and can lead to prompt/flow hijacking in multi-skill or chat-driven environments.

Vague Triggers

Medium
Confidence
89% confidence
Finding
The keyword trigger list is broad and loosely related to the declared skill scope, which can cause the skill to activate or retrieve this document for many unrelated queries. In context, that makes the irrelevant and shopping-biased content more likely to influence responses outside its intended domain, degrading answer quality and increasing the chance of unsolicited recommendations.

Vague Triggers

Medium
Confidence
92% confidence
Finding
The file ends with a broad list of clickable/reply keywords such as '空调', '合同', and '预算项目' without any constraints on when they should trigger. In an agent skill, unconstrained triggers can cause unintended activations and route users into unrelated content paths, increasing the chance of prompt hijacking, scope creep, or confusing behavior even though the content itself appears non-malicious.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The content describes DIY lamp assembly and modification of hanging light fixtures, including wiring-related setup such as installing a ceiling plate, adjusting cord length, and creating a multi-head pendant arrangement, but provides no warning to disconnect power, verify load support, use compliant components, or involve a licensed electrician. In a home-improvement skill, readers may treat this as actionable instruction and attempt unsafe electrical work, creating shock, fire, or falling-fixture hazards.

Missing User Warnings

Low
Confidence
83% confidence
Finding
The text advises making openings for an air conditioner and range hood before finishing walls and floors, but omits basic safety cautions such as checking for hidden electrical wiring, plumbing, structural elements, and proper dust/PPE controls. In a renovation guidance context, readers may interpret this as permission to drill or core through walls themselves, risking injury or property damage.

Static analysis

No suspicious patterns detected.