Security audit

微信AI接入顾问

Security checks across malware telemetry and agentic risk

Overview

This is a documentation and template skill for planning WeChat mini-program AI integration, with no hidden installer or automatic execution behavior found.

Install only if you want guidance for WeChat mini-program AI integration. Treat generated code, mcp.json, AGENTS.md, and SKILL.md as drafts: review permissions, backend URLs, token handling, phone/address collection, payment/order confirmation, and privacy compliance before using them in any real app.

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 (8)

Vague Triggers

Medium

Confidence: 85% confidence
Finding: The README advertises very broad capabilities such as '直接对话即可' and '帮我生成完整项目代码' without clearly constraining scope, trust boundaries, or allowed actions. This can cause the skill to activate on generic requests and encourage users to rely on generated code/configuration artifacts that may be unsuitable, insecure, or overly permissive for their environment.

Missing User Warnings

Low

Confidence: 93% confidence
Finding: The skill explicitly offers to generate '完整项目代码' and configuration artifacts like app.json, AGENTS.md, SKILL.md, and mcp.json, but the README does not warn users that these outputs can affect application behavior, permissions, integrations, or data handling. In a security-sensitive development workflow, omission of these warnings increases the chance that users will apply generated artifacts without adequate review.

Missing User Warnings

Medium

Confidence: 84% confidence
Finding: The document instructs developers how to accept image/file inputs and pass them into AI-facing interfaces, but it omits privacy, consent, and secure-handling guidance. In this skill context, developers may process user-uploaded files containing personal or sensitive information and transmit them onward without clear notice, minimization, or validation, increasing the risk of privacy violations and unsafe data flows.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: This section enumerates sensitive capabilities such as login credentials, phone number access, and network requests, but does not pair them with least-privilege, consent, or handling requirements. In a WeChat AI integration skill, that omission is materially risky because downstream implementers may over-collect identifiers or transmit them broadly in middleware/API flows, creating privacy, compliance, and account-linkage exposure.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The trigger guidance uses broad natural-language phrases such as '帮我生成代码' and '给我完整项目', which are common requests that may appear outside the narrow context intended for this template. This can cause the agent to load code-generation instructions too eagerly, increasing the chance of inappropriate tool/template invocation and unsafe code scaffolding in unrelated conversations.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The file states when it should be read, but does not define boundaries, disambiguation rules, or counterexamples for when it should not be used. In an agent skill, ambiguous activation conditions can lead to misrouting, unnecessary exposure to generation templates, and incorrect code outputs that may not fit the user's actual request or safety constraints.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The trigger condition says to read this file when users ask about broad topics like GEO, structured data, testing, launch checks, or improving AI hit rate. These phrases are wide and overlapping, which can cause the skill to activate in situations where the user did not actually need this checklist, increasing unintended prompt inclusion and context pollution. In a guidance skill like this, the impact is limited because the content is advisory rather than directly executing actions, but it can still reduce reliability and create opportunities for prompt steering through over-broad routing.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The activation guidance includes very broad phrases such as “给我示范” and generic requests for cases/templates, which can cause this reference file to be pulled into many ordinary conversations beyond its intended scope. Over-broad triggers increase the chance of context poisoning or irrelevant instruction injection, especially when the file contains detailed behavioral templates that may steer the agent unnecessarily.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.