Back to skill

Security audit

Wechat Assistant

Security checks for vulnerabilities and agentic risk

Overview

This skill is a local WeChat chat assistant with no exfiltration signs, but it bulk-captures and persists private messages without adequate privacy controls or clear retention boundaries.

Install only if you are comfortable with a local tool reading visible WeChat chats and saving private message content and derived analysis in plaintext files. Use a dedicated folder, avoid broad chat exports, review files before sharing them, and delete or segregate knowledge-base and reply-history files that contain sensitive conversations.

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
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (27)

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The code is narrowly focused on offline analysis of provided chat-message JSON input. Its implemented functions align with the 'analyze chat data' portion of the description: keyword extraction, intent/sentiment analysis, requirement mining, frequency and interaction analysis, and insight generation. However, several prominent declared capabilities are absent: there is no integration with WeChat, no automatic data capture, no reply generation, no approval workflow, no message sending, and no task/reminder automation. Because these missing capabilities are central parts of the declared purpose rather than minor extras, the description materially overstates what the code actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

该代码块的实际功能范围明显比声明狭窄,核心行为只是自动抓取微信聊天记录并导出结果。声明虽然提到了“自动抓取聊天记录”,这一点与代码一致,但其主要描述强调的是“智能分析与自动回复助手”,包括分析、生成回复、发送审核、人脉管理、任务提醒等多项上层能力,而这些在代码中均未出现。代码也没有网络调用、模型推理、消息发送、调度提醒或联系人关系管理逻辑。因此,描述未准确代表此代码块的实际行为,属于描述与行为不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The declared description presents a broad WeChat-focused assistant with automated chat log collection, analysis, reply generation, review/sending, relationship management, and reminders. The supplied code only implements a standalone local knowledge-base utility: it reads/writes a JSON file, maintains samples/intents/templates, performs TF-IDF similarity search, and returns reply suggestions. While reply suggestion/learning is loosely related to part of the description, the major WeChat-specific and operational capabilities are absent. Therefore the code materially under-delivers and has a narrower, different primary purpose than declared.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The declared description presents a broad WeChat assistant covering chat capture, intelligent analysis, need mining, reply generation, review, sending, relationship management, and task reminders. The supplied code does not implement WeChat access, chat scraping, reply drafting, approval flow, or message sending. Its sole purpose is to load JSON data and render visual outputs with matplotlib. Some chart types loosely align with parts of the description—such as relationship maps and task/requirement timelines—but only as presentation of already available data. Because the primary behavior of this code chunk is visualization rather than the declared end-to-end chat analysis and auto-reply assistant functionality, this is a material description-behavior mismatch.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

This script automates extraction of WeChat private chat content and writes it to a JSON file without any consent check, privacy warning, access control, or data minimization. In the context of a chat-assistant skill, this is especially dangerous because it enables bulk collection and persistence of sensitive personal communications that could be misused, leaked, or processed beyond the user’s expectations.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
79% confidence
Finding

The skill requests or implies file read/write behavior through its documented workflows and scripts, but it does not declare any tool scope or permission boundaries. In a skill that processes chat logs and knowledge-base files, missing scope declarations increase the risk of overbroad filesystem access and make it harder to enforce least privilege around sensitive personal data.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The description says to use the skill whenever the user needs to analyze WeChat data, generate replies, manage relationships, or plan reminders, which spans very broad everyday activities without clear boundaries. This lack of specificity increases the chance of unintended invocation because it does not define narrow trigger phrases, scope constraints, or exclusion conditions.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger condition includes broad intents like analyzing chats, managing contacts, auto-replying, or planning reminders, all of which can overlap with ordinary conversation and adjacent assistant tasks. The file does not provide an explicit trigger list or exclusion examples, making invocation criteria ambiguous.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

This JSON knowledge file contains very generic everyday utterances such as "在吗?" and "能帮忙看一下吗" as trigger-like inputs. Because manifest/plain-text trigger guidance applies to JSON files, these examples define an activation scope that overlaps heavily with common speech and provide no constraints or exclusion conditions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The specification defines structures and workflows for collecting, analyzing, storing, reviewing, and sending private WeChat chat content, but it contains no privacy notice, consent requirement, retention guidance, or safety controls. In the context of a messaging assistant handling personal communications, this omission can lead to unauthorized processing or accidental disclosure of sensitive personal data through analysis outputs and auto-reply workflows.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script loads raw WeChat chat records from disk and processes highly sensitive personal communications without any consent prompt, privacy notice, minimization, or validation of how that data should be handled. In this skill context, the input is explicitly private conversation history, so silent ingestion materially increases privacy and compliance risk if users or downstream automation do not realize sensitive content is being analyzed.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The analyzer can persist derived results to an arbitrary output file, and those results include extracted keywords, requirements, contacts, sentiment, and insights derived from private chats. Even if the raw chat log is not copied verbatim, the output can still disclose sensitive relationship, scheduling, and behavioral information, making undisclosed persistence a real privacy exposure.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

This Python file contains the primary description and CLI help text entirely in Chinese, beginning with the module docstring. The policy requires avoiding forced language or locale constraints unless the skill offers opt-in choice or clearly documents a justified regional limitation, which is not present here.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The send path is presented as a simulation, but it still performs a state-changing operation by calling learn_from_conversation(), which can mutate the knowledge base based on message content. This mismatch can mislead operators into believing the action is side-effect free, enabling unintended data poisoning, persistence of sensitive content, or corruption of future reply behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

User-facing natural-language strings, including the module description, errors, and CLI help, are all fixed in Chinese. This forces a specific language/locale without offering the user a choice or documenting a justified locale restriction.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Reply generation pulls outputs from similar historical samples and returns them as suggestions, so previously stored private messages can be surfaced to a different later interaction. In the stated skill context of analyzing and auto-replying to WeChat chats, this semantic data leakage is especially dangerous because the system is explicitly designed to ingest and reuse personal communications.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The learning feature persists raw user inputs and agent replies directly to a local JSON knowledge base, with no consent flow, minimization, redaction, retention control, or encryption. In a WeChat assistant context, chats commonly contain personal, financial, relationship, and business-sensitive content, so silent persistence materially increases privacy and confidentiality risk.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script learns from every conversation by storing plaintext inputs and replies, then rebuilding an index over that content for later retrieval. This creates a durable sensitive-data store from chat content and makes accidental reuse of private information likely, especially in a messaging assistant handling potentially intimate or regulated data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

In dashboard mode, the script writes multiple derived PNG files using suffixed filenames in addition to the provided base output path. This is a safety-relevant side effect beyond the single --output file the user might expect, and there is no warning in code output or argument help describing these extra files.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

All sample inputs and outputs are written in Chinese, with no indication that the skill supports language choice or that the locale restriction is intentional and documented. This can violate language/locale policy if the skill behavior is implicitly constrained to one language without user opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

The skill specification is entirely in Chinese and does not state that the skill is region-specific or that users can choose another language. This can violate language/locale policy when a skill implicitly forces a specific language without opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

Natural-language strings in the module description and CLI/help text are Chinese-only, and the skill does not offer any language or locale selection. Under the stated policy, forcing a specific language without user opt-in is a language/locale policy issue unless clearly documented as region-specific or optional.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

清单描述将该技能定位为可“自动生成回复内容并支持审核发送”,通常意味着在审核后具备实际发送能力。这里 execute_send 只构造成功结果并返回“模拟发送成功”,未与微信或发送通道交互,和描述中的“发送”能力存在语义落差。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The file-level documentation and all user-facing CLI/error messages are written only in Chinese, with no indication that users may select another language or locale. Under the stated policy, forcing a specific language without user opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The file contains user-facing natural-language text in Chinese, including the module description and error/help messaging, without any indication that users can choose a different language. This can violate language/locale policy when a skill imposes a specific language without opt-in or documented justification.

Content

No source excerpt is available for this finding.