Tp4
- Category
- MCP Tool Poisoning
- Confidence
- 96% confidence
- Finding
声明描述的核心功能是内容质量鉴赏与风格/原创性检测,而代码仅做文章类型识别。代码没有任何质量评分维度、AI生成痕迹分析、原创性比对或抄袭检测逻辑,也没有与这些目标相关的数据处理。其主功能与声明存在实质性偏差,因此应判定为描述与实际行为不匹配。
- Content
Security audit
Security checks for vulnerabilities and agentic risk
This skill is a local Chinese article-quality analyzer with documentation and dependency hygiene issues, but no hidden exfiltration, persistence, or destructive behavior was found.
Install only if you are comfortable with a Chinese-language local article analyzer and with non-pinned Python dependencies. Treat its AI/originality results as heuristic scoring rather than proof, and avoid enabling or adding any external LLM processing for private drafts unless the skill clearly discloses what text is sent and asks for consent.
声明描述的核心功能是内容质量鉴赏与风格/原创性检测,而代码仅做文章类型识别。代码没有任何质量评分维度、AI生成痕迹分析、原创性比对或抄袭检测逻辑,也没有与这些目标相关的数据处理。其主功能与声明存在实质性偏差,因此应判定为描述与实际行为不匹配。
声明中的核心能力包括:多维度评估文章质量、检测AI味/大便味、识别原创内容。实际代码只覆盖了其中一部分,即对特定类型文本(情感散文、创意写作、简化小说)进行多维度质量分析。它通过统计情感词、比喻词、段落数、第一人称、对话比例等指标来生成评分和建议,但没有任何用于AI文本检测、风格异常检测、“大便味”检测、原创性验证、抄袭比对、外部检索或内容指纹识别的实现。代码主模块文档也明确写的是“情感散文与小说分析模块”,与声明中更广义且包含AI味/原创识别的能力不一致。因此应判定为描述与实际行为存在明显不匹配。
The declared description suggests a broad article-quality evaluator with AI-text detection and originality recognition. The code instead is narrowly focused on novel/fiction analysis, with functionality centered on literary dimensions such as plot structure, characters, narrative technique, style, and world-building. It also includes spoiler-tolerance handling, spoiler warnings, and spoiler-free summaries—capabilities not mentioned in the description. Conversely, the declared capabilities of detecting AI flavor/low-quality AI writing and identifying original content are not implemented in the supplied code. This is a material purpose mismatch, not just an implementation detail difference.
声明与代码部分重合:代码确实会评估文章质量,并通过原创性指示词、转载指示词和字符唯一率做粗略原创性分析。但声明中的关键能力“检测AI味/大便味”在代码中完全没有对应实现;代码仅基于技术术语密度、标题/列表/代码块数量、步骤词、句长、标点多样性等规则进行技术文章质量评分。此外,代码的分析对象明显限定为技术文章(technical_depth、code_blocks、技术术语密度等),因此其实际主用途比声明更窄且不同。综合判断,描述不能准确代表代码实际行为,存在实质性不匹配。
The skill advertises executable components and optional external LLM integrations but does not declare any tool scope or permission boundaries. In an agent environment, this can lead to overbroad file-read and network access being implicitly available, increasing the chance that article contents or local files are accessed or transmitted without clear user expectations.
The manifest-like metadata lists only '中文' under languages, and the entire skill description/output format is written as Chinese-only behavior without offering a language choice. This can violate language/locale policy when a skill forces a specific language absent user opt-in or a documented regional justification.
The skill states optional OpenAI/Anthropic use but does not warn users that submitted article text may be sent to third-party providers. This creates a privacy and confidentiality risk, especially if users analyze proprietary drafts, internal documents, or personal writing under the assumption processing is local.
This code embeds a fixed Chinese-language evaluative vocabulary, including insulting terminology, in docstrings and user-visible labels. Under the policy scope for all file types, this is a language/locale choice imposed without opt-in and is repeated throughout the skill's natural-language interface.
This code file contains natural-language descriptions and outputs that assume Chinese as the required language, including the module docstring and class documentation. Under the policy rule, forcing a specific language without user opt-in or a documented regional justification is a natural-language policy violation.
This code file contains user-facing natural-language strings that force a specific language/locale experience. The module docstring and command/help text indicate the skill is intended to operate in Chinese only, with no opt-in, fallback, or justification for the constraint.
This code declares the module purpose and user-facing behavior exclusively in Chinese, and the returned analysis labels/messages are also hard-coded in Chinese. Because the skill does not provide any opt-in, fallback, or documented locale constraint, it creates a language policy issue under the rule for forced language/locale behavior.
This code generates user-facing report content, labels, and explanations exclusively in Chinese, including headings, field names, and output text. Because the skill does not offer a language selection or document a justified locale restriction, it may violate language/locale policy expectations requiring user opt-in.
The module description and all user-facing advice templates are written exclusively in Chinese, indicating the skill is designed to produce Chinese-language recommendations by default. There is no visible user opt-in, language selection mechanism, or documented justification that this skill is intended only for a Chinese-language context.
This Python file contains user-facing natural-language descriptions entirely in Chinese, including the module title and evaluation criteria, with no indication that the language is configurable or limited to a China-specific use case. Under the policy, forcing a specific language without user opt-in is a locale-policy concern.
The dependency is specified with a lower bound only, which allows future installs to resolve to different versions over time. This weakens reproducibility and can unintentionally pull in vulnerable or breaking releases from the package index.
jieba>=0.42.1
scikit-learn>=1.0.0
numpy>=1.21.0
Using scikit-learn with only a minimum version makes the environment non-reproducible and prevents reviewers from determining which release will actually be installed. Because scikit-learn has known advisories in some versions, leaving it unpinned increases supply-chain and exposure uncertainty.
jieba>=0.42.1
scikit-learn>=1.0.0
numpy>=1.21.0
The manifest does not pin scikit-learn, and the package has known advisories across some versions, so the actual security posture cannot be verified from this file alone. While this does not prove an exploitable vulnerable version is installed, it is a real supply-chain risk because affected releases could be selected during installation.
The numpy requirement is not pinned to an exact version, so installations may resolve to different releases depending on time and environment. This creates supply-chain uncertainty and may introduce vulnerable or incompatible versions without code changes.
jieba>=0.42.1
scikit-learn>=1.0.0
numpy>=1.21.0
Because numpy is unpinned and has advisories affecting some releases, the installed version cannot be validated from the requirements file. This uncertainty is a genuine security weakness even if no specific vulnerable version is explicitly requested.
The module docstring names the skill in Chinese and describes its purpose only in Chinese, which suggests a fixed language/locale experience. There is no indication that users can opt into another language or that the Chinese-only presentation is required for a specific regional use case.
The skill returns and prints classifications such as '高度可信原创', '轻度疑似AI', and '极强AI味' entirely in Chinese. Because no alternative language or locale option is provided or documented, the skill imposes a specific language on all users.
The module description and user-facing documentation strings are written only in Chinese, which can constitute a language/locale policy issue when no user opt-in or alternative language support is offered. The file does not state that the skill is intentionally restricted to a Chinese-language context or provide any language selection mechanism.
This code performs a local file read based on the --file argument, but there is no visible warning or disclosure that the tool will access local filesystem content. Because the skill processes user documents, a brief disclosure helps set expectations around local data handling.