Security audit
Security checks for vulnerabilities and agentic risk
This Chinese article-review skill is not malicious, but it can automatically rewrite user documents and persist review preferences from broad review triggers.
Install only if you are comfortable with a review skill that may edit article files directly after broad requests like review or proofread. Use explicit instructions such as manual review, suggestions only, or do not write files when you want a read-only critique, and watch for project-local wenqu-skills preference files being created or updated.
The trigger phrases include very common requests such as “审查”, “检查一下”, “看看有没有问题”, and “review”, which are broad enough to match many ordinary interactions outside the intended scope. In an agent system, this can cause unintended activation, hijack user workflow, and lead the skill to process or rewrite content when the user did not explicitly request this specialized review behavior.
The description specifies reviewing Chinese content ("审查中文内容") as the operative scope, which is a language constraint expressed in natural language. The file does not explicitly present this as a user-selectable option or justify the locale restriction as a documented region-specific constraint.
The trigger phrases are broad and overlap with common requests such as 'check this' or 'look for issues,' which can cause unintended invocation in contexts the user did not mean to enter this skill. In a skill that may auto-edit content or write preference files, overbroad activation materially increases the risk of unexpected side effects.
The skill is framed as a review/checking tool, but its default mode says to directly modify the user's article and only report afterward. That creates a capability/expectation mismatch: a user or calling agent may invoke it for analysis but unintentionally grant write-like behavior, which can alter content without explicit per-run consent and increase the chance of unauthorized or harmful edits.
The skill instructs persisting user preferences into files under the project directory, even though persistent writes are not strictly necessary for one-off article review. This expands the skill's side effects beyond immediate review, creating risks of unauthorized file modification, accumulation of sensitive preference data, and cross-session influence on future outputs without clear opt-in.
The entire skill is written as Chinese-only operating guidance and does not indicate that users may choose another language or locale. Under the policy, forcing a specific language without opt-in can be a natural-language policy violation unless the locale limitation is explicitly justified.
The description states this file is a general translation reference for rendering English technical terms into Chinese and avoiding literal English-style phrasing. As written, it imposes a specific language/locale output policy rather than presenting Chinese as an option based on user preference or context.
The instruction says that when English words appear in the main text, the writer should consult exemption lists and this file before deciding translation treatment. This creates a default locale constraint toward Chinese terminology without indicating that the behavior depends on the user's requested language.
该技能具备 Write 和 Edit 权限，且主体内容明确要求“重写问题片段”“呈现人性化版本”，但元数据和说明没有突出提醒这是会直接修改用户文本的操作。若在自动化链路或内联阶段被调用，用户可能以为得到的是审阅意见，实际却发生内容改写，带来静默修改、意义漂移，甚至掩盖 AI 痕迹或弱化原始证据表达的风险。
该技能的触发条件写成“编辑或审阅文本，去除 AI 写作痕迹”，范围非常宽，和通用的审查、润色、proofread 类请求高度重叠。结果是它可能在并非要求“人性化改写”的场景被自动调用，把用户原本只想审查或纠错的内容直接改写，改变原意、语气或证据表述。
The file instructs reviewers to apply a Chinese-specific language policy, including required formatting like '中文名（英文）' and reporting English reuse as an issue. This imposes a fixed language/locale convention without any opt-in, exception scope, or user choice, which matches the language-policy violation criterion.
The document is entirely framed as a Chinese-only translation-style review standard and directs specific Chinese phrasing substitutions throughout. There is no indication that the user can opt into another language or that this locale restriction is justified as a region-specific requirement, which matches the language/locale policy violation criteria.
The phrase "默认跳过。用户说‘顺便看看结构’或‘完整审查’时执行" creates a broad activation condition around common user wording like "顺便看看结构," which can appear in ordinary review requests and unintentionally invoke this structural-review behavior. In a composable writing/review skill, ambiguous triggers can cause the wrong review phase to run, leading to scope creep, unexpected processing, or conflicting outputs across skills.
The trigger phrases are broad and overlap with common conversational requests such as 'overall check' or 'pre-submit check', which can cause the R7 review skill to activate when the user did not specifically intend a change-closure audit. In an agentic workflow, that increases the chance of misrouting tasks, applying the wrong review rubric, or producing misleading 'closure' confirmations based on incomplete context.
No suspicious patterns detected.
