Security audit
Security checks for vulnerabilities and agentic risk
This is a prompt-only frontend architecture review skill whose behavior is disclosed and aligned with analyzing project structure, dependencies, build setup, and maintainability.
Install this if you want Chinese-first frontend architecture review guidance. Be aware that it may activate on broad frontend terms and may ask you to share project structure or config snippets; only provide files you are comfortable having analyzed.
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.
| Tree-shaking | 2 分 | sideEffects 配置正确 |
| 路径别名 | 1 分 | @ 或 ~ 别名配置 |
| Source Map | 1 分 | 生产环境关闭或使用 hidden-source-map |
| 环境变量 | 1 分 | .env 文件 + cross-env / dotenv |
| 缓存策略 | 1 分 | contenthash 文件名 |
**Vite 项目检查清单：**
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.
| Tree-shaking | 2 分 | sideEffects 配置正确 |
| 路径别名 | 1 分 | @ 或 ~ 别名配置 |
| Source Map | 1 分 | 生产环境关闭或使用 hidden-source-map |
| 环境变量 | 1 分 | .env 文件 + cross-env / dotenv |
| 缓存策略 | 1 分 | contenthash 文件名 |
**Vite 项目检查清单：**
The trigger phrase list includes generic terms such as component design, build config, and monorepo that may appear in ordinary conversation or unrelated review requests, causing the skill to activate when the user did not explicitly ask for this architecture-analysis behavior. In a prompt-based skill system, over-broad activation can unexpectedly inject long instruction sets and alter assistant behavior, which is a real prompt-routing security and safety issue even without code execution.
The activation keywords are very broad and include common frontend terms like webpack, vite, monorepo, and component design. This can cause the skill to activate in conversations that only mention these topics casually, leading to unintended prompt takeover, irrelevant architectural judgments, or disclosure requests for project structure/configuration beyond what the user intended.
The intent-based activation rules are underspecified and rely on subjective interpretation of whether a user is asking if a choice is 'reasonable' or 'how to optimize'. In an agent system, ambiguous boundaries increase the chance of misrouting, causing the skill to engage when the user wanted a narrow answer and potentially steering the interaction into broad project analysis or unnecessary data collection.
The skill content is entirely written as a Chinese-only rule engine and its output templates, prompts, and follow-up guidance all assume Chinese responses without offering a language fallback or stating a justified locale restriction. In a general-purpose agent skill, this can cause user intent mismatch, reduce transparency, and create unsafe interaction failures when users cannot understand requests for sensitive project data or remediation advice.
The document presents the skill primarily in Chinese, with only a brief English subtitle, and does not clarify whether outputs or interaction language follow the user's preference. Because locale and language behavior are part of policy scope, the absence of an explicit user-choice statement creates a risk that the skill may default to a specific language without opt-in.
The document title, metadata, and most operating instructions are in Chinese, which can imply a default language behavior without explicit user opt-in. Although English trigger examples are included, the file does not clearly state that output language will follow user preference or that users may choose their locale.
No suspicious patterns detected.