Security audit
Security checks for vulnerabilities and agentic risk
This skill is a straightforward architecture-review report generator with disclosed repository scanning and HTML report creation, and no evidence of hidden execution, exfiltration, persistence, or privilege escalation.
Install this if you want an architecture-review workflow that scans the current codebase and writes a visual HTML report. Review the generated report before sharing it, because it may contain project structure, dependency, and technical-debt details. The skill text is mostly Chinese and appears partially encoding-corrupted, so users who cannot read it may want a translated or cleaned version before relying on it.
The skill content is primarily in Chinese and appears to prescribe a specific language/locale behavior without checking the user's preference. While not a direct code-execution risk, forced locale can mislead users, reduce comprehension of side effects and recommendations, and make consent around scanning or file output less informed.
The skill description is broad enough that ordinary requests about architecture, technical debt, project health, or refactoring could auto-trigger it outside a clearly bounded review workflow. That increases the chance of unexpected deep repository scanning and report generation, which can consume resources and expose project structure when the user did not explicitly request this specific skill.
The skill instructs writing an HTML file into the workspace but does not warn the user beforehand or obtain consent for the filesystem side effect. Unexpected file creation can overwrite artifacts, leak analyzed content into shared workspaces, or violate user expectations about read-only analysis.
The example trigger phrase mirrors normal user speech ('help me analyze project architecture') and lacks constraints that would distinguish casual discussion from an instruction to activate the skill. In systems that route by semantic matching, this can cause unintentional activation and downstream file analysis or report-writing behavior without informed user consent.
No suspicious patterns detected.
