Security audit
Security checks for vulnerabilities and agentic risk
This skill is a coherent code-review aid that reads user-specified documents and source code to produce an implementation verification report, with no evidence of hidden execution, exfiltration, or persistence beyond the report file.
Install only if you are comfortable letting the agent read the PRD/test files and source directories you provide and generate a report containing code paths and snippets. Before running it, specify narrow source directories and confirm the report output path, especially in repositories with sensitive code or documentation.
The manifest description and body instructions are entirely in Chinese and specify outputting a Markdown verification report, but nowhere offer the user a language choice or state that Chinese is optional. This creates a natural-language policy concern because the skill implicitly constrains the interaction/report locale without opt-in or documented justification.
The document contains an internal safety inconsistency: it states the skill is read-only, yet later mandates writing a report file. Such contradictions weaken operator trust and can bypass safeguards that rely on declared capability boundaries, especially in agent environments that distinguish read-only from write-enabled skills.
The skill explicitly promises '只读分析' (read-only analysis) but later instructs the agent to save a Markdown report to disk. That contradiction can mislead users and downstream policy checks, causing unintended filesystem writes in contexts where write access was not expected or approved.
The skill instructs saving a report file but does not clearly warn the user that it will perform filesystem writes. In agentic environments, undisclosed write behavior can create audit, privacy, and integrity risks, especially if reports are written into sensitive project directories or overwrite existing files.
L001-L003 以中文标题和“严格按此模板结构输出”的表述，结合整份模板均为中文，构成对输出语言的默认强制要求。文件中未提供语言切换、用户选择或明确的区域/合规性理由，因此符合语言/locale 策略违规特征。
