为什么没有项目上下文的 AI 代码审查会失败

AI code reviews typically start by scanning diffs and flagging potential issues, such as missing error handling or less descriptive variable names. However, these suggestions often lack practicality because the AI does not understand the specific context of the project. For instance, certain naming conventions may be enforced by the team, error handling might be managed at a global level, and performance may not be the primary concern in the current context. This lack of project understanding is not a model issue but a context issue. To address this, a tool called pi-reviewer has been developed, aimed at providing context-aware code reviews to enhance the effectiveness and relevance of the review process.