使用Codex审查代理筛选面试代码提交
Screen interview code submissions with a Codex review agent
与Codex和ChatGPT完成。评论者使用针对前端代码审查的targeted agents.md与Codex，拒绝更改或添加文件过多的提交.
做了什么
A commenter said they set up ChatGPT Codex with a targeted agents.md file to handle first-pass reviews of coding submissions. The agent reviewed each submission against simple gate rules and automatically rejected ones with too many files changed or added; the human still chose the rules and used the agent as the front-line screener.
效果如何
The post gives no metrics, examples, or comparison to manual review. It only claims this was being used to fight low-quality, AI-assisted submissions earlier in the hiring funnel.
值得注意
The useful catch is how narrow the filter was: they tuned it around obvious review heuristics like file-count and added/changed-file limits, not a full autonomous hiring decision.
试试这个 Prompt
Write an agents.md for [repo or take-home project] that tells you how to screen candidate submissions: flag any submission that changes or adds more than [number] files, doesn't follow [requirement], or lacks tests. Then review the submission in [folder path] against it and give me a pass or reject recommendation with reasons. You're done when I have a short review I can use; don't contact the candidate or post anything anywhere.