---
title: "用 Claude Code 写代码经常被坑（偷偷执行 force push、未捕获异常）。我把真实踩过的坑整理成一套 AI 防御守则与拦截防线。做得还很粗浅，希望能帮大家"
url: "https://claudemods.dev/zh-cn/builds/claude-code-force-push-ai-f93745"
source: "https://github.com/machiko/agent-working-agreements"
source_type: "github"
repo: "https://github.com/machiko/agent-working-agreements"
author: "Reyes"
author_handle: "mr._.children"
stars: 0
category: "工作流与 CI"
categories: ["workflow"]
tags: ["agent-safety", "force-push", "git-safety", "guardrails", "hooks"]
permission_level: null
scan_complete: false
added: "2026-11-09T06:09:10.403Z"
updated: "2026-11-09T03:53:45Z"
locale: "zh-cn"
content_locale: "zh-cn"
---

# 用 Claude Code 写代码经常被坑（偷偷执行 force push、未捕获异常）。我把真实踩过的坑整理成一套 AI 防御守则与拦截防线。做得还很粗浅，希望能帮大家

作者分享一套 Claude Code 的 AI 防御守则与拦截防线，针对 force push、未捕获异常等实际踩坑，并附 GitHub 项目链接。

## Description

用 Claude Code 写代码经常被坑（偷偷执行 force push、未捕获异常）。我把真实踩过的坑整理成一套 AI 防御守则与拦截防线。做得还很粗浅，希望能帮大家少踩坑，非常欢迎 fork、给建议或提 PR 让项目更完整，帮助更多工程师 :D
https://github.com/machiko/agent-working-agreements

## Original README

用 Claude Code 寫扣常被雷（偷跑 force push、漏接例外）。我把真實踩過的坑整理成一套 AI 防禦守則與攔截防線。做得還很粗淺，希望能幫大家少踩雷，非常歡迎 fork、給建議或提 PR 讓專案更完整，幫助更多工程師 :D
https://github.com/machiko/agent-working-agreements