AI资讯大全AIPPXP.CN搜索 ↗
图文教程 · 手把手

Text tutorial: 2 minutes to install "Don't make history again" to cure the bad problem of AI code change图文教程:两分钟装上「别再造史了」,治好 AI 乱改代码的坏毛病

Then do it again: confirm which of the several → unzipped stop-that-shit plug-ins have been shot by your Agent → Use an "export file" small experiment to verify that the hooks are effective → Learn to release legal checksums and other boundary adjustments. Requires Node.js 18 +, supports Claude Code, Codex, OpenCode and other mainstream hosts.

跟着做一遍:确认你的 Agent 中枪了哪几条 → 解压安装 stop-that-shit 插件 → 用一个「导出文件」小实验验证钩子生效 → 学会放行合法校验和等边界调整。需要 Node.js 18+,支持 Claude Code、Codex、OpenCode 等主流宿主。

2026-11-01 更新 · 免费
图文教程:两分钟装上「别再造史了」,治好 AI 乱改代码的坏毛病
1

Step 1: Take a seat on the alignment sign to make sure you need it too第一步:对号入座,确认你也需要它

Let the AI export a file, and it easily generates a .sha256 checksum file - no one asked for it, no one read it. This is H (useless defense).

Let it only do code Review, but there are file modifications in diff - this is I (intention to cross the boundary); fix a small bug and reconstruct half of the module - this is S (range expansion).

Repeat the modified file, repeatedly run the test that has been passed - this is T (task rotation). Shot in any one, this plug-in is worth installing.

中文

让 AI 导出一个文件,它顺手生成了 .sha256 校验和文件——没人要求,也没人读。这是 H(无用防御)。

让它只做代码 Review,diff 里却出现了文件修改——这是 I(意图越界);修一个小 bug 顺手重构了半个模块——这是 S(范围膨胀)。

重复读改过的文件、反复跑已经通过的测试——这是 T(任务打转)。中枪任何一条,这个插件都值得装。

第一步:对号入座,确认你也需要它
2

Step 2: Claude Code Installation (Five Commands)第二步:Claude Code 安装(五条命令)

Download the stop-that-shit source code package (see button above) from this station and extract it to the extracted directory; confirm that node -v shows 18 or higher.

Execute: claude plugin validate. (Checklist)→ claude plugin marketplace add./(Register local directory as plugin marketplace)→ claude plugin install stop-that-shit @ stop-that-shit.

Restart Claude Code to finish loading. After it is installed, it automatically registers eight hooks: the session starts to inject behavior contracts, the classification interception before the tool is called, the sub-agent starts the synchronization contract, etc., without additional configuration.

中文

从本站下载 stop-that-shit 源码包(见上方按钮)并解压,进入解压出的目录;确认 node -v 显示 18 或更高。

依次执行:claude plugin validate .(校验清单)→ claude plugin marketplace add ./(把本地目录注册为插件市场)→ claude plugin install stop-that-shit@stop-that-shit。

重启 Claude Code 完成加载。装好后它会自动注册八个钩子:会话开始注入行为契约、工具调用前分类拦截、子代理启动同步契约等,无需额外配置。

第二步:Claude Code 安装(五条命令)
3

Step 3: Codex/OpenCode Installation and Verification第三步:Codex / OpenCode 安装与验证

Codex user two commands: codex plugin marketplace add lennney/stop-that-shit --ref 0.2.5, then codex plugin add stop-that-shit @ stop-that-shit, restart Codex.

OpenCode, Hermes Agent CLI, Pi, Oh My Pi users: There are corresponding opencode/, hooks/, pi/, omp/adaptive directories in the source code package, access according to the corresponding section INSTALL.md in the package.

Do a small experiment to verify: let the AI "export a copy of report.csv" - it is likely to write a .sha256 before loading, and only this file should be delivered after loading; then let it "only review and not modify" to observe whether the diff is kept clean.

中文

Codex 用户两条命令:codex plugin marketplace add lennney/stop-that-shit --ref 0.2.5,然后 codex plugin add stop-that-shit@stop-that-shit,重启 Codex。

OpenCode、Hermes Agent CLI、Pi、Oh My Pi 用户:源码包里有对应的 opencode/、hooks/、pi/、omp/ 适配目录,按包内 INSTALL.md 对应小节接入。

做个小实验验证:让 AI「导出一份 report.csv」——装之前它很可能顺手写个 .sha256,装之后应该只交付这一个文件;再让它「只 Review 不修改」,观察 diff 是否保持干净。

第三步:Codex / OpenCode 安装与验证
4

Step 4: Border Tune-up第四步:边界调教

Legitimate checksum blocked: If your publishing process does read the checksum and rejects the file if it is inconsistent, release it with the hash = allow command - the rule blocks unread checksums, not all checksums.

Temporary Release Restrictions: Explicit corrections are read by the UserPromptSubmit hook in conversations using $ stop-that-shit directives or natural language (for example, "This time allows you to change your profile").

Review source code re-trust: The author opens all executables (hooks/hooks.json, hooks/stop-that-shit-claude.cjs, src/) for review, and npm test, npm run eval can locally rerun 18 decision use cases.

中文

合法校验和被拦:如果你的发布流程确实会读取 checksum 并在不一致时拒绝文件,用 hash=allow 指令放行——规则拦的是没人读的校验和,不是所有校验和。

临时放开限制:在对话里用 $stop-that-shit 指令或自然语言明确纠正(例如「这次允许你改配置文件」),UserPromptSubmit 钩子会读取这些显式修正。

审查源码再信任:作者把全部可执行面(hooks/hooks.json、hooks/stop-that-shit-claude.cjs、src/)开放审查,npm test、npm run eval 可本地复跑 18 个判定用例。

第四步:边界调教
5

Problem Encountered遇到问题

Unresponsive: confirm that the host has rebooted and the Node is not less than 18; blocking is too strict: all rejections will indicate which rule has been triggered and will be explicitly released as needed. See the download button above for the companion source code package for this article.

中文

装完没反应:确认宿主已重启、Node 不低于 18;拦截太严:所有拒绝都会说明触发了哪条规则,按需显式放行。本文配套源码包见上方下载按钮。

读完接着看 · Keep reading

想马上用起来?去「AI工具」栏挑一个直接下载,或在「AI教程」里跟着图文步骤做一遍。

去 AI工具 → 看 AI教程 →
0阅读0 条评论

阅读与点赞数据保存在你的浏览器本地,欢迎留下你的想法。

评论 文明发言,让讨论更有价值

正能量公益广告今日正能量学一点,用一点;今天种下的种子,会长成明天的能力。去免费下载专区 →广告