AI资讯大全AIPPXP.CN搜索 ↗
插件介绍 · 附完整源码包

stop-that-shit: Chinese open source "Mission Boundary Guard", a red stop seal to shut up when the AI is donestop-that-shit:中文开源「任务边界守卫」,一个红色 STOP 印章让 AI 干完正事就闭嘴

You only ask the Agent to export one file, and it calculates the SHA-256 that no one reads, saying "just in case"; you let it do only Review, and it finds the problem and starts to change the code. This Chinese open source plugin, whose name is even straightforward enough, uses a set of Skill plus Guard hooks to stop the four types of redundant actions of AI in real time in mainstream hosts such as Codex, Claude Code, OpenCode, and Hermes Agent CLI - range expansion, useless defense, intention to cross the border, and task rotation. The official benchmark can save about 65% of the nonsense output tokens, and the technical content is not discounted at all.

你只让 Agent 导出一个文件,它顺手算了份没人读取的 SHA-256,说是「以防万一」;你让它只做 Review,它找到问题就开始改代码。这个连名字都够直白的中文开源插件,用一套 Skill 加 Guard 钩子,在 Codex、Claude Code、OpenCode、Hermes Agent CLI 等主流宿主里实时拦下 AI 的四类多余动作——范围膨胀、无用防御、意图越界、任务打转,官方基准能省约 65% 的废话输出 token,而且技术内容一点不打折。

2026-11-01 更新 · 免费
stop-that-shit:中文开源「任务边界守卫」,一个红色 STOP 印章让 AI 干完正事就闭嘴

Shit Four Deadly Sins: What's It Really Stopping?SHIT 四宗罪:它到底拦什么

S——Scope creep range expansion: fix a problem and reconstruct the irrelevant module. Its principle is that the necessary callers, data migration and testing should be done together, and stop without the extension of the current use.

H - Useless defense: No one reads the checksum, a compatibility layer for the imagined future. Typical example in the document: downstream only reads report.csv, but the export logic additionally writes report.csv.sha256 - directly cut off according to the rules, and the file is delivered as usual.

I - Intention to cross the boundary: The file was changed after the read-only review was said. For the review to report the problem, the modification needs to be authorized, and the boundaries delineated by the user must be counted.

T - Task rotation: No new problems or changes, read, tested and reviewed again. If you can reuse the evidence, reuse it. When you have done it and verified it fully, it will be over.

中文

S——Scope creep 范围膨胀:修一个问题顺手重构无关模块。它的原则是必要的调用方、数据迁移和测试要一起做全,没有当前用途的扩展就停下。

H——无用防御:没人读取的 checksum、为想象中的未来搭的兼容层。文档里的典型例子:下游只读 report.csv,导出逻辑却额外写了 report.csv.sha256——按规则直接砍掉,文件照常交付。

I——意图越界:说了只读 Review,文件还是被改了。审查就报告问题,修改需要授权,用户划的边界必须算数。

T——任务打转:没有新问题或变化,读过、测过、审过的又来一遍。能复用证据就复用,做完、验证充分,就结束。

SHIT 四宗罪:它到底拦什么

It's not a hint, it's eight real hooks.不是提示词,是八个真钩子

Many "anti-deviation" schemes just pile up rules in AGENTS.md. The author tried it - one for each gas, and the rule file itself began to make history. This plugin is changed to register the real hook at the host level: SessionStart and SubagentStart inject the current contract into the new session, PreToolUse classifies the upcoming action, and returns permission deny directly if necessary.

PostToolUse reads delegation results, PostToolUseFailure reserves capacity until completion of unconfirmed, PermissionDenied releases reservations for unstarted calls, SessionEnd keeps only pending activities without assuming completion. In other words, it governs behavior, not the mouth.

The project comes with 18 public Bad/Good decision use cases and a complete chain of evidence (EVIDENCE.md), including historical Codex/GPT-5.6 run against - not a rule of thumb.

中文

很多「防跑偏」方案只是往 AGENTS.md 里堆规则,作者试过——每气一次补一条,规则文件自己也开始造史。这个插件改成在宿主层面注册真实钩子:SessionStart 和 SubagentStart 把当前契约注入新会话,PreToolUse 对即将执行的动作分类、必要时直接返回 permission deny。

PostToolUse 读取委派结果、PostToolUseFailure 在完成未证实前保留容量、PermissionDenied 释放未启动调用的预留、SessionEnd 只保留未决活动而不假设完成。也就是说它管的是行为,不是嘴。

项目自带 18 个公开的 Bad/Good 判定用例和完整证据链(EVIDENCE.md),包括历史 Codex / GPT-5.6 对照运行——不是拍脑袋定的规矩。

不是提示词,是八个真钩子

Installation: Mainstream Host Full Coverage安装:主流宿主全覆盖

Claude Code route: Unpack the source code package of this site into the directory, execute claude plugin validate., claude plugin marketplace add./, claude plugin install stop-that-shit @ stop-that-shit, and restart after loading. Node.js 18 + is required.

Codex route: codex plugin marketplace add lennney/stop-that-shit --ref 0.2.5, then codex plugin add stop-that-shit @ stop-that-shit, restart takes effect.

OpenCode, Hermes Agent CLI, Pi, Oh My Pi also have corresponding adaptive directories (opencode/, hooks/, pi/, omp/), accessed according to one command INSTALL.md. Users who will look at the source code can review all executable aspects such as hooks/stop-that-shit-claude.cjs, and the author explicitly encourages you to verify before trusting.

中文

Claude Code 路线:解压本站源码包进入目录,依次执行 claude plugin validate . 、claude plugin marketplace add ./ 、claude plugin install stop-that-shit@stop-that-shit,装完重启。需要 Node.js 18 以上。

Codex 路线:codex plugin marketplace add lennney/stop-that-shit --ref 0.2.5,再 codex plugin add stop-that-shit@stop-that-shit,重启生效。

OpenCode、Hermes Agent CLI、Pi、Oh My Pi 也都有对应适配目录(opencode/、hooks/、pi/、omp/),照 INSTALL.md 一条命令接入。会看源码的用户可以审查 hooks/stop-that-shit-claude.cjs 等全部可执行面,作者明确鼓励你先验证再信任。

安装:主流宿主全覆盖

Problem Encountered遇到问题

Unresponsive after loading: first confirm that the host has restarted and the Node version is not less than 18; the legal checksum is blocked by error: if the publishing process really reads the checksum and rejects the file if it is inconsistent, use hash = allow to release it. See the download button above for the companion source code package for this article.

中文

装完没反应:先确认宿主已重启、Node 版本不低于 18;合法校验和被误拦:如果发布流程真的会读取 checksum 并在不一致时拒绝文件,用 hash=allow 放行。本文配套源码包见上方下载按钮。

资源下载 · Download

⬇ Download · 点击下载:stop-that-shit 0.2.5 完整源码包

MIT 许可 · 含 Claude Code / Codex / OpenCode 全套适配与 18 个判定用例

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