Skills CLI
Installs, updates and removes agent skills for Claude Code, Codex, Cursor and more
by Vercel Labs
Stop That Shit
by LennneyPluginWrite code
A Skill and Guard that set task boundaries for your coding agent
The problem it solves
You ask your coding agent for one small change and it adds checksums nobody reads, refactors nearby code and repeats checks it already ran.
For DevelopersCodex usersClaude Code users
- Before
- You keep adding rules to AGENTS.md and still spend your time supervising the agent.
- After
- Once a directive arms the Guard, it stops or asks on covered actions, such as a write in review mode or a new dependency, unless you allow them.
- Stars
- ★ 3.7k
- License
- MIT
- Cost
- Free
- For
- Codex
- Contents
- Includes scripts
Overview
Stop That Shit is a Skill plus a Guard for AI coding agents. The Skill tells the agent to finish the task you gave and skip unrequested extras such as unread checksums. Once a directive arms it, the Guard checks a few explicit boundaries, such as writes in review mode, new dependencies and writes outside files=, and returns a denial, and you can allow one on purpose, for example with hash=allow. It suits developers who use Codex, Claude Code or OpenCode.
You steer it with one-line directives, and its decisions are logged on your own machine as metadata only.
Try these first
Start with the first one: it is a small task you can check easily.
- $stop-that-shit review -- Review this repository. Report findings; do not edit.First try · read-only
- $stop-that-shit change -- Fix the failing config test.edits files
- $stop-that-shit status
Example prompts written by Agent Flows Hub.
What you get
From the creator's documentation.
| Output | What it is | Reads |
|---|---|---|
| Review mode | Reports findings without editing any file | Your project |
| Task boundaries | When armed, returns a denial for covered actions such as a write in review mode or outside files= | Agent tool calls |
| Allow on purpose | Authorize a checksum or a dependency change with a directive such as hash=allow | Your directive |
| Local decision log | Stores metadata-only events about Guard decisions in a host data folder | Your computer |
| Hook commands | Runs Node.js hook commands that you inspect before they are enabled | Your computer |
Good to know
You ask a coding agent for a small change and it adds extras you did not request. This plugin gives the agent a task-boundary Skill and, once armed, a Guard that checks a few explicit boundaries.
Limits
- After install the Guard only observes and blocks nothing until a review, answer, monitor or change directive arms it.
- The Skill is advisory; the Guard checks only explicit, covered boundaries and the host effect of a denial is unobserved.
- The Guard needs Node.js 18 or newer, and Codex asks you to inspect each packaged hook in /hooks before it is enabled.
Notes by Agent Flows Hub (original project name: Stop That Shit), written from the creator's documentation, not from running it ourselves. Stop That Shit: README · Stop That Shit: install guide
Stuck setting it up?
Tell us what happened. We take a look first; if it needs paid work, you get an estimate before anything starts.
More to write code
Other plugins
- Telegram for Claude Code Lets you message your running Claude Code session from Telegram and get replies there
- text-to-cad Lets your AI agent design 3D parts from a description and export files ready to print
- UI Design Skills Helps you choose color palettes, type scales, grids and layouts for app or web screens
- Understand Anything Turns a codebase or notes folder into a clickable map with plain-English explanations
- Zapier for Claude Code Walks you through connecting Zapier to Claude Code and running your first app action
- Animate Makes short animated explainer videos in one of seven drawing styles, rendered to MP4