Stop That Shit
A skill plus hooks that keep a coding agent on task and cut speculative defenses, unread checksums and scope creep
Medium risk
We rate an entry medium when the tool runs code, makes network calls or reads project files. Check what exactly it does before installing.
Why this level
- Installs hooks that intercept the agent's tool calls and can deny an action
- Runs local Node.js code within the agent lifecycle
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/stop-that-shitDetects the agents on your machine, checks the risk and pins the version.
Other ways to install
Run one by one in the Claude Code chat
/plugin marketplace add lennney/stop-that-shit
/plugin install stop-that-shit@stop-that-shitClone lennney/stop-that-shit, then in Claude Code run claude plugin marketplace add ./ and claude plugin install stop-that-shit@stop-that-shit, restart the agent after install and review the hook executables.
Other ways from the author
claude plugin marketplace add ./
claude plugin install stop-that-shit@stop-that-shitClone the repository first, run the commands from its root, then restart Claude Code. Node.js 18 or newer is required.
This is third-party code. Review the repository files before installing.
What it does
The project has two parts: a skill with rules and guard hooks. The skill advises the agent to finish the work the task needs while dropping what it does not: unread checksums, compatibility layers for an imagined future, edits during a read-only review, and repeated verification with no new question. The hooks plug into the agent lifecycle, classify an action before a tool call and can return a permission deny, and they also track subtask delegation and scope boundaries. Rules are set with directives in the prompt text and run locally without network calls. Supported hosts are Claude Code, Codex, OpenCode, Hermes Agent CLI and Pi.
Who it is for. For developers tired of babysitting a coding agent who want machine-enforced boundaries instead of long rules in AGENTS.md.
Good fit when
- The agent tends to rewrite neighboring modules and expand the task scope
- The agent adds unneeded defenses and checksums nobody reads
- You need to machine-block edits during review and redundant re-verification
Not a fit when
- Your agent is not among the supported hosts
- You want plain guidance without installing hooks into the agent lifecycle
Example request
Only review this module and list the issues, do not change any filesLimitations
The skill alone only advises and does not guarantee model behavior. Machine enforcement comes from the hooks, which need Node.js 18 or newer, and the set of intercepted events depends on the host. Codex records trust against the hook definition hash, so every install and update must be reviewed by hand. The authors recommend inspecting the hook and adapter executables before trusting them.
How to disable. Remove the plugin through your host's plugin commands, for example claude plugin, and delete the hook entries from the configuration. Restart the agent afterwards.
Security check
- Installs hooks that intercept the agent's tool calls and can deny an action
- Runs local Node.js code within the agent lifecycle
README in short
The README frames the problem as SHIT behavior: scope creep, hashing and hypothetical hardening, intent violation and task thrashing with no result. In response the project offers a stop ladder where the agent first understands the current responsibility, starts with a direct solution, closes a concrete gap, judges defenses by their effect and finishes after verification. Installation is a plugin via a local marketplace in Claude Code and via the repository marketplace in Codex, both needing Node.js 18 or newer. It ships eighteen public good and bad decision cases, tests and evidence records. MIT licensed.
FAQ
Does this make the agent weaker?
No. The skill requires finishing the work the task needs, including caller edits, migrations and tests. It removes only what the task does not require.
Does the skill work without the hooks?
Yes, the skill advises even without hooks, but the machine-enforced denials come from the hooks on supported hosts.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent