Stop That Shit

Finish the task. Know when to stop.

One export came with an unread SHA-256 file. It prompted a question: what does each extra step an agent adds actually change?

Stop That Shit is a task-boundary Skill + Guard for AI coding agents. The Skill uses the Stop Ladder to guide implementation and acceptance; the Guard checks explicit task, file, dependency, hash, and subagent boundaries on supported Host paths. STSS guides rewriting, auditing, and drafting proposals, articles, case studies, and captions.

In this article

The same checksum. A different reason to keep it.

One scribe passes a scroll to another across a low table, leaving an unused bundle aside.

Not consumed

result.csv and an unused checksum

The next step reads only the output file. The unused checksum can be omitted.

Used to verify

File → checksum → release verifier

The release verifier reads the checksum and rejects a mismatch. Keep this step.

Two scenarios: the checksum’s role in the current delivery changes the decision. See the Stop Ladder below for the full criteria.

How it works

For people working with coding agents

Define the task and acceptance

What must be delivered, and which actions are allowed.

The Skill evaluates necessity

Use the Stop Ladder to find evidence that the current task needs this step.

Guide the agent’s choices

Complete requested work and necessary follow-through; omit unsupported extras.

The Guard checks boundaries

Check the declared mode, file, and action boundaries.

Check covered actions

With an explicit task mode armed, return an allow or deny result.

The Skill guides implementation and acceptance. The Guard observes by default and checks supported Host action paths after an explicit task mode is armed.Read the Skill and Guard responsibilities

Follow the Stop Ladder. Complete the current responsibility.

Establish the result and existing guarantees, start with a direct solution, close concrete gaps, judge defenses by their effect, then verify and finish.

  • What result, boundaries, and existing guarantees must the task satisfy?
  • Can existing code, standard libraries, platform features, or installed dependencies do it?
  • Which concrete gap still needs to be closed?
  • What does this defense detect, and how does rejection, recovery, or diagnosis change the result?
  • Are the result and necessary follow-through complete, with sufficient relevant verification?

Complete the requested work and necessary follow-through. Stop extra work that has no current purpose.

Read the evidence and method

Installation and support

Documented version: 0.2.5

The Codex commands below pin a release tag. After restart, compare /hooks in the CLI TUI with that tag's Hook manifest and trust the commands you reviewed. GitHub Release has the complete changes. STSS can also be installed on its own.

CODEX / POWERSHELL
codex plugin marketplace add lennney/stop-that-shit --ref 0.2.5
codex plugin add stop-that-shit@stop-that-shit

After Hook review, use $stop-that-shit review or $stop-that-shit change in a new task to make authority explicit. The core Guard needs Node.js 18+. Pi 0.84.4 itself needs Node.js 22.19+; the recorded Oh My Pi 18.4.4 checks used Bun 1.3.14. See the versioned INSTALL.md for each Host’s requirements and installation steps.

One task boundary, 6 Host adapters

Codex
Plugin + packaged Hooks
Claude Code
Plugin + native lifecycle Hooks
OpenCode
V1 / V2 plugins + before-action checks
Hermes Agent CLI
Native plugin callbacks
Pi
Native Extension + before-action checks
Oh My Pi
Explicit OMP Extension + native /sts controls

Install guide on GitHub

Common questions

Getting started

What does it do?

Stop That Shit is a task-boundary Skill + Guard for AI coding agents. The Skill uses the Stop Ladder to guide implementation and acceptance; the Guard checks explicit task, file, dependency, hash, and subagent boundaries on supported Host paths. STSS guides rewriting, auditing, and drafting proposals, articles, case studies, and captions.

Is it a Skill or a plugin?

It is both. The standalone Skill provides the Stop Ladder and task modes such as review and change as advisory guidance. The complete plugin also packages the Guard and Host adapters, which check explicit boundaries on supported Hook paths.

How do I install it in Codex?

Use the current GitHub Release and that tag's INSTALL.md to install or update. Codex requires Node.js 18+: pin the marketplace to the tag shown in the install section, restart, and compare /hooks in a fresh CLI TUI with the packaged Hook manifest before trusting it. Use review or change to declare authority in a new task.

Which workflows is it useful for?

Run a review/change smoke in a disposable repository, then use paired Bad and Good Cases to judge false allows and false denials. EVIDENCE.md separates tests, adapter behavior, and final Host effects; those records do not prove a general improvement in model behavior.

Which coding agents are supported?

The stable release provides the adapters listed in the installation section, including the explicit Oh My Pi Extension and OpenCode V1 and V2. Installation, trust, restart, and event coverage differ by Host. Separate Pi child processes do not inherit the parent contract; see the installation section and versioned INSTALL.md.

Behavior and limits

How does it differ from AGENTS.md?

AGENTS.md is the right place for durable repository architecture, commands, and conventions. Stop That Shit does not replace that context. It lets the user declare per-task review/change, file, dependency, hash, and subagent boundaries and checks explicit actions on supported Hook paths.

What does SHIT stand for?

SHIT is not one execution mode. It names four kinds of task drift: Scope creep, Hashing and hypothetical hardening, Intent violation, and Task thrashing. The actual task modes include review, answer, monitor, and change.

What are its limitations?

It can inspect only Hook events that the Host exposes with enough input. It is not a security sandbox and does not label every migration, retry, or cache as unnecessary. A returned denial records the Guard response; it is not observation of the Host's final effect.

Does it ban every hash or checksum?

No. A release checksum read by a verifier has a consumer and is necessary for that acceptance path. The current task can authorize an identifiable new hash action with hash=allow; the request and consumer still determine whether it belongs.

Versions and sources

What is STSS?

STSS is a standalone writing Skill for proposals, articles, case studies, captions, and new drafts. Use rewrite to edit text and audit to report problems. Its Sentence Consumer Test, Claim Ledger, and Claim Diff help remove unused disclaimers, hedge stacks, and self-defense while preserving conditions, attribution, and responsibility that affect the reader’s decision.

Where are the source and releases?

The public source, releases, and versioned download documentation are at github.com/lennney/stop-that-shit. This page is the bilingual product guide maintained by lennney; use the GitHub tag and release as the source of truth for code, versions, and installation steps.

Install guide on GitHub

Keep reading

Codex Exported One File, Then Added an Unused SHA-256

How Defensive Engineering Grows Inside an AI Agent Task

Stop That Shit vs AGENTS.md: Skill, Plugin, and Task Boundaries

View the source

Back to the project story