Guide · Agent skill
How to use the Verify and Stop agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Verify and Stop is an agent skill from the caveman repository that makes an AI agent act as a focused checker. It is meant for confirming that work which is already finished meets its acceptance conditions, without the agent wandering into new changes or extra tests. This guide explains what the skill does, how to install it, how to trigger it, and what to watch for, so you can use it for a quick completion check or a final gate before merging.
Verify and StopJuliusBrussee/caveman · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Download the skill ZIP from our page.
- 2Choose an install location. Pick personal (~/.claude/skills/) or project (.claude/skills/ in your project).
- 3Unzip the skill. Unzip the downloaded file into the chosen skills folder.
- 4Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 5Write your acceptance conditions. Describe the finished work and the criteria it must meet, in concrete terms.
- 6Ask for a validation-only check. Request a completion check and state that you do not want code changes.
- 7Read the report. Review the commands run, what passed or failed, and any remaining risk.
What is an agent skill?
An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a request matches the skill's description. The skill tells the agent how to approach a specific kind of task.
The verify-and-stop skill is licensed under Apache-2.0 and is published in the JuliusBrussee/caveman repository under skills/verify-and-stop.
What the skill makes the agent do
The skill turns the agent into a validator. It translates your acceptance conditions into the smallest set of checks that can prove they are met, runs focused checks before wider ones, and stops as soon as the proof is complete.
It reuses earlier results when the repository state still matches, and it does not edit product code unless your request includes fixes. It also avoids adding polish, cleanup, or unrelated tests once the criteria pass.
- Reuses still-current results when the repository state matches
- Runs focused checks before wider gates
- Reports pass, fail, unavailable, and blocked outcomes exactly
- Does not edit product code unless fixes are requested
- Stops once acceptance proof is complete
- Reports commands run, results, and unresolved risk only
How to install the skill
Download the ZIP of the skill from our page. You can install it in one of two ways.
For a personal install, unzip it into ~/.claude/skills/. For a project-only install, unzip it into the .claude/skills/ folder inside your project.
In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Project: unzip into .claude/skills/ in your project
- Claude apps: Settings → Capabilities → Skills, then upload
How to trigger the skill with natural requests
You do not need a special command. Describe a validation-only task in plain language, and the agent should match it to the skill's description, which covers validation-only tasks, completion checks, focused gate runs, and last-mile proof.
Be explicit that the work is finished and you want proof, not changes. If you do want fixes, say so in the request, because the skill only edits product code when the request includes fixes.
- Check that this finished feature meets its acceptance criteria and stop once proven.
- Run a completion check on the login work before I merge. Do not change any code.
- Give me a final verification gate for the export feature and report what remains risky.
- Verify the billing changes and fix anything that fails.
What the skill needs and what it does not do
The source lists no requirements. It does not run scripts and does not use the network. Its files are agents/openai.yaml and SKILL.md.
The skill is not for building or fixing features. Because it forbids product code edits unless fixes are requested, you need a different approach for development work. It also deliberately limits checks to what is sufficient to prove the acceptance criteria, so it is not a tool for broad test coverage.
- Requirements: none listed in the source
- Runs scripts: no
- Uses network: no
Tips for getting useful results
State your acceptance conditions clearly in the request, since the agent translates them into the proof set. Vague criteria lead to vague proof.
Ask for the report you need. The skill reports commands, results, and unresolved risk only, so ask specifically if you want more detail on a failure.
- Write acceptance conditions as concrete, checkable statements
- Keep fixes out of validation requests unless you want edits made
- Expect unavailable or blocked checks to be reported as such rather than guessed
What people use it for
- Confirming a finished feature meets its acceptance criteria before a merge
- Running a final verification gate before handing work off to another team
- Checking that a last-mile change is proven without adding extra polish or tests
- Getting a clear pass, fail, unavailable, or blocked status for each criterion
Frequently asked questions
Do I need anything installed to use the verify-and-stop skill?
The source lists no requirements for this skill. It does not run scripts and does not use the network. You install it by placing the unzipped skill folder in ~/.claude/skills/ or a project's .claude/skills/, or by uploading it in the Claude apps.
Is the verify-and-stop skill free?
The skill is licensed under Apache-2.0, as listed on this page. Check the repository's licence file for the full terms.
Can verify-and-stop fix bugs it finds?
Not by default. The skill tells the agent not to edit product code unless the verification request includes fixes. If you want fixes, say so explicitly in your request.
How do I install verify-and-stop in Claude?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or .claude/skills/ inside a project. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
Is verify-and-stop a replacement for a full test suite?
No. The skill deliberately runs only the smallest set of checks needed to prove the acceptance criteria and does not aim for broad test coverage. Use a full test suite when you need wide coverage.
What is the difference between verify-and-stop and a general coding skill?
Verify-and-stop is for confirming finished work and stops once proof is complete. It explicitly avoids feature building or fixing, so a development-focused skill is better suited to changing code.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Verify and Stop suits developers and teams who want a quick, scoped completion check or a final gate on finished work, with no scope creep and no code edits. It is a poor fit if you need the agent to build or fix features, or if you need broad test coverage, so pick a different skill for those tasks.
Download the skill