Guide · Agent skill
How to use the Audit, Verify, Explain agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Audit, Verify, Explain is an agent skill, a folder of instructions that tells an AI agent how to handle reviews and claims. Once it is installed, the agent audits what changed, checks it with direct evidence such as tests or logs, and explains the result in simple grade-5 language. This guide shows you how to install it, how to ask for it in plain requests, and what it will and will not do.
Audit, Verify, ExplainMengTo/Skills · MIT Download the skill Quick start
- 1Download the ZIP. Download the skill ZIP from this page.
- 2Unzip into the right folder. For personal use, unzip into ~/.claude/skills/. For one project, unzip into that project's .claude/skills/ folder.
- 3Or upload in the Claude apps. In the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.
- 4Make a review request. Ask in plain words, such as review this change or check whether this fix works, and name the evidence to look at.
- 5Read the verdict. Check the What changed, Why it matters, How I verified it, and What is still not proven sections.
What the skill makes the agent do
The skill treats every review or claim as three jobs: audit what changed or is being claimed, verify it with direct evidence, and explain it to a smart reader who is new to the topic.
The main rule is that the agent should not skip verification when files, commands, logs, tests, screenshots, or source data are available. It also should not present something as verified when it was only inferred. The agent is asked to say what was checked and what is still not proven.
- Audit: find the real source of truth, such as the diff and touched files for code changes, or the before and after behaviour for bug fixes.
- Verify: prefer automated tests, builds, or validators first, then running the app, then logs and live output, then static code reading, and label anything else as inference.
- Explain: use short sentences, define technical terms in plain words, and keep important caveats visible.
- Output shape: the skill suggests sections titled What changed, Why it matters, How I verified it, and What is still not proven. Very small answers can use a short paragraph instead.
Install the Audit, Verify, Explain skill
Download the skill ZIP from this page. Then choose one of the two install options below, depending on where you want the skill to be available.
For a personal setup, unzip the download into your ~/.claude/skills/ folder. For a single project, unzip it into that project's .claude/skills/ folder. Either way, the skill folder should contain SKILL.md.
If you use the Claude apps instead of a folder on your computer, open Settings, go to Capabilities, and upload the skill under Skills.
- Personal use: unzip into ~/.claude/skills/
- Project use: unzip into the project's .claude/skills/
- Claude apps: Settings → Capabilities → Skills, then upload
Trigger the skill with natural requests
You do not need a special command. The skill's description lists the kinds of requests it is meant for, so you can ask in ordinary words. Good triggers include asking the agent to review, audit, check, or verify something, to explain a change or a fix, to summarise test results, or to say whether something works.
You can also ask for a translation of technical findings into plain language for a non-technical reader. Be specific about what to look at, such as a diff, a test run, or a log, because the skill depends on real evidence.
- Review this change and tell me what could break.
- Check whether this bug fix really works and say what you tested.
- Summarise these test results for my manager in plain language.
- Verify that the checkout button no longer submits twice, and separate what you saw from what you guessed.
- Explain this fix to a non-technical reader.
Example: checking a checkout bug fix
The demo folder in the skill shows a realistic case. The input is a note saying the checkout button used to submit twice on a slow connection, and a guard was added to disable the button while the request is pending. The evidence listed is a targeted unit test output, a browser recording of one successful checkout, and the changed-file diff.
The expected result in the demo is a PASS status. The button locks after the first click, so a second order is not created, and the focused test and browser recording support that. The same result also states a remaining risk: retry behaviour after a failed payment was not exercised. That is the kind of honest gap the skill is designed to keep visible.
What the skill needs and how it stays safe
The listing says the skill requires nothing in particular. It does not run scripts, and it does not use the network. It is a set of written instructions in SKILL.md, plus an agents/openai.yaml file and demo files.
The skill cannot add tools of its own. It sets a process and a writing style. The agent still needs access to your files, tests, or logs to verify anything. If it has no access, the skill asks the agent to label its answer as inference.
- No scripts are run by the skill.
- No network use is listed.
- The skill is general guidance, not a specialist tool.
- Files included: SKILL.md, agents/openai.yaml, and the demo folder (expected-output.md, index.html, input.md, PROMPT.md).
Tips for getting useful answers
Give the agent the evidence up front. Point it to the diff, the test command output, or the log file. The more direct evidence it can read, the stronger the verdict can be.
Ask for the split between facts and judgment. The skill already asks for this, but you can make it explicit when you need to share the result with someone else.
- Name the files, test names, or logs you want checked.
- Say who will read the summary, so the explanation fits them.
- Ask what is still not proven when you need a clear list of risks.
- When tests cover only part of a workflow, expect the agent to say that the checked parts work, not that everything works.
What people use it for
- Getting a second look at a code change or bug fix before merging, with the agent reading the diff and related tests.
- Summarising a test run for a manager who does not read code, with the checked and unchecked parts clearly split.
- Checking a browser or app behaviour against a recording or logs, and noting any remaining risk.
- Turning a technical finding into a plain-language note for a non-technical reader.
Frequently asked questions
Is the Audit, Verify, Explain skill free to use?
The listing states the licence is MIT. The page does not list any price, so check the listing itself for current terms before relying on it.
What do I need to install the Audit, Verify, Explain skill?
The listing says nothing specific is required. Install it by unzipping the download into ~/.claude/skills/ or a project's .claude/skills/ folder, or by uploading it under Settings, Capabilities, Skills in the Claude apps.
Does the skill run scripts or use the internet?
No. The listing says it runs no scripts and uses no network. It is a set of written instructions in SKILL.md.
Why does the agent say something is only inferred?
The skill requires direct evidence such as tests, running the app, or logs. If the agent cannot access these, it is designed to label its answer as inference rather than say it works.
Can this skill do a security penetration test or legal compliance check?
No. The listing says it is not for deep specialist reviews such as security penetration testing or legal compliance, because it only sets a process and style and adds no domain tools.
How is this different from just asking an agent to review my code?
The skill gives the agent a fixed process: audit the source of truth, verify with evidence in a set order, then explain in plain language with a clear split between what was checked and what is still unproven.
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
Use this skill if you want a second look at code changes, bug fixes, or test runs, or you need plain-language summaries for non-technical readers, and you can give the agent access to the files, tests, or logs. Pick something else if you need a deep specialist review such as security testing or legal compliance, or if the agent cannot reach any evidence at all.
Download the skill