Guide · Agent skill
How to use the Product Flow Audit agent skill (audit)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The audit skill is an agent skill that lets an AI agent review a product experience such as a signup, checkout or settings path. Instead of giving a general opinion, the agent opens the product in a browser, walks each step, saves and checks a screenshot of every screen, and writes a numbered critique tied to that evidence. This guide explains what the skill does, how to install it, how to trigger it, and what it needs to run.
Product Flow AuditXiaomiMiMo/MiMo-Code · MIT Download the skill Quick start
- 1Download the ZIP. Download the audit skill ZIP from this page.
- 2Unzip into a skills folder. Unzip into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project.
- 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 4Confirm a browser tool is available. Make sure the Browser skill, Chrome or Computer Use can capture the product.
- 5Make a natural request. Ask the agent to audit, review or critique a named flow, such as checkout or onboarding.
- 6Read the inline report. Check the numbered steps, screenshots, findings and evidence limits.
- 7Decide on Figma. Answer the one-time offer to plot the flow in Figma only if you want a board.
What is an agent skill, and what does audit make the agent do?
An agent skill is a folder of instructions that an AI agent loads when a request matches the skill's description. The audit skill is one such skill, taken from the product-design bundle in the MiMo-Code repository.
When triggered, the agent identifies the product or surface and the flow or task, captures the flow screen by screen, inspects each screenshot, and returns an inline report. The report includes the screenshots, a numbered step list, UX and design findings tied to steps or screenshots, accessibility risks tied to steps or screenshots, and clear limits on what could not be checked from screenshots alone.
- Captures screenshots of each step before writing notes
- Rejects screenshots that are blank, loading, cropped, blocked or showing the wrong state
- Covers usability, visual design and accessibility risks
- Notes what it could not verify from screenshots alone
- Creates a Figma board only if you explicitly ask for one
Install the audit skill
Download the ZIP for the skill from this page. Then choose one of the two install routes below.
For a personal install, unzip the file into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Single project: unzip into .claude/skills/ inside the project
- Claude apps: Settings → Capabilities → Skills, then upload
Check what the skill needs before you run it
The skill needs a way to capture the browser and access to the product you want audited. The source lists the Browser skill, Chrome or Computer Use as capture tools. The agent cannot audit a flow it cannot reach.
The bundle also references companion skills (index, user-context, critical-overrides and browser) and a bundled design audit framework reference file. The skill runs scripts and uses the network, so review the safety notes below first.
- A browser capture tool: the Browser skill, Chrome, or Computer Use
- Access to the product being audited, such as a live or staging build
- Optional: Figma tools and skills, only if you ask for a Figma board
- Bundled files: SKILL.md, references/design-audit-framework.md, agents/openai.yaml
How do you trigger the audit skill with a natural request?
You do not need a special command. Describe the experience and use a review word such as audit, review, critique, inspect, assess, analyze, evaluate or feedback. The skill's description lists these triggers, and the agent should load it when your request matches.
Name the flow and, if you can, the product URL or surface. Ask for a specific path, such as onboarding or checkout, so the agent knows which steps to walk.
- "Audit the checkout flow on our staging site, from cart to confirmation."
- "Review the onboarding path for new accounts and give me UX and accessibility findings."
- "Critique the settings path for changing a password."
- "Give me feedback on the signup funnel with screenshots of each step."
What does the report look like, and what does it not cover?
The default output is an inline report with screenshots shown in the chat, in flow order. It gives an overall verdict, numbered steps, the highest-impact changes and the evidence limits. Every finding points to the screenshot or step that supports it.
The skill does not claim full accessibility compliance from screenshots alone. Accessibility risks are described as what is visible in the screenshots and what still needs testing. After the report, the agent asks once whether you want the flow plotted in Figma.
- Use it for: a quick, evidence-backed review of onboarding or checkout
- Use it for: a structured UX critique of a live staging build
- Do not use it for: full WCAG compliance certification
- Do not use it for: flows behind logins or paywalls the agent cannot access; it stops and reports the blocker
- Do not use it for: a design file or prototype with no capturable product
Safety notes: scripts and network access
The skill runs scripts and uses the network. Its instructions say to run the user-context preflight script when local shell access is available. The skill also loads the browser and drives the flow by clicking, typing and pressing keys.
Because the agent acts in a real browser, point it only at products you are allowed to test. Use a staging build where possible, and do not give it credentials you would not want an agent to use. Figma is used only when you explicitly ask for it.
- Runs scripts and uses network access, as listed in the bundle
- Loads the user-context preflight script when shell access exists
- Clicks, types and presses keys during capture, so use test accounts and staging environments
- Figma tools are needed only for a board you explicitly request
Tips for better audits
Give the agent a clear start point and end point. If the flow needs a test account, have it ready. The skill rejects screenshots that show login walls, cookie dialogs or loading states, so a clean start improves the result.
If you have saved reference material such as a product URL, design system or brand assets, mention it. The skill uses saved references as grounding, but only inspects what the current task needs.
- Name the exact flow and its first and last screens
- Use a staging or test build with a test account
- Mention a design system or brand guide if you have one
- Ask for accessibility findings explicitly if you need them, and plan separate testing for compliance
What people use it for
- Reviewing an onboarding path for a new account to find friction and unclear next actions
- Critiquing a checkout flow on a staging build before a release
- Checking a settings path, such as changing a password, for clarity and error recovery
- Preparing screenshots and notes for a team design review
- Getting a structured first-pass accessibility risk list before manual testing
Frequently asked questions
How do I install the audit agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
What does the audit skill need to run?
It needs a browser capture tool such as the Browser skill, Chrome or Computer Use, plus access to the product being audited. Figma tools are optional and needed only if you ask for a Figma board.
Is the audit skill free, and what is its licence?
The source lists the licence as MIT for the MiMo-Code repository. The source does not state pricing, so check the repository and your own Claude plan for any costs.
Can the audit skill check full WCAG accessibility compliance?
No. The skill states it cannot claim full accessibility compliance from screenshots alone. It reports visible accessibility risks and notes what still needs testing.
Why does the audit stop partway through a flow?
If the flow sits behind a login or paywall the agent cannot access, or a step cannot be captured, the skill stops and reports the blocker rather than guessing. It also rejects screenshots that are blank, loading or showing the wrong state.
Can it audit a Figma file or prototype instead of a live product?
Not on its own. The skill relies on captured screenshots from a live or capturable product as evidence. Reviewing a design file or prototype without a capturable product is outside its scope.
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 the audit skill if you want a quick, screenshot-backed critique of a live onboarding, checkout or settings flow, and you have a browser tool and access to the product. Pick something else if you need full WCAG compliance certification, need to review a design file without a live build, or cannot give the agent access to the flow.
Download the skill