agentleFS
SantiagoVqzSantiagoVqz/claude-skills/skills/3-verify/ui-check/SKILL.md

ui-check

Skill2 starsChanged 12 days agoMITView on GitHub

Drive the app in a browser or an iOS simulator to check a change, then give the user a short list of what passed and what to check by hand. Use when the user runs /ui-check, or asks for a browser check of a change.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the target: the issue and its criteria, or the text the user gave.

    line 13

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

UI check

/ui-check runs the app, does the main flow of a change in a browser, and reports. It changes no code. The user decides when it runs. The server and the tab stay open until the user answers, so the user can look for themselves.

1. Pick the flow

Read the target: the issue and its criteria, or the text the user gave. With no argument, use the issue in the current branch name.

Write the flow as a short list of steps, the way a user does the feature from start to end. Cover the criteria that a browser can show. A criterion that a browser cannot show stays out. A screen width, such as 390px, is a parameter of a step, not a new step.

Completion: you have the flow, five steps or fewer.

2. Run

Print the flow as a numbered list in your message, before any tool call. Print it also when a caller gave the flow: the user cannot see thinking or the caller's arguments.

Read the run section of the README or AGENTS.md, and start each service that the pages need first, such as the database or the API. Send one line before each service that you start, because the user sees nothing else meanwhile. When the worktree's database container is missing, say so before you start anything. Then start the dev server in the current worktree, in the background, with no fixed port. Read the URL from the server's ready line.

Check that the dev database holds the rows that the flow needs. Take the database name and the user from the provision output of the worktree, and write them as literals, because the guard refuses a $VAR. Run the migrate command on the fork first, so that it has the branch's columns. Print no value from an env file, because a line such as PROD_DATABASE_URL holds a password. To see which variables exist, list the names only: grep -oE '^[A-Z_]+=' <file>.

When rows are missing and the database is a worktree fork, with a branch suffix such as <app>_<branch>, make the fewest rows that the flow needs, and ask no question, because the fork is disposable. Use the project's seed script or the API. When neither can make the rows, use SQL in the fork: read \d <table> first, read the allowed values from an existing row or the model, and give each row the timestamps that the list orders by. Restore each row that you change for one step, such as a role or a module flag. For any other database, ask the user first. The report names the rows that you made and the rows that you restored.

For a native iOS app, read SIMULATOR.md and follow it in place of the server and browser parts of this step.

Load the Claude in Chrome tools and browser_batch in one ToolSearch call, and open a new tab. When the extension is not connected, ask once in this session with the question tool: "Connect the extension" or "Use DevTools". The extension uses the user's Chrome, so it keeps their sign-ins. DevTools starts signed out. On "Connect the extension", load the tools again. On "Use DevTools", or when the extension still does not connect, use the Chrome DevTools tools, and read DEVTOOLS.md first.

Do each step of the flow in one browser_batch call. When a step needs a ref, do one find call first, then the batch. End each batch with a read of the page, and compare it with what the step expects. Before you mark a step ✗, read the URL and the console.

When the app goes to a sign-in page, ask the user in the chat to sign in in that window, and end your turn. Do not poll with wait_for: the sign-in can take longer than its timeout. Type no password yourself. When the user replies, do the step again.

Completion: the numbered flow is in the chat above the first tool call, and each step passed or failed. The dev server and the tab stay open.

ExcuseAnswer
"I have the flow in mind."The user cannot see thinking. Print the numbered flow before the first tool call.
"Another session holds the browser, so I kill it."Its check runs in that browser. Connect the extension, or skip the browser check.
"The extension is not connected, so I skip the check."Ask, then use DevTools. Only a failure of both skips the check, and the report says so.

3. Report

Print the two lists in your message, one line each, also when a caller ran you. When a caller skill ran you, stop here: the caller asks the question and stops the server. When the user ran you, or you started yourself with no caller skill, ask with the question tool: "Approve" or "Change".

Verified:
- ✓ <step> · <what you saw>
- ✗ <step> · <what you saw instead>

Data: <the rows that you made, or "none">

Check yourself at <URL>:
1. <what to do> · <what the user sees>

"Check yourself" has three steps or fewer: what a browser check cannot judge, such as the look and the feel.

Completion: the lists are in the chat. When the user ran you, they answered, and you stopped the dev server and closed the tab.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.