mantis
/mantis
Look at the UI before calling it done. Opens the running app in a real browser, checks every screen size, light and dark mode, and other languages, and reports what is actually broken: page overflow, cut-off text, overlapping controls, unreadable contrast, dark-mode leaks, tiny tap targets, console errors, broken images, layout shift. Use after any frontend change, before saying a UI task is done, when a UI, layout or screenshot check fails (a red check:ui, a failing mobile e2e test, a release blocked by one), or when the user says "check the UI", "does it work on mobile", "dark mode looks off
- Author
- vulinh1704 ↗
- Repository
- mantis ↗
- Licence
- MIT
- Last repository change
- 30.10.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 1
No signals detected. Repository changed on 30.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Look at the UI before calling it done. Opens the running app in a real browser, checks every screen size, light and dark mode, and other languages, and reports what is actually broken: page overflow, cut-off text, overlapping controls, unreadable contrast, dark-mode leaks, tiny tap targets, console errors, broken images, layout shift. Use after any frontend change, before saying a UI task is done, when a UI, layout or screenshot check fails (a red check:ui, a failing mobile e2e test, a release blocked by one), or when the user says "check the UI", "does it work on mobile", "dark mode looks off
Triggers from the description
Original author description
Look at the UI before calling it done. Opens the running app in a real browser, checks every screen size, light and dark mode, and other languages, and reports what is actually broken: page overflow, cut-off text, overlapping controls, unreadable contrast, dark-mode leaks, tiny tap targets, console errors, broken images, layout shift. Use after any frontend change, before saying a UI task is done, when a UI, layout or screenshot check fails (a red check:ui, a failing mobile e2e test, a release blocked by one), or when the user says "check the UI", "does it work on mobile", "dark mode looks off
How it works
- 1. Get the app running
- 2. Look
- 3. Read the evidence, then look at it
- 4. Decide
- 5. Report back
- Limits
SKILL.md preview
---
name: mantis
description: >
Look at the UI before calling it done. Opens the running app in a real browser,
checks every screen size, light and dark mode, and other languages, and reports
what is actually broken: page overflow, cut-off text, overlapping controls,
unreadable contrast, dark-mode leaks, tiny tap targets, console errors, broken
images, layout shift. Use after any frontend change, before saying a UI task is
done, when a UI, layout or screenshot check fails (a red check:ui, a failing
mobile e2e test, a release blocked by one), or when the user says "check the
UI", "does it work on mobile", "dark mode looks off", /mantis.
---
# Mantis
A mantis shrimp sees 12 color channels. People see 3. A coding agent that never
opens the browser sees 0.
**Rule:** a UI change is not done until you have looked at it rendered. "The code
looks right" is not evidence. A Mantis run with zero errors is.
## 1. Get the app running
- Use the project's own dev command (`npm run dev`, `pnpm dev`, `rails s`, …) and
note the URL. If it is already running, reuse it.
- List the routes your change touches. Add `/` when the change is global
(layout, theme, fonts, CSS reset).
- Pages behind a login or onboarding: pass `--storage-state <file>`, a Playwright
storage state with the session's cookies and localStorage. Ask the user for one,
or make it: `npx playwright@1.64.1 codegen --save-storage=.mantis/auth.json <login-url>`
(the user signs in and closes the window), or a short Playwright script that signs
in with a dev or seed user and calls `context.storageState({ path })`. When the
app only checks a localStorage flag (onboarding done), write the JSON by hand.
The file holds a live session: keep it in `.mantis/` or elsewhere out of git.
- If you can't reach a page, say so in your report instead of skipping it silently.
## 2. Look
```bash
mantis look http://localhost:3000 --routes /,/settings
```
If `mantis` isn't installed, run `npx -y @vulinh1704/mantis@0.7.1 look …` instead.
Without Node or a browser, use the Docker image (`ghcr.io/vulinh1704/mantis`, see the
README for how it reaches a dev server on the host).
- `--locales en,vi` when the app has translations. Longer languages (Vietnamese,
German) break layouts that look fine in English.
- `--locale-strategy` when the app doesn't follow the browser language (the report
says a locale is "not localized"). Check how the app switches language first:
`path:/{locale}` (`/vi/settings`), `query:lang`, `cookie:NEXT_LOCALE` (next-intl)
or `storage:i18nextLng` (i18next).
- `--locales en,pseudo` when the app has no translations yet (or one language
only). The pseudo cell shows every string about 40% longer, accented and
bracketed; cut or overflowing text there will break in real translations. Fix it
the same way (let the box grow or wrap), never by shortening the pseudo text.
- `--locales en,pseudo-rtl` when the app will ship Arabic, Hebrew or Persian but
doesn't yet: the pseudo text laid out right to left. Overflow on the left and
controls stuck to the wrong side usually come from physical CSS (`margin-left`,
`left:`, `text-align: left`); fix them with logical properties
(`margin-inline-start`, `inset-inline-start`, `text-align: start`).
- `--dark-strategy class:dark` or `--dark-strategy attr:data-theme=dark` when dark
mode is a class or attribute toggle rather than the OS setting. Check how the app
switches themes before choosing.
- Keep the default screens (mobile 360, tablet 768, desktop 1440). Never drop
mobile to make a report pass.
- Before you change styles, run `mantis learn <url>` (same routes) on the
unchanged app when `.mantis/style.json` doesn't exist yet. Every later `look`
then reports `off-style` warnings: colors, font sizes and radii that are near
copies of the house values, and fonts the app uses nowhere else.
- Exit code: `0` no errors, `1` errors found, `2` the page could not be loaded
(is the server up, on that port?).
## 3. Read the evidence, then look at it
1. Read `.mantis/latest/report.md`: issues grouped by type, with the selector, the
screens where each appears, and the screenshot file. When an earlier run of the
same URL exists, the report also compares with it: what was fixed, what is new,
and any `cover-up` (an issue hidden instead of fixed).
2. Open the screenshots of the worst screens in `.mantis/latest/shots/`. Errors
are boxed in red and warnings in amber, each with its issue number. Look at the
image before deciding; a selector alone can mislead.
3. `.mantis/latest/contact-sheet.png` shows every screen at once.
4. A clean report means clean for what Mantis measures, not "nothing is wrong".
Scan the sheet and the shots for what it can't judge: text over photos or
gradients, a control that is missing, spacing that looks off. Say what you saw.
5. One bug can hide another: an invisible toolbar can't overlap anything. After a
fix, look again before you call the report final.
## 4. Decide
- Errors in code you changed: fix them now with the **mantis-fix** skill, then
look again with the same command.
- Errors that were already there before your change: tell the user what you saw.
Don't silently rewrite unrelated code.
- Warnings: fix them when cheap and in scope; otherwise list them.
## 5. Report back
Say what Mantis checked (routes × screens × themes × languages), the error counts
before → after, and where the contact sheet is. Say plainly what you could not
check.
## Limits
- Mantis checks that the rendered page works, not whether it is beautiful. Pair it
with a design skill for taste.
- It doesn't click through flows or test hover and focus states yet. Visit a route
that shows the menu or dialog you changed, or say it was not checked.
- It doesn't measure animation smoothness. For load speed (LCP, blocking time,
slow API calls, third parties), use the **mantis-perf** skill.
- Contrast on canvas apps and under transparent layers is read from the
screenshot. Text over photos, CSS background images and gradients is not
checked: look at those yourself.
Versions
The database has no other skills named mantis.
Database snapshot: 13.11.2026 · package version d2a62cc3edc1 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100mantis-fix
Fix the UI problems Mantis found at their root cause instead of hiding them, then look again until the report is clean. Covers page overflow, cut-off text, overlapping controls, contrast, dark-mode leaks, tap targets, console errors, broken images and layout shift. Use after… - 100color-contrast
Use when choosing a palette, checking WCAG contrast, fixing unreadable text, or making light and dark themes that stay accessible. - 90verify-ui
Check a UI change in a real browser before calling it done. Starts or finds the dev server, opens the changed pages, takes screenshots at phone and desktop widths in light and dark mode, reads console errors, and looks for layout bugs that typecheck and curl miss. Use after any… - 98build-and-verify
Build a WinUI 3 app from the command line, install it, and check light and dark window snapshots of every page before calling a change done. Use after any UI code change, when a build fails, or when the user asks to run, screenshot, or check the app. - 100color-mode-and-theme
Light, dark or both, and when a theme selector is worth it. Use when defining a base palette or deciding on dark mode support. - 100build-and-verify
Build a SwiftUI macOS app from the command line, install it, and check light and dark window snapshots of every page before calling a change done. Use after any UI code change, when a build fails, or when the user asks to run, screenshot, or check the app.