The commands
Four commands.
Judge your front end, redo it, then hold every pull request to it. No front yet? Mock one up. In Claude Code.
/aeom judge
Judge your front end.
When before you change anything.
- Reads your product, then replays its key journeys in a real browser, on phone and desktop.
- Measures every screen: contrast, clickable elements, sideways scroll, console errors, the header on scroll, text that wraps or gets cut, tap targets, text size, line length, zoom, images, accessible names, keyboard focus, layout shift.
- Holds each screen to eight design principles, each journey to six of its own.
- Says what fails, in plain words. Changes nothing.
Claude Code · an example notes app
/aeom judgeFor 4 screens and 2 key journeys:Understand the product no agent the session ≈ 25 minJudge the screens and the journeys 6 agents ≈ 0.4 M tokens ≈ 10 minTotal, at most 6 agents ≈ 0.4 M tokens ≈ 35 mingoMeasured: 4 screens at 390, 1280 px, 16 checks and 8 principles each; 2 journeys on 6 principles. ✗ / contrast: .hint "Saved": text contrast is 2.6:1, needs 4.5:1✗ /notes overflow: the page is 512 px wide in a 390 px window, so it scrolls sideways✗ /notes/:id hierarchy: two buttons of the same weight compete for the eye✗ /settings states: an empty list shows nothing and says nothing✗ Share a note breaks at step 3 at 390 px: no button "Share" on the screen✗ Write a note same-words: the link says "New", the page "Draft", the button "Create" 6 things to redo.Next: three art directions by default, drawn from the product sheet.
/aeom redo
Redo your front end.
When the judge finds it falls short.
- Says what the run will cost, waits for your go, then judges like
/aeom judge. - Draws three art directions from your product's own world, and you pick one.
- Rebuilds every screen and journey on it, keeps only what got better, never adds a feature.
- Writes the standard of your front end, and opens a pull request.
Claude Code · an example notes app
/aeom redoFor 4 screens and 2 key journeys:Total, at most 22 agents ≈ 1.9 M tokens ≈ 1 h 50go6 things to redo. Three directions, drawn from the product: Field notebook · Library card · Index tabsWhich one should the front take?Library cardKit merged on Library card.✓ / kept✓ /notes kept✓ /notes/:id kept✗ /settings sent back: contrast got worse at 390 px✓ Share a note reaches its end, in 3 steps instead of 5Standard written: .aeom/standard.mdPull request opened: #12 The front redone on Library card
/aeom review Soon
Check a pull request.
When before you merge, whoever wrote it.
- Holds the screens it touches to the standard of your front end.
- Flags a colour or a font outside its tokens, a component redone by hand, a different header.
- Checks the journeys it crosses still reach their end.
- Gives a clear verdict, as a GitHub check.
Soon · what it will say
/aeom reviewfeat/export against main, held to .aeom/standard.md: ✗ ExportButton.tsx:14 #7c3aed is not a colour of the standard✗ ExportButton.tsx a button redone by hand: the standard's is ui/button.tsx✗ reports/layout.tsx:8 a header of its own: the standard's is ui/header.tsx✓ Export a report still reaches its end 3 things outside the standard.
/aeom build Soon
Mock up a front end.
When you have a spec or an idea, and no front yet.
- Reads your spec, or a prompt of one line, and drafts the product sheet and its key journeys from it.
- Says what the run will cost, and waits for your go.
- Draws three art directions from your product's own world, and you pick one.
- Mocks up the screens of each key journey on it: front only, on your API or on fake data named as such.
- Replays each journey to its end, writes the standard of your front end, and opens a pull request.
Soon · what it will say
/aeom build --from-scratch "A logbook for a climbing gym: members log their climbs, route setters see what gets climbed."No front yet. Product sheet drafted from your prompt: .aeom/product.md3 key journeys: Log a climb · See my progress · See what gets climbedWhat the run will cost, phase by phase.goThree directions, drawn from the product: Route tape · Chalk board · Grade chartWhich one should the front take?Route tapeKit built on Route tape.✓ Log a climb reaches its end, on fake data✓ See my progress reaches its end, on fake data✓ See what gets climbed reaches its end, on fake dataFake data, named as such, in one module: src/fake/Standard written: .aeom/standard.mdPull request opened: #1 A first front, on Route tape
18 commands underneath.
What /aeom runs on: each one measures, counts or writes, and the judging is left to agents. Run any of them alone, aeom <command> --help says how.
- Understand
productjourneycapturesheet- Judge
checkprinciplesjudgeverdict- Redo
estimatetournamentguardsnapshotcomparerun- Hand over
reportfeaturesstandardpr
Try it.
From the source
git clone https://github.com/The-Vibe-Company/all-eyes-on-mecd all-eyes-on-me && pnpm install && pnpm buildThen /aeom judge in Claude Code for a first look, /aeom redo to redo what fails. The rest is in the README.