Open World Explorer — a Breath of the Wild-grade sandbox in one HTML file, where the grass grew to 152% of the player's height and nobody noticed until it was measured
A single-file open world benchmarked against Breath of the Wild: procedurally generated terrain with erosion, a climb-anything traversal system, a paraglider that rides real thermals, fire that propagates through vegetation and generates the updraft you glide on, physics-driven emergent puzzle shrines, day/night and weather, and a stamina economy that ties it together. The critic measured the grass against the player capsule and found blades rendering at 44 to 152 percent of player height — a field of waist-to-overhead slabs. Fixed to 11-23%. Then the terrain overcorrected into spiky cones and had to be retuned again.
What this is
An open world in a single HTML file. The terrain is procedurally generated with ridged noise, hydraulic-style erosion and terracing, so valleys have floors and ridgelines have shoulders. You can climb essentially anything, governed by a stamina economy that makes route selection a real decision. You have a paraglider.
The part worth caring about is that the systems compose. Fire spreads through connected vegetation. Burning vegetation produces a thermal updraft. That updraft gives your paraglider lift. None of those three things knows about the puzzle design — but together they solve shrines. There is a day/night cycle, weather, and temperature that feeds back into the stamina model.
Why this is mind-blowing
The best catch in this build is almost funny: the grass was taller than the player.
Every screenshot looked lush. A dense green field reads as "good grass" instantly, and every reviewer — human or agent — nodded past it. So the critic stopped looking and measured blade height against the player capsule instead.
Grass was rendering at 44% to 152% of player height. The character was wading through a field of waist-high to overhead green slabs, and it looked fine, because "lush" and "correct" are different properties and only one of them is visible at a glance. Retuned to 11-23%, which is what grass actually looks like from eye level.
Then the terrain fix broke the terrain, which is exactly why you loop rather than review. The initial pass corrected the world for being too flat — and overcorrected it into spiky cones. Technically more dramatic. Visually ridiculous. A second critique retuned the ridge exponent from 2.0 to 1.65 and added terracing and flat valley floors. A single review pass would have confidently shipped either the pancake or the spikes. Only a loop that re-critiques after every fix converges on the version in between.
And the systems chain genuinely chains. Verified end to end in a real browser, not asserted: set fire to vegetation, watch it propagate through connected plants, watch the burn generate a thermal, and then ride it — the HUD reading "Glide — THERMAL +13.6m/s" mid-ascent. A shrine was then solved by that same mechanic with the log confirming "Shrine solved! Fire climbed the trellis."
Nothing scripted that. The fire system doesn't know what a shrine is. The thermal doesn't know what a paraglider is. They just each behave correctly and the puzzle falls out of the overlap — which is the entire trick Breath of the Wild is built on, and the only part of it worth copying.
Prompt
I want you to build an open world exploration game at the level of The
Legend of Zelda: Breath of the Wild. It should be utterly perfect, visually
beautiful, with every single thing done at AAA quality — from the terrain
generation to the traversal to the systemic interactions to anything you
could think of.
Procedurally generated terrain with real erosion, not just noise. Climb
anything, with a stamina economy that makes route choice matter. A
paraglider. And most importantly: SYSTEMS THAT COMPOSE. Fire should spread
through vegetation. Burning vegetation should create a thermal updraft. That
updraft should give you lift on the paraglider. Puzzles should be solvable
by combining those systems in ways I never explicitly scripted.
Day/night cycle, weather, temperature. Shrines that are physics puzzles.
Fan out sub-agents and have sub-agents tackle each one individually so that
the world is utterly perfect. You should /loop on each item and have a
separate sub-agent check it VISUALLY in a real browser. That separate
sub-agent should be a really harsh critic, and if it doesn't look and feel
triple A, it should keep going.
Do not let the critic judge scale by eye. Make it MEASURE things against the
player character — how tall is the grass relative to the player capsule, how
steep is the terrain, how far is the draw distance. "Looks lush" is not an
acceptable report. And make it actually play the systems chain end to end:
light a fire, confirm it propagates, confirm the thermal forms, confirm the
paraglider gains altitude, and quote the numbers off the HUD.
Don't stop until each sub-agent is utterly wowed when compared with actual
Breath of the Wild. It should literally compare them side by side blind and
say which one looks better. ThreeJS, ONE self-contained HTML file. /loop
until it's utterly perfect. Fan out sub-agents and ultracode.Paste this into Claude, Cursor, or Copilot. Change one thing that matters to you.
What I learned shipping it
- How to catch scale bugs that are invisible to judgement — measure world objects as a ratio of player height rather than asking whether a screenshot 'looks right', which is how grass at 152% of player height ships unnoticed
- Why a single review pass ships either the pancake or the spikes: correcting flat terrain overshot into cones, and only re-critiquing after each fix converged on ridge exponent 1.65 with terracing and flat valley floors
- How to build emergent puzzles from composed physics — fire propagation feeding a thermal updraft feeding paraglider lift — so shrines are solved by system interaction rather than scripted triggers