[
  {
    "id": "mythos-craft",
    "index": "01",
    "kind": "html",
    "title": {
      "zh": "神话境界",
      "en": "Mythos Craft"
    },
    "tagline": {
      "zh": "单个 HTML 文件，一次生成一款《我的世界》式的可玩 3D 体素世界。",
      "en": "A playable Minecraft-style 3D voxel world, generated as a single HTML file in one shot."
    },
    "prompt": {
      "zh": "你是一个顶级的资深游戏开发者。请为我使用单 HTML5 和 Canvas（或 Three.js）编写一个《我的世界》克隆版。请包含以下核心功能：\n\n1. 一个基于柏林噪声（Perlin Noise）生成的无限或可滚动的 3D 地形。\n2. 第一人称视角，支持使用 WASD 移动和鼠标环视。\n3. 支持至少 3 种方块类型（如：泥土、草地、石头等），并支持鼠标左键破坏方块，右键放置方块。材质要尽可能还原。\n4. 要有一个内置的、极其丰富的神话幻想中的建筑和地形等，由你自行设计。",
      "en": "You are a top-tier veteran game developer. Build me a Minecraft clone using a single HTML5 file and Canvas (or Three.js). Include the following core features:\n\n1. An infinite or scrollable 3D terrain generated from Perlin noise.\n2. A first-person view with WASD movement and mouse look.\n3. At least 3 block types (e.g. dirt, grass, stone), with left-click to break and right-click to place blocks. Keep the textures as faithful as possible.\n4. A built-in, extremely rich set of mythic-fantasy buildings and terrain, designed at your own discretion."
    },
    "tags": [
      "HTML",
      "3D",
      "GAME",
      "ONE-SHOT"
    ]
  },
  {
    "id": "pelican-cycling",
    "index": "02",
    "kind": "svg",
    "title": {
      "zh": "鹈鹕海边骑行",
      "en": "Pelican by the Sea"
    },
    "tagline": {
      "zh": "经典 SVG 测评题：让模型用代码「画」出一只骑自行车的鹈鹕。",
      "en": "The classic SVG eval: make the model \"draw\" a pelican riding a bicycle, in code."
    },
    "prompt": {
      "zh": "制作一个逼真的、生动的、在海边骑着自行车的鹈鹕 <svg>",
      "en": "Create a realistic, lively <svg> of a pelican riding a bicycle by the sea."
    },
    "tags": [
      "SVG",
      "ILLUSTRATION",
      "ONE-SHOT"
    ]
  },
  {
    "id": "skeleton-watch",
    "index": "03",
    "kind": "html",
    "title": {
      "zh": "镂空机械表",
      "en": "Skeleton Watch"
    },
    "tagline": {
      "zh": "单个 HTML 文件，一次生成一只可旋转、可爆炸视图的 3D 镂空机械腕表，机芯全可见。",
      "en": "A single HTML file: a rotatable, exploded-view 3D skeleton mechanical watch with a fully visible movement, generated in one shot."
    },
    "prompt": {
      "zh": "创建一个精美、高细节的交互式 3D 镂空机械表。\n\n核心要求：\n• 完全可见的复杂机芯：包括主发条、齿轮系、带游丝的摆轮、擒纵机构、夹板、红宝石轴承等。\n• 真实的金属材质、蓝宝石水晶玻璃、反射、阴影，以及奢华腕表摄影级别的照明效果。\n• 荧光指针与刻度。\n• 交互式控制：支持旋转手表、切换爆炸视图、悬停/点击时高亮显示的零件标签、机芯信息面板。\n• 流畅的动画，达到专业级产品展示质量。\n• 深色优雅背景，带有细微的反射效果。\n\n在一次回复中尽可能真实且令人惊叹地完成。输出为一个完整的、自包含的单 HTML 文件，使用 Three.js (CDN)，所有代码均采用内联形式。",
      "en": "Create an exquisite, highly detailed, interactive 3D skeleton mechanical watch.\n\nCore requirements:\n• A fully visible, complex movement: mainspring, gear train, balance wheel with hairspring, escapement, bridges, ruby bearings, and so on.\n• Realistic metal materials, sapphire crystal glass, reflections, shadows, and luxury-watch photography-grade lighting.\n• Luminous hands and markers.\n• Interactive controls: rotate the watch, toggle an exploded view, part labels highlighted on hover/click, and a movement info panel.\n• Smooth animation at professional product-showcase quality.\n• A dark, elegant background with subtle reflections.\n\nMake it as realistic and stunning as possible in a single response. Output a complete, self-contained single HTML file using Three.js (CDN), with all code inline."
    },
    "tags": [
      "HTML",
      "3D",
      "THREE.JS",
      "ONE-SHOT"
    ]
  },
  {
    "id": "cs-dust2",
    "index": "04",
    "kind": "react",
    "title": {
      "zh": "CS Dust2 5v5",
      "en": "CS Dust2 5v5"
    },
    "tagline": {
      "zh": "React + three.js 从零手写的 3D 第一人称射击原型，复刻 Dust2 的 5v5 对战。",
      "en": "A from-scratch 3D FPS in React + three.js: a 5v5 battle on a rebuilt Dust2."
    },
    "prompt": {
      "zh": "用 **React + TypeScript + three.js** 从零开发一个可在浏览器直接运行的 3D 第一人称射击原型：复刻 Dust2 核心区域的 5v5 对战。你可以自行搭建项目、安装依赖、运行调试并迭代，直到它能正常游玩。\n\n## 交付形式\n\n- 一个标准 **Vite + React 18 + TypeScript** 项目，含完整 `package.json`、`vite.config.ts`、`index.html`、`src/`，`npm install && npm run dev` 即可打开游玩。\n- **不使用任何外部资产文件**（3D 模型、贴图、音频一律不用）。地图、角色、武器全部在代码里用几何体 procedural 生成，声音用 Web Audio 程序化合成。第三方库只用 three / react / react-dom；寻路、状态机、物理、渲染循环都自己实现，不引入现成游戏引擎。\n\n## 1. Dust2 地图\n\n在 3D 世界里搭出空间尺度合理的 Dust2 经典区域，**必须完整包含**：T 出生点、CT 出生点、A大、A点、中门（含可穿过的中门门体）、猫道、B洞、B点，并保证这些区域按 Dust2 的实际走位彼此连通。要么全部包含，不要交出缺区域的残缺版本。所有墙体、箱子都有物理碰撞，玩家和 AI 都无法穿墙。\n\n## 2. 角色\n\n在没有外部模型的前提下，用代码搭出**类人型角色**：必须有独立的头、躯干、双臂、双腿——**严禁用胶囊体或圆柱体等基础几何体敷衍成一根柱子**。CT 与 T 两个阵营在外观上要能一眼区分，角色默认是持枪姿态，武器随角色朝向移动。\n\n## 3. 武器系统\n\n做一套可扩展的武器数据系统，支持在主武器 / 副武器 / 近战之间切换，并有对应的第一人称手持模型。至少还原以下手感差异：\n\n- **AK-47**：高伤害，开火后坐力剧烈（准星扩散 + 视角上扬明显）。\n- **M4A4**：中等伤害，后坐力低，射速高。\n- **AWP**：单发致命，射速极低，**右键开镜**（3D 视野拉近 + 2D 瞄准镜遮罩覆盖）。\n- **手枪**：至少 Glock（T 默认）、USP（CT 默认）、沙鹰，用于手枪局；伤害低于步枪、沙鹰介于两者之间且弹匣小。\n- **近战刀**。\n\n命中判定按 **Hitbox 分区**（头 / 胸 / 腹 / 手臂 / 腿），伤害各不相同，**爆头伤害是身体的两倍**。有护甲时按护甲减伤。\n\n## 4. 手枪局\n\n支持\"手枪局\"开局：全员只带默认手枪、没有主武器、护甲按手枪局配置。这是标准 CS 的第一回合形态。\n\n## 5. C4 炸弹与回合胜负\n\n- 回合开始时 C4 交给某个 T；持包者阵亡后炸弹掉落，可被其他 T 捡起。\n- **下包**：持包 T 在 A点或 B点范围内可安放 C4，安放后进入倒计时。\n- **拆包**：CT 站在已安放的 C4 处，持续一段时间完成拆除。\n- **回合结算**：T 清空 CT、或 C4 爆炸 → T 胜；CT 清空 T（且未下包）、或成功拆包 → CT 胜。\n- 玩家操控的角色阵亡后，可切换到存活队友的视角 / 接管其操控（bot）。\n\n## 6. 5v5 AI 战场\n\n场上共 10 人：1 名玩家 + 4 名 AI 队友，对抗 5 名 AI 敌人，分属 CT / T 两阵营。AI 需要：能在地图内自主寻路巡逻、不卡墙；有视野索敌判定（被墙挡住看不见）；遭遇敌方阵营自动开火交战；并参与 C4 的下包 / 拆包流程（进攻方推包安放、防守方前往拆除）。\n\n## 7. 操作与 HUD\n\n- 流畅的第一人称控制：WASD 移动、空格跳跃、指针锁定鼠标转视角。\n- 现代 2D UI 覆盖层：十字准星、生命值 / 护甲、动态弹药面板（开火递减、换弹恢复）、右上角实时击杀播报（Killfeed）。\n- **小地图（minimap）**：常驻显示 Dust2 俯视轮廓，实时标出自己、队友、被看见的敌人的点位，以及 C4 的位置。\n\n## 8. 声音\n\n游戏要有音效反馈。由于不使用外部音频文件，用 **Web Audio API 程序化合成** 声音即可。至少包含：不同武器的开火声（AK / M4 / AWP / 手枪要有区分）、换弹声、脚步声、AWP 开镜声、命中反馈音、C4 安放 / 拆除 / 爆炸声，以及击杀提示音。\n\n## 架构要求\n\n这是工业级挑战。你要独立完成技术选型、架构设计、数据结构封装，以及渲染循环、物理帧更新、React 状态生命周期三者之间的数据同步。全局状态引擎、武器系统、AI 状态机、3D 渲染与物理组件等关键系统都要真正实现，而非留空占位。",
      "en": "Build a browser-playable 3D first-person shooter prototype from scratch using **React + TypeScript + three.js**: a 5v5 battle recreating the core areas of Dust2. You may set up the project, install dependencies, run, debug and iterate until it is properly playable.\n\n## Deliverable\n\n- A standard **Vite + React 18 + TypeScript** project with a complete `package.json`, `vite.config.ts`, `index.html` and `src/`; `npm install && npm run dev` opens it to play.\n- **No external asset files whatsoever** (no 3D models, textures or audio). The map, characters and weapons are all generated procedurally from geometry in code, and sound is synthesized programmatically with Web Audio. The only third-party libraries are three / react / react-dom; pathfinding, state machines, physics and the render loop are all implemented yourself, with no off-the-shelf game engine.\n\n## 1. Dust2 map\n\nBuild the classic Dust2 areas at a sensible spatial scale in the 3D world, and you **must fully include**: T spawn, CT spawn, Long A, A site, Mid (with a passable set of Mid Doors), Catwalk, B tunnels and B site, connected to one another along Dust2's real routes. Include them all — do not submit a partial version missing areas. All walls and boxes have physical collision; neither players nor AI can pass through walls.\n\n## 2. Characters\n\nWith no external models, build **humanoid characters** in code: each must have a distinct head, torso, both arms and both legs — **do NOT cheat with a capsule or cylinder as a single column body**. The CT and T teams must be distinguishable at a glance; characters default to a weapon-holding pose and the weapon moves with the character's facing.\n\n## 3. Weapon system\n\nBuild an extensible weapon data system supporting switching between primary / secondary / melee, each with its own first-person held model. Reproduce at least these handling differences:\n\n- **AK-47**: high damage, heavy recoil after firing (pronounced crosshair spread + upward view kick).\n- **M4A4**: medium damage, low recoil, high fire rate.\n- **AWP**: one-shot lethal, very low fire rate, **right-click to scope** (3D zoom-in + a 2D scope overlay).\n- **Pistols**: at least Glock (T default), USP (CT default) and Deagle for pistol rounds; lower damage than rifles, Deagle in between with a small magazine.\n- **A melee knife**.\n\nHit detection uses **hitbox zones** (head / chest / stomach / arm / leg) with different damage each, and **headshot damage is twice body damage**. Armor reduces damage when present.\n\n## 4. Pistol round\n\nSupport a \"pistol round\" opening: everyone carries only their default pistol, no primary weapon, armor per pistol-round rules. This is the standard CS first round.\n\n## 5. C4 bomb and round win/loss\n\n- At round start the C4 is given to a T; when the carrier dies the bomb drops and can be picked up by another T.\n- **Plant**: a T carrying the bomb can plant the C4 within the A or B site, starting a countdown after planting.\n- **Defuse**: a CT standing at a planted C4 defuses it over a set time.\n- **Round resolution**: T wins by eliminating all CTs or the C4 exploding; CT wins by eliminating all Ts (with no plant) or by successfully defusing.\n- After your controlled character dies, you can switch to a living teammate's view / take over that bot.\n\n## 6. 5v5 AI battlefield\n\nTen players on the field: 1 human player + 4 AI teammates against 5 AI enemies, split into CT / T teams. The AI must: navigate and patrol the map autonomously without getting stuck on walls; have line-of-sight target detection (blocked by walls); automatically open fire on the enemy team on contact; and take part in the C4 plant / defuse flow (attackers push and plant, defenders go to defuse).\n\n## 7. Controls and HUD\n\n- Smooth first-person controls: WASD movement, space to jump, pointer-locked mouse to look.\n- A modern 2D UI overlay: crosshair, health / armor, a dynamic ammo panel (decrements on fire, restored on reload), and a live kill feed in the top-right.\n- A **minimap**: a persistent top-down Dust2 outline marking your own, teammates' and visible enemies' positions in real time, plus the C4's location.\n\n## 8. Sound\n\nThe game needs sound feedback. Since no external audio files are used, synthesize sound programmatically with the **Web Audio API**. Include at least: distinct firing sounds per weapon (AK / M4 / AWP / pistols differentiated), reload, footsteps, AWP scope-in, hit feedback, C4 plant / defuse / explosion, and a kill sound.\n\n## Architecture\n\nThis is an industrial-grade challenge. You own the technology choices, architecture, and data-structure design, plus the data sync between the render loop, the physics frame update, and the React state lifecycle. The global state engine, weapon system, AI state machine, and 3D-rendering and physics components must be genuinely implemented, not left as placeholders."
    },
    "tags": [
      "REACT",
      "3D",
      "THREE.JS",
      "GAME"
    ]
  },
  {
    "id": "turf-war",
    "index": "05",
    "kind": "agent",
    "title": {
      "zh": "圈地对决",
      "en": "Turf War"
    },
    "tagline": {
      "zh": "每个模型一次性写出一个圈地智能体,在同一张网格战场上真刀真枪对战——排名不靠投票,靠客观胜负。",
      "en": "Each model writes a territory-grabbing agent in one shot; agents battle head-to-head on the same grid — ranked by objective wins, not votes."
    },
    "prompt": {
      "zh": "你要为一个 1v1 的「圈地对决」(Turf War) 环境编写一个 JavaScript 智能体。这是一个双人零和、双方同时行动的博弈。你的目标只有一个:在与多个未知对手、多张随机地图的循环赛中拿到尽可能高的胜率。对手由同等水平的 AI 模型编写——不要依赖「对手会犯错」的脆弱套路,任何固定模式都可能被针对;但也不要为求稳放弃进攻,平局只得 0.5 分,一个只求不败的智能体拿不到高排名。这是你唯一一次提交:没有反馈、没有试跑、没有重试,成绩公开计入排行榜。请输出你能写出的最强智能体。\n\n【环境规则】\n\n- 棋盘为 40×40 网格,坐标 (x, y),x 向右,y 向下,范围 0–39。每局按种子生成若干 180° 中心对称的岩石格(不可通行),出生点附近不会有岩石。\n- 双方从对角出生点出发,初始拥有出生点周围 3×3 领地。每 tick 双方同时移动一格,方向为 up / down / left / right 之一;全场共 1200 tick。\n- 在自己领地外移动会留下尾巴。回到自己领地的瞬间,你的尾巴以及被「尾巴 + 领地」围住的区域全部变成你的领地(包括对方领地)。围住判定:从棋盘边界向内 flood fill(岩石可穿过但不可被占有),无法到达的格子归你。\n- 任何人(包括你自己)踩到某玩家的尾巴,该尾巴的主人立即死亡。双方同 tick 进入同一格或互相穿过,双方同时死亡。撞岩石或出界,自己死亡。击杀判定基于本 tick 开始前的尾巴状态,双方完全对等。\n- 死亡后尾巴消失、领地保留;3 tick 后在出生点重生并重新获得周围 3×3 领地(覆盖任何归属),重生后 10 tick 无敌(不会死亡也不能击杀)。\n- 同 tick 结算顺序固定:移动 → 死亡判定 → 圈地 → 重生计时。1200 tick 结束时领地格数多者获胜,相等为平局。\n\n【接口】你必须输出一个 JS 文件,顶层定义:\n\n```js\nfunction createAgent(config) {\n  // config = { size: 40, totalTicks: 1200 }\n  // 可以在闭包里保存任意记忆(允许并鼓励在对局中在线建模对手)\n  return {\n    act(obs) {\n      return \"up\"; // \"up\" | \"down\" | \"left\" | \"right\"\n    }\n  };\n}\n```\n\nobs 每 tick 传入,字段:\n\n- obs.tick:当前 tick,从 0 起\n- obs.grid[y][x]:0 空地 · 1 我的领地 · 2 我的尾巴 · 3 对方领地 · 4 对方尾巴 · 5 岩石。grid 数组每 tick 会被复用,不要跨 tick 持有引用,需要历史请自行拷贝\n- obs.me / obs.opp:{ x, y, dir, alive, respawnIn, invincibleFor }\n- obs.scores:{ me, opp },双方当前领地格数\n- obs.rng():[0,1) 确定性随机数;需要随机时只准用它\n\n另外,请在返回的对象里附带 name 字段(与 act 并列):给你的算法起一个不超过 5 个字、有意思的名字,例如 { name: \"...\", act(obs) {...} }。\n\n视角是自我中心的:无论执哪一边,你都是 me。信息完全公开,没有战争迷雾。非法动作、抛异常或超出算力预算,该 tick 视为沿原方向直行。\n\n【运行时约束】\n\n- 纯 JavaScript(ES2020),单文件;禁止 import/require、网络、存储、Math.random、Date(运行环境会封禁它们)。\n- 算力限制是确定性的:你的代码会被自动插桩计数,每 tick 预算为 200,000 次「循环迭代 + 函数调用」。参考:一次 40×40 全图 BFS 约 2,000–6,000 次。超出预算该 tick 视为直行。\n- 完成本任务只依靠你自身的知识与推理:禁止联网搜索、禁止查阅外部文档或代码库、禁止调用任何 skill/插件/外部工具、禁止试运行代码。\n- 输出前静态自查:合法 ES2020、顶层定义了 createAgent、无 import / Math.random / Date。\n\n【评测】你的智能体将与其他 AI 模型编写的智能体进行多局循环赛,每局地形随机、先后手轮换,按胜负用 Bradley-Terry 排名。请深思熟虑后一次性输出唯一、完整、可直接运行的 JS 文件,不要输出任何解释文字。",
      "en": "Write a JavaScript agent for a 1v1 \"Turf War\" environment. This is a two-player, zero-sum, simultaneous-move game. Your sole objective is to maximize your win rate in a round-robin against multiple unknown opponents across randomized maps. Opponents are written by equally capable AI models — do not rely on brittle patterns that assume opponent mistakes, as any fixed pattern can be exploited; but do not trade wins for safety either: a draw scores only 0.5, and an agent that merely avoids losing will not rank. This is your only submission: no feedback, no test runs, no retries, and the result is public on a leaderboard. Output the strongest agent you can write.\n\n[Environment]\n\n- The board is a 40×40 grid, coordinates (x, y), x rightward, y downward, range 0–39. Each match seeds a set of 180°-centrally-symmetric rock cells (impassable); no rocks appear near the spawns.\n- Both players start from opposite corners owning a 3×3 territory around their spawn. Each tick both players move one cell simultaneously (up / down / left / right); a match lasts 1200 ticks.\n- Moving outside your own territory leaves a trail. The moment you re-enter your own territory, your trail plus every region enclosed by trail + territory becomes yours (including the opponent's territory). Enclosure test: flood fill inward from the board border (rocks are transparent to the flood but can never be owned); unreachable cells are captured.\n- If anyone (including yourself) steps on a player's trail, that trail's owner dies immediately. Entering the same cell on the same tick, or swapping cells, kills both players. Hitting a rock or leaving the board kills you. Kill checks use the trail state from the start of the tick, so both sides are perfectly symmetric.\n- On death your trail vanishes but your territory is kept; you respawn at your spawn point after 3 ticks with a fresh 3×3 territory (overwriting any ownership) and 10 ticks of invincibility (you cannot die or kill).\n- Per-tick resolution order is fixed: moves → deaths → capture → respawn countdown. After 1200 ticks the larger territory wins; equal is a draw.\n\n[Interface] You must output one JS file that defines at top level:\n\n```js\nfunction createAgent(config) {\n  // config = { size: 40, totalTicks: 1200 }\n  // keep any memory in the closure (in-match opponent modeling is allowed and encouraged)\n  return {\n    act(obs) {\n      return \"up\"; // \"up\" | \"down\" | \"left\" | \"right\"\n    }\n  };\n}\n```\n\nobs is passed every tick:\n\n- obs.tick: current tick, starting at 0\n- obs.grid[y][x]: 0 empty · 1 my territory · 2 my trail · 3 opponent territory · 4 opponent trail · 5 rock. Grid arrays are reused across ticks — do not hold references across ticks; copy if you need history\n- obs.me / obs.opp: { x, y, dir, alive, respawnIn, invincibleFor }\n- obs.scores: { me, opp }, current territory cell counts\n- obs.rng(): deterministic random in [0,1); it is the only allowed source of randomness\n\nAlso include a name field on the returned object (next to act): give your algorithm a fun, memorable name of at most one short word (~10 characters), e.g. { name: \"...\", act(obs) {...} }.\n\nThe view is ego-centric: whichever side you play, you are me. Information is fully public — no fog of war. An invalid action, a thrown exception, or exceeding the compute budget makes you continue straight for that tick.\n\n[Runtime]\n\n- Pure JavaScript (ES2020), single file; import/require, network, storage, Math.random and Date are forbidden (the runtime blocks them).\n- The compute limit is deterministic: your code is automatically instrumented, with a per-tick budget of 200,000 loop iterations + function calls. Reference: one full-board 40×40 BFS costs about 2,000–6,000. Exceeding the budget means you walk straight that tick.\n- Rely solely on your own knowledge and reasoning: no web search, no consulting external docs or codebases, no skills/plugins/external tools, no test-running code.\n- Static self-check before you output: valid ES2020, createAgent defined at top level, no import / Math.random / Date.\n\n[Evaluation] Your agent plays a multi-game round-robin against agents written by other AI models, with randomized terrain and alternating sides, ranked by Bradley-Terry on match outcomes. Think hard, then output a single, complete, directly runnable JS file — no explanations."
    },
    "tags": [
      "JS",
      "AGENT",
      "1V1",
      "ONE-SHOT"
    ],
    "credit": {
      "text": "Inspired by Jiayi Weng, Learning Beyond Gradients",
      "url": "https://trinkle23897.github.io/learning-beyond-gradients/"
    }
  }
]