Show HN：我用 4 天时间，没写一行代码就做出了一个多人游戏
1 分•作者: fcavalcantirj•6 天前
我让 Claude Code 循环运行了 4 天。它自主构建了一个多人策略游戏。我只在最后修改了代码进行润色（约 5%）。
工作原理：
1. 让 Claude 采访我关于游戏设计的问题
2. 从对话中，它生成了大约 248 个结构化的 JSON 任务
3. 创建了一个简单的循环：Claude 选择一个任务，实现、测试、提交，重复
4. 限制为 3 个任务，暂停 15 分钟，以避免耗尽我的 API 配额
脚本：
claude --permission-mode acceptEdits "@specs/prd-v1.json \
1. 阅读 PRD。选择优先级最高的任务。\
2. 实现。运行测试。\
3. 更新进度。提交。推送。\
一次只处理一个任务。"
结果：<a href="https://jarls-production.up.railway.app" rel="nofollow">https://jarls-production.up.railway.app</a>
代码：<a href="https://github.com/fcavalcantirj/jarls" rel="nofollow">https://github.com/fcavalcantirj/jarls</a>
循环概念：<a href="https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents" rel="nofollow">https://www.anthropic.com/engineering/effective-harnesses-fo...</a>
虽然不完美——有错误，界面也不美观。但 95% 是自主完成的。
瓶颈不再是编码。而是前期清晰的思考。
查看原文
I ran Claude Code in a loop for 4 days. It built a multiplayer strategy game autonomously. I only touched the code at the end for polish (~5%).<p>How it worked:<p>1. Asked Claude to interview me about the game design
2. From that conversation, it generated ~248 structured tasks as JSON
3. Created a simple loop: Claude picks a task, implements, tests, commits, repeats
4. Throttled to 3 tasks, 15min pause, to not kill my API quota
The script:
claude --permission-mode acceptEdits "@specs/prd-v1.json \
1. Read PRD. Pick highest-priority task. \
2. Implement. Run tests. \
3. Update progress. Commit. Push. \
ONLY ONE TASK AT A TIME."
Result: <a href="https://jarls-production.up.railway.app" rel="nofollow">https://jarls-production.up.railway.app</a>
Code: <a href="https://github.com/fcavalcantirj/jarls" rel="nofollow">https://github.com/fcavalcantirj/jarls</a>
The loop concept: <a href="https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents" rel="nofollow">https://www.anthropic.com/engineering/effective-harnesses-fo...</a><p>Not perfect — had bugs, ugly screens. But 95% was autonomous.<p>The bottleneck isn't coding anymore. It's thinking clearly upfront.