Grok Bot 工作坊公开:用仓库约束替代人工审代码
SpaceXAI 公开 Grok Bot 工作坊,核心是用 600+ PR 重构仓库、靠 lint 和 CI 把品味写进约束,让 agent 合并 20 个 PR 也无需逐行审。
This is f*king insane. SpaceXAI just made their whole Grok Bot workshop public and the part everyone skips is the setup: "You spend the tokens to set up a code base so that even the dumbest agents can do a good job" 600+ PRs to refactor Grok Bot into a new architecture Then woke up to 20 PRs merged by her agents and stopped reading the code That only works because the repo enforces the taste: lints for bad patterns and compiler diagnostics, CI red on the same mistake Most people pay for a smarter model. she paid once for a repo where a dumber one cannot screw up Tokens → constraints → trust → no review The setup is the product. the model is just the engine Save this before you babysit agents for another month
这条是引用着另一条写的:x.com/silentguyy66/status/2108957020187468171