让 Agent 发现战略性技术错误

One thing that scares me about coding with AI is that it's hard to feel your strategic mistakes I.e. I STILL remember mistakes from my early career, in ~2018: - using a complex form library instead of browser built-ins - using a downright dumb webpack + PHP combo These mistakes cost us valuable time. And when we rectified them, the tasks around them got faster. But often, we didn't even see them as mistakes! We just thought "this is the way we do it". We stuck to the tactical side of our job, and didn't consider high-level alternatives. Too many things to ship, too much pressure, too little interest in fixing it. I see the same attitude in agents. If you tell them to do a task, they will do that task. They aren't RL'd to think outside the box, or propose strategic alternatives. So the human needs to be on the lookout for these strategic opportunities. But how do you do it, if you don't feel the pain of working in them? I'm planning on building this into my /retro skill (still in-progress). That'll at least give the agent an opportunity to identify refactor opportunities. We'll see how it goes, but I'm still nervous about it.