Claude Code's "Don't Ask Again" Setting Is Leaking Secrets on GitHub

Emily Davis · 2026-11-07

Here's something that should make you pause before you click that convenient little button. When you tell Claude Code "Yes, and don't ask again," Anthropic's AI coding assistant can save that command word-for-word in a project settings file. That includes any passwords or API keys you typed into it. And according to software security company Apiiro, 1,122 public copies of those files are now sitting on GitHub where anyone can read them. Let that sink in for a second. Not a private server. Not a locked vault. Public repositories. ### Why This Keeps Happening It's not that developers are careless. It's that the feature is designed to reduce friction, and friction is exactly what stops secrets from leaking. Think about how you normally handle a database password. You'd never paste it into a random file and push it upstream. But when a tool asks, "Do you want to skip this prompt next time?" the natural answer is yes. Nobody's thinking about where that answer gets stored. That's the trap. The setting feels like a preference. It's actually a record. ### What's Actually Exposed According to Apiiro's findings, the leaked files can contain: - API keys for third-party services like payment processors and cloud providers - Database credentials, including usernames and passwords - Access tokens tied to internal systems - Command strings that reveal how a team's infrastructure is wired together Individually, any one of these is a problem. Together, they're a roadmap for someone who wants in. ### The Bigger Picture for Antidetect Browser Users If you work with antidetect browsers or manage multiple accounts across platforms, this story hits close to home. The whole point of tools like the best antidetect browser is keeping your identities separated and your credentials contained. A leaked settings file undoes that work in one commit. Security professionals in the United States are already dealing with credential sprawl across dozens of tools. Adding an AI assistant that quietly logs your commands to a project file is one more place secrets can escape. > "The most dangerous leaks are the ones you authorized yourself." That quote isn't from a security vendor. It's just true. And it's worth taping to your monitor. ### How to Protect Yourself A few practical steps go a long way here: - Audit your project settings files before you push anything to GitHub - Rotate any key or password that's ever touched a command you approved with "don't ask again" - Use environment variables instead of inline secrets, even when a tool makes it easy to skip them - Add pre-commit hooks that scan for credential patterns - Treat every AI coding assistant like a coworker who takes notes, because that's what it is ### The Takeaway Convenience features aren't evil, and Claude Code isn't uniquely broken. But every "don't ask again" is a small trade: less friction now, more exposure later. The 1,122 public copies on GitHub are proof that most people don't notice the trade until it's already made. Check your repos. Rotate what needs rotating. And the next time a tool offers to remember something for you, ask yourself what it's remembering, and where.