A security vulnerability has been discovered in Claude Code's "don't ask again" feature, which inadvertently saves sensitive information like tokens and passwords into public repositories. This occurs when users approve shell commands without further prompting, causing the configuration file `.claude/settings.local.json` to store secrets. While Claude Code attempts to exclude this file from Git, it only applies to the user's global git excludes, not the project's `.gitignore`, leading to potential exposure when the repository is shared. AI
IMPACT This vulnerability highlights a common security risk in AI coding assistants that remember user permissions, potentially exposing sensitive data if not properly managed.
RANK_REASON The item discusses a security vulnerability in a specific AI coding tool and a related open-source tool for auditing such vulnerabilities.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →