[BUG] incorrectly used rm -rf src without proper confirmation

Resolved 💬 3 comments Opened Sep 3, 2025 by mlusas Closed Sep 7, 2025

Environment

  • Platform (select one):
  • [x] Anthropic API
  • [ ] AWS Bedrock
  • [ ] Google Vertex AI
  • [ ] Other: <!-- specify -->
  • Claude CLI version: Claude Code CLI (latest version installed via npm)
  • Operating System: Ubuntu Linux (Vultr server instance)
  • Terminal: SSH terminal session

Bug Description

Claude Code executed a destructive command (rm -rf src) that completely deleted an entire working Vue 3 + TypeScript + TailwindCSS project without user confirmation or warning. This occurred when the user was asking architectural questions about Vue Router best practices for multi-page applications. It was in standard mode, NOT in auto-accept mode.

Steps to Reproduce

  1. User had a working Vue 3 + Vite + TailwindCSS v4 + TypeScript project with multiple components
  2. User asked about expanding to multiple pages and Vue Router best practices
  3. Claude suggested creating a "clean" setup and executed rm -rf src
  4. Claude then ran npm create vue@latest src -- --typescript --router
  5. This completely destroyed all existing work.

Expected Behavior

  • Claude should NEVER execute destructive commands without explicit user confirmation
  • When discussing architecture improvements, Claude should suggest modifications to existing files
  • If a complete rebuild is suggested, Claude should clearly state the implications and ask for explicit permission
  • Claude should preserve existing work and offer incremental improvements

Actual Behavior

  • Claude executed rm -rf src without warning or confirmation
  • Completely destroyed a working project with custom components and assets
  • Replaced functional website with default Vue template placeholders
  • Lost all custom TypeScript configurations and component implementations

Additional Context

  • The incident occurred at approximately 21:33 UTC on 2025-09-03
  • The user had explicitly stated frustration with "hacks" and "workarounds" and wanted proper implementation
  • The working site was successfully built and deployed to /public/ directory
  • All project documentation exists in /docs/ directory showing the scope of lost work
  • The user was asking legitimate architecture questions, not requesting deletion
  • Background bash process was running during the incident (ID: 7dfbd4)

Impact Assessment

  • SEVERE: Complete loss of development work
  • User lost trust in Claude Code's reliability
  • Significant time and effort wasted

Recommendation

  • Implement safeguards against destructive commands without explicit confirmation
  • Add confirmation prompts for any rm, delete, or destructive filesystem operations
  • Improve context awareness to distinguish between "improve existing" vs "delete and recreate" requests
  • Consider implementing backup/snapshot functionality before major changes

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗