Guides to running AI coding agents safely

How the safety features of Claude Code, Codex, Cursor and Gemini CLI really work, where they stop, and what to do about it. Written by the maker of Rollcage, and useful whether or not you use it.

• Claude Code security

  Is Claude Code safe? What its permissions, sandbox and hooks protect, the gaps they leave, and a checklist to close them.

  Read more →

• Claude Code hooks

  How hooks work, with PreToolUse examples that block .env reads, environment dumps and destructive commands.

  Read more →

• Claude Code permissions

  Permission modes, allow, ask and deny rules in settings.json, how to protect .env files, and where rules stop.

  Read more →

• Claude Code YOLO mode

  What --dangerously-skip-permissions does, what can go wrong, the equivalents in other agents, and how to run it with a safety net.

  Read more →

• Claude Code sandbox

  The built-in /sandbox, Docker and dev containers, how Codex compares, and what a sandbox still doesn't protect.

  Read more →

• Undo Claude Code changes

  /rewind and checkpoints, what they miss, and how to recover files deleted by rm or git reset --hard on a Mac.

  Read more →

• Claude Code usage and cost

  /usage, /cost and /context, the transcripts behind them, and how to see tokens and cost per day, project and model.

  Read more →

• Codex usage

  /status in the CLI, the usage page in your ChatGPT settings, and the local session files that record every token.

  Read more →

• Cursor security

  How Cursor hooks, auto-run (YOLO mode), .cursorignore and privacy mode work, and what each leaves open.

  Read more →

• Vibe coding security

  The real risks of vibe coding, in the code you ship and on your own machine, with a practical checklist.

  Read more →

• AI coding agent security

  The threat model, five layers of guardrails, and how Claude Code, Codex, Cursor and Gemini CLI compare.

  Read more →