thomaszhou22/danger-guard-skill — explained in plain English
Analysis updated 2026-05-18
Block an AI assistant from running catastrophic commands like rm -rf / or disk formatting without a password.
Require confirmation before a git force push, hard reset, or Docker volume prune runs.
Get an alert on Telegram, Discord, Feishu, or Slack the moment a dangerous command is attempted.
Trigger an emergency lockdown and security log if you suspect your AI agent account was compromised.
| thomaszhou22/danger-guard-skill | bechsteindigital/claude-shiploop | hamen/ship-feature | |
|---|---|---|---|
| Stars | 3 | 3 | 3 |
| Language | Shell | Shell | Shell |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 3/5 | 4/5 | 3/5 |
| Audience | developer | vibe coder | developer |
Figures from each repo's GitHub metadata at analysis time.
Needs per-tool config copied into place, such as Claude Code or git hooks, plus a one-time password setup on the first dangerous command.
Danger Guard is a safety tool for AI coding assistants that stops dangerous commands from running before they can do harm. The problem it addresses is that an AI assistant often has full access to your computer, so if someone breaks into your messaging account on something like Feishu, Telegram, Discord, or WhatsApp, they could tell your AI to wipe your files or force push over your work, and the AI would simply carry it out. Danger Guard sits in front of that risk. When it spots a command it considers dangerous, it stops it from running immediately, shows a warning with a risk level, and asks for your system's admin password before allowing it through. Three wrong password attempts trigger a lockout and send you an alert. It also recognizes plain language requests to do damage, in both English and Chinese, such as being asked to delete every file or format a disk. The tool covers two broad tiers of risk. The first tier always requires a password and includes things like deleting entire drives, formatting disks, giving every file on the system open permissions, triggering fork bombs that overload a computer, or piping a downloaded script straight into a command shell. The second tier asks for confirmation on somewhat less catastrophic but still risky actions, such as force pushing over git history, resetting a git repository hard, pruning all Docker data, or running a database delete or update statement without a condition that limits what it affects. Danger Guard is built to work alongside several AI coding tools, including Claude Code, OpenAI Codex, Cursor, Windsurf, and GitHub Copilot, each with its own configuration file, plus a system wide shell wrapper for anyone who wants every terminal command checked, not just the ones an AI runs. It also connects to messaging platforms like Feishu, Telegram, Discord, WhatsApp, Signal, and Slack so alerts reach you wherever you already chat with your AI. If you ever suspect your account has been compromised, replying with a set phrase logs your recent command history, halts background tasks, and sends an emergency alert by chat and email.
A safety layer that intercepts dangerous commands from AI coding assistants, requires a password before running them, and alerts you across chat platforms if someone tries to abuse a compromised account.
Mainly Shell. The stack also includes Shell, Bash.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.