Backup First AI — AI skill for Claude Code
Safety guardrails for Claude Code: a drop-in CLAUDE.md, example sandbox/permission settings, and a backup-first checklist for adopting AI coding agents responsibly.
How to install Backup First AI
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open jordankrueger/backup-first-ai and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Backup First AI does
Safety guardrails for Claude Code: a drop-in CLAUDE.md, example sandbox/permission settings, and a backup-first checklist for adopting AI coding agents responsibly. Built for non-engineers.
Alternatives in AI
- Openclaw Backup — One-click backup & restore for OpenClaw instances — workspace, credentials, skills, agent history 660 ★
- AI SDK Cpp — The AI Toolkit for Modern C++ 183 ★
- Hazmat — Containment for AI agents - user isolation, sandboxed execution, network controls, backup/rollback 167 ★
README
Claude Code guardrails
[](LICENSE) [](https://code.claude.com/docs) [](CONTRIBUTING.md) [](https://github.com/jordankrueger/backup-first-ai/discussions)
A starter set of safety defaults for [Claude Code](https://code.claude.com/docs), built for people who are not security engineers.
Copy a file, flip on one setting, do a short checklist. It is a strong starting point you can put in place in a few minutes — not a complete shield, and the honest limits are spelled out below.
The one idea
**Everything the AI touches should be backed up in a way it can't delete.**
You cannot fully stop an AI coding agent from making a mistake or being tricked into one. The agent reads files, runs commands, and installs software on its own, and there is no setting that guarantees it always behaves. So do not bet your safety on perfect behavior.
Bet it on this instead: assume the worst day happens, and make that day survivable. If the thing the agent can destroy is also backed up somewhere the agent cannot reach, a disaster becomes an inconvenience.
That single move matters more than every other tip in this repo combined.
Why this exists
Modern AI coding agents are not autocomplete. They open files, run shell commands, install dependencies, and take actions without a human approving each step. That is exactly why they are useful, and also why they occasionally cause real damage. Documented incidents include agents deleting live databases, leaking credentials into logs, installing malicious packages they invented, and following hidden instructions planted in a file.
You do not need to understand any of that deeply to be safe. You need a few sensible defaults
Related Skills
Agent Settings Backup Script
Git-versioned backup tool for AI coding agent config folders (Claude Code, Cursor, Codex, etc.) with size-base
Claude Auto API
Claude Code settings.json auto-config tool to quickly switch API_KEY, AUTH_TOKEN, and model configs across mul
Coding Agent Settings Backup
CLI that backs up, versions, and restores configuration folders for AI coding agents (Claude Code, Codex, Curs
Askproof Skill
Verify AI agent output with actionable evidence for non-engineers. Build better prompts by auditing AI claims
Ceo Info
Live preflight — resolve framework paths + writability, print effective settings, optional cheap model round-t
Nightward
Local-first TUI for auditing AI agent state, MCP config, and dotfiles backup safety
Related Agents
Devroute Researcher
Researches options for a missing development capability identified by DevRoute (for example browser testing or
Agent Engineer
AI Agent Engineer — builds agent orchestration systems, evaluation harnesses, autonomous loops, and safety gua
Light Security ClaudeSettingsAudit
Audit Claude Code config files (settings.json, settings.local.json, .mcp.json) for unsafe or malicious content