Agent Safety & Hooks Pack

Stop coding agents leaking secrets, deleting things or burning your budget.

Zero-dependency Node.js hooks for Claude Code and Cursor. They hard-block destructive commands, ask before risky ones (push, deploy, publish, sudo, deletes), stop agents writing secrets into files or touching .env and key files, cap tool calls and tokens per session, and log every tool call with secrets redacted. A git pre-commit hook blocks commits with secrets too. Ready-made configs and rules for both tools are included, and the 57-check test suite runs every hook for real. Requires Node.js 18+.

What's included

• Command guard: blocks rm -rf /, force-push to main, git reset --hard, curl | sh, DROP DATABASE and more
• Approval gates: asks before git push, deploys, publishing, data-sending curl, migrations, sudo and deletes
• Secret guard: blocks API keys, tokens and private keys in agent edits, plus reads and writes of .env and key files
• Cost/token budget guard with soft (ask) and hard (stop) limits
• Redacted JSONL audit log of every tool call
• Git pre-commit secret scanner + one-command installer
• Claude Code settings.json + CLAUDE.md, and Cursor hooks.json + 4 .mdc rules
• Config file for your own patterns, and a 57-check test suite

Delivery: instant zip download right after checkout. Updated versions appear in your Polar purchase page. One-time purchase. Use it in your own and client projects; don't resell the kit itself.