# Mk Careful

> Session-scoped guardrails warning before rm -rf, DROP TABLE, force-push, git reset --hard, kubectl delete, etc; user can override. NOT for scoping edits to a directory (mk:freeze).

- Skill: `ngocsangyem/mk-careful-3` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add ngocsangyem/mk-careful-3`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ngocsangyem/mk-careful-3/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: ngocsangyem (https://skillmd.com/u/ngocsangyem)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/ngocsangyem/mk-careful-3

---


# /careful — Destructive Command Guardrails

Safety mode is now **active**. Every bash command will be checked for destructive
patterns before running. If a destructive command is detected, you'll be warned
and can choose to proceed or cancel.

```bash
mkdir -p .meowkit/telemetry
echo '{"skill":"careful","ts":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","repo":"'$(basename "$(git rev-parse --show-toplevel 2>/dev/null)" 2>/dev/null || echo "unknown")'"}'  >> .meowkit/telemetry/skill-usage.jsonl 2>/dev/null || true
```

## What's protected

See [references/destructive-patterns.md](references/destructive-patterns.md) for full pattern list.

## How it works

The hook reads the command from the tool input JSON, checks it against the
patterns above, and returns `permissionDecision: "ask"` with a warning message
if a match is found. You can always override the warning and proceed.

To deactivate, end the conversation or start a new one. Hooks are session-scoped.

## Hooks

- **PreToolUse on Bash**: Warns before destructive commands (rm -rf, DROP TABLE, force-push, reset --hard, kubectl delete)
- Session-scoped — only active when `mk:careful` is invoked
- User can override each warning individually
- **Interaction with mk:investigate**: When careful is active during an investigation, destructive-Bash warnings still fire. Debugging commands that touch state require explicit user confirmation per warning — do not bypass.

## Gotchas

- **False positives on legitimate operations**: Pattern matching `rm` or `drop` in file content, not commands → Check command context, not just string presence
- **Overly broad regex blocking development**: Guard triggers on test fixtures or documentation mentioning destructive commands → Scope guards to actual Bash tool invocations only