skills.rest
sanjanb
sanjanbCOMMUNITY
Explore all skills by this author →
GitHub

careful

Intercepts destructive bash commands and prompts for confirmation before execution.

1|Updated Jul 23, 2026
Search Tags:#bash#safety#guardrails#hook#confirmation#destructive commands#rm -rf
One-click install
npx skills add https://github.com/sanjanb/my-agent-harness --skill careful-sanjanb
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: careful
Source: https://github.com/sanjanb/my-agent-harness/tree/main/skills/careful
Command: npx skills add https://github.com/sanjanb/my-agent-harness --skill careful-sanjanb

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? AI agents and developers can accidentally run destructive commands like rm -rf, DROP TABLE, or git push --force, causing irreversible data loss or production outages. This Skill adds a safety checkpoint that warns before such commands execute. ## Core Features & Use Cases - Destructive Command Detection: A PreToolUse hook scans every Bash command for dangerous patterns including recursive deletes, SQL DROP/TRUNCATE, git force-push, git reset --hard, kubectl delete, and docker prune. - Safe Exceptions: Routine cleanup of build artifacts like node_modules, dist, .next, and pycache passes through without warnings. - User Override: Each warning returns an "ask" permission decision, so you can review the risk and choose to proceed or cancel. - Use Case: While debugging a live production issue, activate safety mode so any kubectl delete or force-push suggested by the agent triggers a confirmation prompt instead of executing immediately. ## Quick Start Ask the agent to enable careful mode with a phrase like "be careful" or "safety mode" before running commands in a production environment.

Frequently Asked Questions about careful

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I prevent accidental rm -rf or destructive commands from an AI agent?▼

Enable a PreToolUse hook that scans every Bash command for destructive patterns before execution. When a match like rm -rf or DROP TABLE is found, the hook returns an ask decision so you can confirm or cancel the command.

What destructive commands does the careful hook detect?▼

It detects recursive deletes (rm -r), SQL DROP TABLE/DATABASE and TRUNCATE, git push --force, git reset --hard, git checkout/restore of the working tree, kubectl delete, and docker rm -f or system prune.

Can I still delete node_modules or build folders without warnings?▼

Yes. Standalone recursive deletes of build artifacts such as node_modules, .next, dist, __pycache__, .cache, build, .turbo, and coverage are whitelisted and pass without prompting.

Can I override a destructive command warning?▼

Yes. The hook returns an ask permission decision rather than a hard block, so you can review the warning message and choose to proceed with the command or cancel it.

How do I turn off safety mode after activating it?▼

The hooks are session-scoped, so safety mode deactivates automatically when you end the conversation or start a new session. No manual cleanup is required.

Related Skills

Explore More Software Engineering Skills→
walid-moswalid-mos
2

git-guardrails

Install a Git PreToolUse hook to block destructive commands.

Community
Intermediate
full-stack-bizfull-stack-biz
26

hook-creator

Build and validate Claude Code plugin hooks with production safety checks.

Official
Advanced
arbazkhan971arbazkhan971
26

ship

Automate shipping code changes with pre-ship checks, dry-run, execution, verification, and rollback.

Community
Advanced
otrebuotrebu
4

claude-permissions

Configure Claude Code permissions and sandboxing in settings.json.

Community
Advanced