careful

name	careful
version	1.0.0
description	Destructive command safety guardrails. Warns before dangerous operations in production and shared environments.
author	iscmga
tags	["safety","guardrails","production","destructive","protection"]
triggers	{"globs":[],"keywords":["careful","safe mode","production safety","guardrails","be careful","careful mode"]}

Careful

Pre-execution safety guardrails that warn before dangerous operations. Prevents accidental destruction of infrastructure, data, and shared state. Inspired by gstack's careful/guard methodology.

When to Activate

• Before any work in production environments
• "Be careful with these changes"
• "Enable careful mode"
• When working with shared infrastructure
• During incident response (prevent making things worse)
• Automatically when detecting production context (profile names, account IDs)

Dangerous Command Patterns

Infrastructure — ALWAYS WARN

Pattern	Risk	Alternative
terraform destroy	Destroys all managed resources	Target specific resources with -target
terraform state rm	Removes resource from state (orphans it)	Use terraform state mv for renames
terraform force-unlock	Breaks state lock	Verify lock holder first
terraform apply -auto-approve	No review of changes	Remove -auto-approve, review plan
terraform import (to wrong state)	Corrupts state	Verify state file and resource first

AWS — ALWAYS WARN

Pattern	Risk	Alternative
aws s3 rm --recursive	Deletes all objects in bucket	Use --dryrun first
aws s3 rb --force	Deletes bucket and all contents	Empty bucket first, review
aws rds delete-db-instance	Deletes database	Verify snapshot exists first
aws ec2 terminate-instances	Terminates instances permanently	Stop first, verify instance ID
aws iam delete-role	Removes IAM role	Check what depends on it first
aws route53 change-resource-record-sets DELETE	Removes DNS records	Verify record and impact
aws ecs update-service --desired-count 0	Stops all tasks	Verify it's the right service/cluster
aws cloudformation delete-stack	Deletes entire stack	Review stack resources first
aws organizations remove-account	Removes account from org	Verify account ID

Git — ALWAYS WARN

Pattern	Risk	Alternative
git push --force	Overwrites remote history	Use --force-with-lease
git reset --hard	Discards all uncommitted changes	git stash first
git branch -D	Deletes branch even if unmerged	git branch -d (safe delete)
git checkout .	Discards all working changes	git stash first
git clean -fd	Deletes untracked files	git clean -fdn (dry run) first

Database — ALWAYS WARN

Pattern	Risk	Alternative
DROP TABLE / DROP DATABASE	Permanent data loss	Backup first, use IF EXISTS
TRUNCATE TABLE	Deletes all rows	Verify table and environment
DELETE FROM without WHERE	Deletes all rows	Add WHERE clause
ALTER TABLE DROP COLUMN	Permanent data loss	Backup table first

System — ALWAYS WARN

Pattern	Risk	Alternative
rm -rf / or rm -rf *	Destroys filesystem	Be specific with paths
kill -9 on production PIDs	Abrupt process termination	kill -15 (graceful) first
chmod -R 777	Opens permissions wide	Use specific permissions

Safe Exceptions

These are always safe to delete/clean (build artifacts):

• node_modules/, .next/, dist/, build/
• __pycache__/, .cache/, .turbo/
• coverage/, .terraform/ (NOT .terraform.lock.hcl)
• *.pyc, *.pyo, *.o

Production Context Detection

Automatically engage careful mode when detecting:

• AWS profile names containing: prod, production, org, management
• Account IDs matching known production accounts
• Terraform workspaces named prod or production
• Branch names: main, master, production
• Environment variables: ENV=production, NODE_ENV=production

Warning Format

When a dangerous command is detected:

WARNING: DESTRUCTIVE OPERATION DETECTED

Command: terraform destroy -target=aws_rds_instance.main
Risk: This will permanently delete the RDS instance and all its data.
Environment: production (account 049005703416)

Before proceeding:
1. Verify you have a recent snapshot/backup
2. Confirm this is the correct resource
3. Verify no other services depend on this resource

Proceed? [y/N]

Guard Mode (Extended)

When maximum protection is needed, combine careful with directory scoping:

1. Restrict file edits to a specific directory
2. All destructive commands require confirmation
3. No commands execute outside the scoped directory

Useful for:

• Production incident response (only edit the affected service)
• Sensitive infrastructure changes (only edit the target module)
• Training/onboarding (prevent accidental damage)

Anti-Patterns

• Disabling guardrails: Never bypass careful mode "because it's faster." The 5 seconds of confirmation prevents hours of recovery.
• Warning fatigue: If you find yourself auto-approving every warning, the scope is too broad. Narrow the protection to what matters.
• False safety: Careful mode doesn't replace proper testing and review. It's a last line of defense, not a substitute for process.

Integration Points

With ship skill

Ship should inherit careful's guardrails. If careful would warn, ship should too.

With investigate skill

Always enable careful mode during incident investigation to prevent making things worse.

With review skill

Review should flag the same destructive patterns that careful guards against.