careful

/careful — Safety Guardrail for Destructive Commands

Role: Safety Guardrail Inputs: none Outputs: updates .magneto/safety.json Audit actions: careful.activated, careful.warned, careful.overridden

---

What /careful Does

Activates command scanning. When /careful is ON and you propose a destructive shell command, you MUST:

1. Warn the user explicitly, naming the command and its destructive pattern.
2. Require explicit confirmation ("yes run this") — "okay" or "proceed" are NOT sufficient.
3. Record the override via magneto audit add careful.overridden --payload '{"command":"..."}'.
4. If the user declines, STOP and propose a safer alternative.

Destructive Patterns Scanned

Severity	Pattern
critical	rm -rf recursive removal
critical	DROP TABLE/DATABASE/SCHEMA
critical	dd if=... of=/dev/..., mkfs
critical	Fork bomb patterns
critical	kubectl delete ns/namespace/cluster
critical	terraform destroy
high	TRUNCATE TABLE
high	git push --force / -f
high	git reset --hard
high	git clean -dfx
high	chmod -R 777

Full list in .magneto/safety.json after running magneto careful.

Activation

Run:

magneto careful          # enable
magneto careful --off    # disable
magneto careful status   # show current state
magneto careful check '<command>'   # test a command without running it

Integration Contract (for host agents)

Before executing any shell command, read .magneto/safety.json. If careful: true:

# Pseudocode for host
state = read(".magneto/safety.json")
if state["careful"]:
    matches = scan_for_destructive_patterns(command)
    if matches:
        warn_user(matches)
        if not explicit_confirmation():
            abort()
        record_override(command, reason)

Magneto's CLI provides magneto careful check '<command>' for programmatic testing — exit code 0 means safe, 2 means matches found.