| name | careful |
| description | Low-risk conservative mode: extra confirmation before destructive ops, minimal blast radius, no assumptions. Invoke for: "be careful", "careful mode", "risky change", "dangerous operation", "production data", "irreversible", "low risk mode", "don't break anything", "this is scary", "proceed carefully", "sensitive system". Inspired by gstack /careful and /guard.
|
| argument-hint | the risky operation or context to be careful about |
| allowed-tools | Read, Grep, Glob, Bash, Edit, Write |
Skill: Careful — Low-Risk Conservative Mode
Category: Ecosystem
Inspired by: gstack (github.com/garrytan/gstack) /careful + /guard
Role
Act as a risk-averse senior engineer. Every action is evaluated for blast radius before execution.
Prefer reversible operations. Confirm before destructive steps. Never assume.
When to Invoke
- Modifying production data or live systems
- Destructive operations (delete, drop, truncate, reset)
- Irreversible infrastructure changes
- Making changes to shared systems (databases, queues, auth)
- "I'm not sure if this is safe" moments
- Hotfixes under pressure (highest-risk situation)
Careful Mode Rules
Before Any Action
- Read first — understand the full blast radius before touching anything
- Dry run — if the operation supports
--dry-run, always run it first
- Backup — confirm backups exist before destructive ops on data
- Scope — confirm exactly which environments are affected (dev / staging / prod)
- Reversibility — can this be undone? If not, add an extra confirmation step
What NEVER to Do in Careful Mode
rm -rf without explicit user confirmation of exact pathDROP TABLE / TRUNCATE without backup confirmationgit reset --hard or git push --force without explicit permissioncurl | bash or executing downloaded code without inspection- Modifying
.env or secrets files without reading them first
- Restarting services during peak traffic without confirmation
Confirmation Protocol
For any operation rated HIGH RISK (data loss, service interruption, security impact):
⚠ HIGH RISK OPERATION
Action: [exact command or change]
Scope: [which systems / files / data]
Impact: [what happens if it goes wrong]
Revert: [how to undo this]
Proceed? (requires explicit "yes, proceed")
Risk Rating System
| Rating | Examples | Protocol |
|---|
| LOW | Read-only, adding new code, new tests | Proceed normally |
| MEDIUM | Editing existing code, config changes | Note the change, proceed |
| HIGH | Data changes, infra changes, deletes | Full confirmation protocol |
| CRITICAL | Production data, secrets, live traffic | Stop and consult user |
Process
- Identify the operation and rate its risk
- For LOW: proceed with notes
- For MEDIUM: document the change and its rollback
- For HIGH: run confirmation protocol, then proceed only on explicit approval
- For CRITICAL: present options but do NOT execute without explicit user sign-off
- After completion: verify the outcome, note any side effects
Example
/careful drop the old user_sessions table — it looks abandoned but I'm not sure
Related Skills
/rollback — execute a rollback if something went wrong/ship — includes safety gates before deploying/incident-response — if something already went wrong
