name: careful version: 0.1.0 description: Safety guardrails for destructive commands. (gstack) triggers:
- be careful
- warn before destructive
- safety mode allowed-tools:
- Bash
- Read
hooks:
PreToolUse:
- matcher: "Bash"
hooks:
- type: command command: "bash $HOME/.claude/skills/gstack/careful/bin/check-careful.sh" statusMessage: "Checking for destructive commands..."
- matcher: "Bash"
hooks:
When to invoke this skill
Warns before rm -rf, DROP TABLE, force-push, git reset --hard, kubectl delete, and similar destructive operations. User can override each warning. Use when touching prod, debugging live systems, or working in a shared environment. Use when asked to "be careful", "safety mode", "prod mode", or "careful mode".
/careful — Destructive Command Guardrails
Safety mode is now active. Every bash command will be checked for destructive patterns before running. If a destructive command is detected, you'll be warned and can choose to proceed or cancel.
mkdir -p ~/.gstack/analytics
echo '{"skill":"careful","ts":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","repo":"'$(basename "$(git rev-parse --show-toplevel 2>/dev/null)" 2>/dev/null || echo "unknown")'"}' >> ~/.gstack/analytics/skill-usage.jsonl 2>/dev/null || true
What's protected
| Pattern | Example | Risk |
|---|---|---|
rm -rf / rm -r / rm --recursive | rm -rf /var/data | Recursive delete |
DROP TABLE / DROP DATABASE | DROP TABLE users; | Data loss |
TRUNCATE | TRUNCATE orders; | Data loss |
git push --force / -f | git push -f origin main | History rewrite |
git reset --hard | git reset --hard HEAD~3 | Uncommitted work loss |
git checkout . / git restore . | git checkout . | Uncommitted work loss |
kubectl delete | kubectl delete pod | Production impact |
docker rm -f / docker system prune | docker system prune -a | Container/image loss |
Safe exceptions
These patterns are allowed without warning:
rm -rf node_modules/.next/dist/__pycache__/.cache/build/.turbo/coverage
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.
Overview
careful is a free skill on OpenRuna. Safety guardrails for destructive commands. (gstack)
What this skill does
Reach for "careful" whenever you need a reliable skill for real work across ChatGPT, Claude, Gemini, and Cursor. Safety guardrails for destructive commands. (gstack) OpenRuna cross-links it to related prompts, agents, and tools, which makes assembling a full workflow around it straightforward. Paste it straight into a chat, drop it into a system prompt, or store it as a reusable skill.
Use cases
- Fork it as a baseline and layer in your own project context, constraints, and examples.
- Keep it in a shared library as the canonical version of this skill for your organisation.
- Combine it with related tools and prompts in the same OpenRuna category to build an end-to-end workflow.
- Reach for it during planning or review sessions when you want consistent, AI-assisted structure.
Example output
Running this skill produces output shaped like the source material below: <!-- AUTO-GENERATED from SKILL.md.tmpl — do not edit directly --> <!-- Regenerate: bun run gen:skill-docs --> ## When to invoke this skill Warns before rm -rf, DROP TABLE, force-push, git reset --hard, kubectl delete, and similar destructive operations. User can override each warning. Use when touching prod, debugging live systems, or working in a shared environment. Use when asked to "be careful", "safety mode", "prod mode", or "careful mode". # /careful — Destructive Command Guardrails Safety mode is now **active**. Every bash command will be checked for destructive patterns before runn… Results vary by model and temperature; treat the first response as a draft and refine it with follow-up prompts.
Tips by platform
Claude
With Claude, drop this skill into Project knowledge so every chat in the project inherits it. Ask Claude to restate the goal first, then run — it catches edge cases early.
ChatGPT
In ChatGPT, start a fresh chat and paste this skill verbatim, then follow up with "apply this to [your context]." Pick a current GPT model for coding or reasoning tasks.
Cursor
Add this skill to your Cursor rules and invoke it from Agent mode for repeatable results. Link back to its OpenRuna page in the rule so the source stays discoverable.
Frequently asked questions
- What is "careful"?
- It is a skill listed on OpenRuna — Safety guardrails for destructive commands. (gstack) You can copy and adapt it for ChatGPT, Claude, Cursor, or any other AI assistant.
- Is "careful" free to use?
- Most OpenRuna resources are open or CC0-licensed. Check the license shown on this page before commercial use; premium collections are clearly marked as such.
- How do I get the best results from this skill?
- Replace any placeholders, add your project context, and ask the model to confirm its assumptions first. Iterate over 2–3 follow-up turns rather than expecting a perfect first response.
- Does "careful" work with both Claude and ChatGPT?
- Yes — it is model-agnostic text, so it runs on Claude, ChatGPT, Gemini, and Cursor. The tips on this page cover each of those assistants specifically.
- Where can I find resources related to "careful"?
- Scroll to the Related resources section on this page, or open the matching category hub on OpenRuna to find connected prompts, tools, agents, and datasets in the same topic area.
Related resources
- Tool
v0 Prompts and Tools — ReadFile
Reads file contents intelligently - returns complete files when small, paginated chunks, or targeted chunks when large based on your query. **How it works:** • **Small files** (≤2000 lines) - Returns complete content • **Large files** (>2000 lines) - Uses AI to find and return relevant chunks based on query • **Binary files** - Returns images, handles blob content appropriately • Any lines longer than 2000 characters are truncated for readability • Start line and end line can be provided to rea
- Tool
v0 Prompts and Tools — LSRepo
Lists files and directories in the repository. Returns file paths sorted alphabetically with optional pattern-based filtering. Common use cases: • Explore repository structure and understand project layout • Find files in specific directories (e.g., 'src/', 'components/') • Locate configuration files, documentation, or specific file types • Get overview of available files before diving into specific areas Tips: • Use specific paths to narrow down results (max 200 entries returned) • Combine wi
- Tool
Traycer AI — grep_search
Fast text-based regex search that finds exact pattern matches within files or directories, utilizing the ripgrep command for efficient searching. Results will be formatted in the style of ripgrep and can be configured to include line numbers and content. To avoid overwhelming output, the results are capped at 50 matches. Use the include patterns to filter the search scope by file type or specific paths. This is best for finding exact text matches or regex patterns. More precise than codebase sea
- Tool
Trae — search_codebase
This tool is Trae's context engine. It: 1. Takes in a natural language description of the code you are looking for; 2. Uses a proprietary retrieval/embedding model suite that produces the highest-quality recall of relevant code snippets from across the codebase; 3. Maintains a real-time index of the codebase, so the results are always up-to-date and reflects the current state of the codebase; 4. Can retrieve across different programming languages; 5. Only reflects the current state of the codeba
- Tool
Trae — todo_write
Use this tool to create and manage a structured task list for your current coding session. This helps you track progress, organize complex tasks, and demonstrate thoroughness to the user. It also helps the user understand the progress of the task and overall progress of their requests.
- Tool
Same.dev — task_agent
Launches a highly capable task agent in the USER's workspace. Usage notes: 1. When the agent is done, it will return a report of its actions. This report is also visible to USER, so you don't have to repeat any overlapping information. 2. Each agent invocation is stateless and doesn't have access to your chat history with USER. You will not be able to send additional messages to the agent, nor will the agent be able to communicate with you outside of its final report. Therefore, your prompt shou
- Tool
Replit — shell_command_application_feedback_tool
This tool allows you to execute interactive shell commands and ask questions about the output or behavior of CLI applications or interactive Python programs. ## Rules of usage: 1. Provide clear, concise interactive commands to execute and specific questions about the results or interaction. 2. Ask one question at a time about the interactive behavior or output. 3. Focus on interactive functionality, user input/output, and real-time behavior. 4. Specify the exact command to run, including any ne
- Tool
Replit — str_replace_editor
Custom editing tool for viewing, creating and editing files * State is persistent across command calls and discussions with the user * If `path` is a file, `view` displays the result of applying `cat -n`. If `path` is a directory, `view` lists non-hidden files and directories up to 2 levels deep * The `create` command cannot be used if the specified `path` already exists as a file * If a `command` generates a long output, it will be truncated and marked with `<response clipped>` * The `undo_edi