菜单
Use when about to do dangerous operations (deploys, database migrations, force pushes, production changes) — activates on-demand hook that blocks destructive commands. Also use when the user says "be careful", "careful mode", or wants extra safety. Use /my-careful again to deactivate.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when you need to commit and push changes — stages files, commits with co-author tags, pulls with rebase, resolves conflicts, and pushes. Also supports "pr" mode for creating PRs and "clean" mode for pruning merged branches. Also use for "commit", "push", "sync", or "create PR".
Use when the user wants short-and-sweet answers to one or more questions. Restates each question as a bold heading, then answers in 2–5 lines per question. Bullets and small tables allowed when they help. Also use for "quick question", "qq", "answer briefly", or any "/my-qq <question>" invocation.
Use when starting implementation on any project — reads planning docs, dispatches sub-agents, and creates or reiterates docs/acceptance-criteria.md with measurable criteria per feature. Must run before writing implementation code. Also use for "create acceptance criteria", "what are the requirements", or "define done".
Use when starting a new feature or project from a raw idea — asks targeted questions and produces a structured PRD at docs/prd-<slug>.md. Sits upstream of /my-prompt. Also use for "write a PRD", "product requirements", or "what should we build".
Use when code needs rigorous review before shipping — spawns skeptic/believer/referee agents in iterative rounds until findings converge to nitpicks. Use after implementing features, before merging PRs, or when the user asks for a "thorough review", "adversarial review", or "tear this apart".
Use when facing a complex task that benefits from multiple specialized agents working in parallel — spawns, delegates, tracks status, and collects results. Also use when the user says "use agents", "multi-agent", or "parallelize this".
| name | my-careful |
| description | Use when about to do dangerous operations (deploys, database migrations, force pushes, production changes) — activates on-demand hook that blocks destructive commands. Also use when the user says "be careful", "careful mode", or wants extra safety. Use /my-careful again to deactivate. |
| argument-hint | < on | off > |
Toggles a safety mode that blocks destructive commands via a PreToolUse hook. When active, commands matching dangerous patterns are rejected with an explanation.
/tmp/claude-careful-mode.flag.touch /tmp/claude-careful-mode.flag
Tell the user:
Careful mode ON — destructive commands will be blocked.
Blocked patterns: rm -rf, git push --force, git reset --hard, DROP/TRUNCATE/DELETE without WHERE, docker system prune, kill -9, chmod 777, > /dev/null redirection of important files.
Run /my-careful again to deactivate.
rm -f /tmp/claude-careful-mode.flag
Tell the user:
Careful mode OFF — all commands allowed.
/tmp/ so it survives within a session but gets cleared on reboot — careful mode won't persist across machine restartsdocker compose down is NOT blocked (it's a normal shutdown) — only docker system prune is blockedThe hook at ~/.claude/hooks/careful-mode-guard.sh checks for these patterns when the flag file exists:
| Pattern | Why |
|---|---|
rm -rf / or rm -rf ~ or rm -rf . | Catastrophic deletion |
git push --force or git push -f | Overwrites remote history |
git reset --hard | Discards uncommitted work |
git clean -fd | Deletes untracked files |
git checkout -- . or git restore . | Discards all changes |
DROP TABLE / DROP DATABASE | Database destruction |
TRUNCATE | Data loss |
DELETE FROM without WHERE | Full table wipe |
docker system prune | Removes all unused Docker data |
kill -9 | Ungraceful process kill |
chmod 777 | Insecure permissions |
mkfs | Filesystem format |
dd if= | Raw disk write |