# Claude Code Safety Net — Block Destructive Commands

> cc-safety-net adds a safety gate for Claude Code tool calls: explain/score risky commands (rm -rf, git resets) and validate custom rules (1,330★).

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

```bash
npx cc-safety-net doctor
npx cc-safety-net explain "rm -rf /"
npx cc-safety-net explain --json "git checkout -- file.txt"
npx -y cc-safety-net --verify-config
```

## Intro

cc-safety-net adds a safety gate for Claude Code tool calls: explain/score risky commands (rm -rf, git resets) and validate custom rules (1,330★).

**Best for:** Teams letting agents run shell commands but needing guardrails, auditability, and consistent “deny by default” rules

**Works with:** Claude Code native hooks/plugins, npx-based installs, and JSON outputs for automated gating

**Setup time:** 5-12 minutes

## Main

- **Command risk UX**: provides explainers for destructive commands before execution.
- **Custom rules**: supports defining policy rules (block global npm installs, dangerous docker actions, etc.).
- **Validation gate**: documents a `--verify-config` step so config changes are checked before you rely on them.
- **Agent-friendly outputs**: includes JSON mode so an agent runner can decide allow/deny deterministically.

### FAQ

- **Is this only for Claude Code?**: It is designed around Claude Code hooks/commands, but the patterns are reusable for other agent runners.
- **How do I start safely?**: Run `doctor`, then try `explain` on a few commands you commonly see in agent sessions.
- **Why validate configs?**: A broken rule file can silently disable protections; validation keeps the gate trustworthy.

## Source & Thanks

> Source: https://github.com/kenryu42/claude-code-safety-net
> License: MIT
> GitHub stars: 1330 · forks: 62

---

<!-- ZH -->

## Quick Use

```bash
npx cc-safety-net doctor
npx cc-safety-net explain "rm -rf /"
npx cc-safety-net explain --json "git checkout -- file.txt"
npx -y cc-safety-net --verify-config
```

## Intro

cc-safety-net 为 Claude Code 的工具调用加安全闸：解释/评估高风险命令（rm -rf、git reset 等），支持自定义规则校验并提供 JSON 输出（1,330★）。

**Best for:** 允许 agent 执行 shell 命令但需要护栏、可审计与一致规则（默认拒绝危险动作）的团队

**Works with:** Claude Code 原生 hooks/plugins、npx 安装方式，以及可用于自动化的 JSON 输出

**Setup time:** 5-12 minutes

## Main

- **命令风险解释**：在执行前解释高风险命令的影响，减少误操作。
- **支持自定义规则**：可配置策略（例如阻止全局 npm 安装、危险 docker 指令等）。
- **配置校验闸门**：提供 `--verify-config`，避免手改规则导致失效。
- **JSON 友好**：支持 JSON 模式，让 allow/deny 决策更可复现。

### FAQ

- **只适用于 Claude Code 吗？**：主要围绕 Claude Code hooks/commands，但思路也可迁移到其他 agent runner。
- **如何安全起步？**：先跑 `doctor`，再对常见命令做 `explain` 看看风险分类是否符合预期。
- **为什么要校验配置？**：规则文件一旦写错可能导致护栏失效；校验能保证闸门可信。

## Source & Thanks

> Source: https://github.com/kenryu42/claude-code-safety-net
> License: MIT
> GitHub stars: 1330 · forks: 62


---
Source: https://tokrepo.com/en/workflows/claude-code-safety-net-block-destructive-commands
Author: Script Depot