git-guardrails-claude-code

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

Audit 0.4.1 · source sha 72a124f48772 · 29 Oct 2026Report an issue or request a re-scan

• Publishervinvcn GitHub userOwner of the repository this skill is read from
• Source foldervinvcn/mattpocock-skills-zh-cn/skills/misc/git-guardrails-claude-codeThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdict72a124f48772A new audit runs when the source changes
• Repository statusNot archivedPushed today
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: 设置 Claude Code hooks,在危险 git commands(push、reset --hard、clean、branch -D 等)执行前阻止它们。适用于用户想防止破坏性 git 操作、添加 git safety hooks,或在 Claude Code 中阻止 git push/reset 时。

---
name: git-guardrails-claude-code
description: 设置 Claude Code hooks,在危险 git commands(push、reset --hard、clean、branch -D 等)执行前阻止它们。适用于用户想防止破坏性 git 操作、添加 git safety hooks,或在 Claude Code 中阻止 git push/reset 时。
---

# 设置 Git Guardrails

设置一个 PreToolUse hook,在 Claude 执行危险 git commands 前拦截并阻止它们。

## 会被阻止的命令

- `git push`(包括 `--force` 在内的所有 variants)
- `git reset --hard`
- `git clean -f` / `git clean -fd`
- `git branch -D`
- `git checkout .` / `git restore .`

被阻止时,Claude 会看到一条 message,说明它无权访问这些 commands。

## 步骤

### 1. 询问安装范围

询问用户:只为**当前 project** 安装(`.claude/settings.json`),还是为**所有 projects** 安装(`~/.claude/settings.json`)?

### 2. 复制 hook 脚本

bundled script 位于:[scripts/block-dangerous-git.sh](scripts/block-dangerous-git.sh)

根据 scope 复制到目标位置:

- **Project**: `.claude/hooks/block-dangerous-git.sh`
- **Global**: `~/.claude/hooks/block-dangerous-git.sh`

用 `chmod +x` 让它可执行。

### 3. 把 hook 加入 settings

添加到对应 settings file:

**Project** (`.claude/settings.json`):

```json
{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/block-dangerous-git.sh"
          }
        ]
      }
    ]
  }
}
```

**Global** (`~/.claude/settings.json`):

```json
{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "~/.claude/hooks/block-dangerous-git.sh"
          }
        ]
      }
    ]
  }
}
```

如果 settings file 已存在,把 hook merge 到现有 `hooks.PreToolUse` array 中,不要覆盖其他 settings。

### 4. 询问是否定制

询问用户是否要在 blocked list 中添加或移除 patterns。相应编辑复制后的 script。

### 5. 验证

运行快速测试:

```bash
echo '{"tool_input":{"command":"git push origin main"}}' | <path-to-script>
```

应以 code 2 退出,并向 stderr 打印 BLOCKED message。

Excerpt, first 1,841 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

1. 2026-10-29
   ✓SAFE B 89 / 100

   first audit · source 72a124f48772

Questions about git-guardrails-claude-code

• github.com/vinvcn/mattpocock-skills-zh-cn/tree/main/skills/misc/ — the audited source at commit 72a124f48772 and every finding aboveObserved · 29 Oct 2026

Spotted something wrong? Report a correction · Methodology