双子座审稿人

工程开发 社区

技能解读

当前原文的自动摘译,尚未人工审校。

执行协议

调用此技能时,您必须: 从对话上下文中识别审核目标: 查找提到的 PR/MR 编号 查找提到的提交哈希值 查找代码片段或实施计划

上下文文件模板

使用以下结构创建 /tmp/gemini_context.txt: [用 - 和 + 显示已删除/添加的行的差异] [显示差异]

执行示例

场景:使用 MR 44 调用 /code-review 命令 遵循上面的执行协议,下面是一个具体示例:

关键规则

始终使用 --approval-mode plan - 只读模式阻止所有工具执行 始终使用 -p 标志 - 使用 stdin 输入时需要提示 切勿使用 -y/YOLO 模式 - 自动批准危险操作(rm、git重置等) 始终使用 -o text - 干净的文本输出格式 不要使用 --sandbox -与企业 UID 范围存在兼容性问题 (4212687 > 60000)

故障排除

如果未安装 Gemini:向用户报告 Gemini CLI 不可用 跳过 Gemini 检查并在组合输出中注明 如果 Gemini 返回错误:在输出中包含错误

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
工程开发
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@tomevault-io · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
需手动接入
是否需要外部接口密钥
不需要
兼容的系统
Docker
运行环境要求
Docker
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
  • 命令执行
网络行为
仅限本地
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md

# Gemini Reviewer Skill

**IMPORTANT:** This skill executes automatically. When invoked, follow these steps precisely.

## Execution Protocol

When this skill is invoked, you MUST:

1. **Identify the review target** from conversation context:
   - Look for PR/MR numbers mentioned
   - Look for commit hashes mentioned
   - Look for code snippets or implementation plans
   - Check recent tool calls (glab mr view, gh pr view, git diff, etc.)

2. **Prepare comprehensive review context** file:
   - Include MR/PR title and description
   - Include all code changes (diff)
   - Include relevant file contents
   - Include upstream issues or comments
   - Include helper functions or related code

3. **Execute Gemini review** using Bash tool with read-only security:
   ```bash
   gemini --approval-mode plan -o text -p "You are an expert code reviewer. Review this [MR/PR/code] and provide detailed feedback on correctness, security, performance, code quality, and testing. Be specific and reference line numbers or code sections." < /tmp/gemini_context.txt
   ```

   **SECURITY NOTE**:
   - `--approval-mode plan`: Read-only mode preventing all tool execution
   - Gemini can only analyze the provided context and return text feedback
   - No file system modifications, git operations, or shell commands possible
   - Note: Docker `--sandbox` not used due to enterprise UID compatibility issues

4. **Return the complete Gemini output** - do not summarize or filter

## Context File Template

Create `/tmp/gemini_context.txt` with this structure:

```
Review [GitLab MR/GitHub PR/Commit] #[NUMBER]: "[TITLE]"

## CONTEXT

**Purpose:** [What this change does]

**Related Issues:** [Links to upstream issues, bugs, etc.]

**Rationale:** [Why this change is needed]

**CI Status:** [Pipeline/test results if available]

## CHANGES SUMMARY

[High-level summary of what files changed and why]

## DETAILED CODE CHANGES

### File: [filename]
```[language]
[Show diff with - and + for removed/added lines]
```

### File: [filename2]
```[language]
[Show diff]
```

## HELPER FUNCTION CONTEXT

[Include any related helper functions or dependencies that aren't in the diff]

## REVIEW FOCUS

Please review focusing on:
1. Correctness: Logic errors, edge cases
2. Security: Input validation, vulnerabilities
3. Performance: Scalability, efficiency concerns
4. Edge Cases: Null handling, error conditions
5. Architecture: Design soundness, alternatives
6. Code Quality: Readability, maintainability
7. Testing: What tests should be added?
8. Configuration: Impact of config changes

Provide specific, actionable feedback with line references.
```

## Example Execution

**Scenario:** /code-review command called with MR 44

Following the execution protocol above, here's a concrete example:

```bash
# Step 2: Prepare context file
cat > /tmp/gemini_context.txt << 'EOF'
Review GitLab Merge Request #44: "Tuning rag score thresholds"

## CONTEXT
[Full MR description, purpose, rationale]

## DETAILED CODE CHANGES
[Complete diff from glab mr diff]

## REVIEW FOCUS
[Specific areas to examine]
EOF

# Step 3: Execute Gemini in read-only plan mode
gemini --approval-mode plan -o text -p "You are an expert code reviewer. Review this merge request and provide detailed feedback on correctness, security, performance, and code quality. Be specific and reference line numbers." < /tmp/gemini_context.txt
```

## Critical Rules

- **ALWAYS use `--approval-mode plan`** - Read-only mode preventing all tool execution
- **ALWAYS use `-p` flag** - Required for prompts when using stdin input
- **NEVER use `-y`/YOLO mode** - Auto-approves dangerous operations (rm, git reset, etc.)
- **ALWAYS use `-o text`** - Clean text output format
- **DO NOT use `--sandbox`** - Has compatibility issues with enterprise UID ranges (4212687 > 60000)
- **DO NOT** summarize or filter Gemini's output - return it verbatim
- **DO gather sufficient context** - include diffs, descriptions, related code
- **DO focus the prompt** - tell Gemini what to look for
- **DO use temporary files** (`/tmp/gemini_context.txt`) for large context

## Troubleshooting

**If Gemini is not installed:**
- Report to user that Gemini CLI is not available
- Skip Gemini review and note it in the combined output

**If Gemini returns an error:**
- Include the error in the output
- Note that Gemini review failed but continue with other reviews

## Integration with /code-review

This skill is automatically invoked by `/code-review` command alongside Claude's code-reviewer agent. Both run in parallel for comprehensive coverage.

---
> Source: [ccronca/dotfiles](https://github.com/ccronca/dotfiles) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-05-23 -->

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。