gcop-rs

English

简体中文

English

简体中文

Appearance

Custom Prompts ​

gcop-rs allows you to customize the prompts sent to AI for both commit message generation and code review.

Why Customize Prompts? ​

  • Language: Generate commit messages in Chinese or other languages
  • Style: Match your team's commit message format
  • Focus: Emphasize specific review criteria (security, performance, etc.)
  • Context: Add project-specific guidelines

Auto-Completion ​

gcop-rs automatically appends missing required sections to your custom prompts, so you can focus on writing instructions without worrying about placeholders.

Commit Prompts ​

If your custom prompt doesn't contain {diff}, gcop-rs will automatically append:

  • Git diff section with {diff} placeholder
  • Context section with {files_changed}, {insertions}, {deletions}

Example - Simplified custom prompt:

toml
[commit]
custom_prompt = "Generate commit message in Chinese, be concise"

This will be automatically expanded to:

Generate commit message in Chinese, be concise

## Git Diff:
```
<actual diff content>
```

## Context:
- Files: src/main.rs, src/lib.rs
- Changes: +45 -12

Review Prompts ​

For review prompts, gcop-rs will:

  1. Append {diff} section if missing
  2. Always append JSON output format specification

Example - Simplified custom prompt:

toml
[review]
custom_prompt = "Review this code for security vulnerabilities, focus on SQL injection and XSS"

This will be automatically expanded to include the diff and JSON format instructions.

Template Variables ​

Commit Message Prompts ​

Available placeholders:

  • {diff} - Complete git diff content
  • {files_changed} - Comma-separated list of changed files
  • {insertions} - Number of lines added
  • {deletions} - Number of lines deleted
  • {branch_name} - Current branch name (if available)
  • {branch_info} - Formatted branch info ("- Branch: xxx" or empty)

Code Review Prompts ​

Available placeholders:

  • {diff} - Code diff content

Examples ​

Chinese Commit Messages ​

toml
[commit]
custom_prompt = """
你是一个专业的软件工程师，正在审查 git diff 以生成简洁的中文提交信息。

## Git Diff:
{diff}

## 上下文:
- 修改的文件: {files_changed}
- 新增行数: {insertions}
- 删除行数: {deletions}
- 分支名称: {branch_name}

## 要求:
1. 使用中文生成提交信息
2. 第一行：类型(范围): 简要描述（不超过50字）
3. 空一行
4. 正文：说明改动的原因和影响

常见类型：feat（新功能）、fix（修复）、docs（文档）、refactor（重构）

只输出提交信息，不要解释。
"""

Simple Commit Messages ​

toml
[commit]
custom_prompt = """
Generate a one-line commit message for:

{diff}

Files: {files_changed} (+{insertions} -{deletions})

Format: <type>: <description>
Keep it under 72 characters.
"""

Security-Focused Review ​

toml
[review]
custom_prompt = """
You are a security expert. Review this code for vulnerabilities:

{diff}

Focus on:
1. SQL Injection
2. XSS (Cross-Site Scripting)
3. CSRF
4. Authentication/Authorization flaws
5. Insecure data handling

Output in JSON format:
{{
  "summary": "Security assessment",
  "issues": [
    {{
      "severity": "critical" | "warning" | "info",
      "description": "Issue description",
      "file": "filename",
      "line": line_number
    }}
  ],
  "suggestions": ["Security recommendation"]
}}
"""

Performance-Focused Review ​

toml
[review]
custom_prompt = """
Review this code for performance issues:

{diff}

Check for:
- Inefficient algorithms
- Memory leaks
- Unnecessary allocations
- N+1 queries
- Blocking operations

Return JSON with findings.
"""

Best Practices ​

1. Keep Prompts Focused ​

Shorter, focused prompts often work better than very long ones.

2. Specify Output Format ​

For code review, always request JSON format to ensure proper parsing.

3. Test with Verbose Mode ​

Use gcop-rs -v commit to see the actual prompt sent to AI.

4. Iterate ​

Try different prompts and compare results to find what works best.

Default Prompts ​

If you don't specify custom_prompt, gcop-rs uses built-in defaults:

  • Commit: Generates conventional commit messages in English
  • Review: Comprehensive code review in English with JSON output

See src/llm/prompt.rs in the source code for the exact default templates.

Troubleshooting ​

Issue: LLM Returns Wrong Format ​

If the AI doesn't follow your prompt format:

  1. Be more explicit about the required format
  2. Add examples in the prompt
  3. Use --verbose to debug
  4. Try a different model or adjust temperature

Issue: Non-English Output ​

Some models may ignore language instructions. Try:

  • Being more explicit: "You MUST respond in Chinese"
  • Adding examples
  • Using a different model

See Also ​