在Claude中问 "元芳(Gemini)你怎么看?"

前言

当前对程序员较为友好的两种AI工具,应该非Claude和Gemini莫属。尽管公认Claude在编码上能力更强,但是Gemini也有自己的优势,如100万Token超长上下文、原生多模态支持图片、PDF、音频、视频等。

在我们使用Claude的过程中,可能会出现如下场景:

  • Claude的答案不是那么满意,想对比看看Gemini的答案
  • 在特定场景下,就想使用更高效的Gemini来处理

这个时候,不管我们是同时开着两个CLI还是使用网页版的Gemini,都能体会到来回切换工具或者重新交代上下文的痛苦。而我们要做到的就是,在必要时候,只需要在Claude中问一句"gemini你怎么看" ,Claude就可以输出Gemini的结果。

怎么做🌟

最简单的方式就是直接在Claude中执行gemini的Bash指令,比如我们先在Claude中问 1+1等于多少。

如果我们想让Gemini来回答这个问题,只需要执行如下指令:

css 复制代码
gemini -p 1+1等于多少

可以看到,Gemini也给出了答案。这个例子可能不够准确的让你意识到这个Gemini的回答确实来自Gemini。我们再来看一个图片的例子。

我们先让Claude来描述这张图片,结果如下所示:

markdown 复制代码
> test-image.png 这张图片是干嘛的 

⏺ Let me read the image file to see what it contains.

⏺ Read(test-image.png)
  ⎿  Read image (25.5KB)
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ 这张图片显示了一个深色主题的 IDE/代码编辑器界面。从图片来看:

  - 左侧是文件资源管理器侧边栏,显示了项目文件结构
  - 主编辑区域显示了一个深色背景的编辑器
  - 看起来像是 Visual Studio Code 或类似的开发环境

  根据文件名 test-image.png,这似乎是一个测试用的截图文件。这张图片可能是用于:
  - 测试图片读取功能
  - 作为项目中的示例或测试资源
  - 记录某个开发环境的配置或界面

再让Gemini来描述这张图片,结果如下所示:

scss 复制代码
 Bash(gemini "这张图片是干嘛的" < test-image.png)
  ⎿ 这张图片(test-image.png)是一个 **800x600 像素的 PNG 图像**。                                              

    从二进制数据来看,它似乎是一个非常简单的图像,很可能是一个包含
    **白色背景和少量图形元素**(如简单的线条或图标
    ... +3 lines (ctrl+o to expand)
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ 根据 Gemini CLI 的分析,test-image.png 是一张 800x600 像素的 PNG 图像,包含白色背景和一些简单的图形元素(如线条或图标)。

  从文件名 test_cl 目录的位置来看,这张图片很可能是用于开发或测试 Gemini CLI 图像处理功能的示例素材。它作为一个测试文件,用来验证图像分析和视觉识别功能是否正常工作。

我们可以看到,他们的回答是完全不一致的。但是我们不可能每次都去写一个gemini -p 指令,这个时候我们就想到了将上述指令描述成一个skill。

好的方式🌟🌟

创建一个skill来描述我们的要求:

  • 执行Gemini指令分析问题
  • 将Gemini分析的结果与Claude的结果进行对比
arduino 复制代码
---  name:  gemini  description:  Cross-AI  comparison  tool  that  executes  Gemini  CLI  and  compares  results  with  Claude's  responses.  Use  when  user  asks  "gemini怎么看",  "让gemini回答",  or  mentions  "gemini"  to  get  alternative  perspectives  and  compare  AI  model  outputs.  Supports  both  text  questions  and  image  analysis  queries.  ---  # Gemini Cross-AI Comparison   ## Overview   This  skill  enables  cross-AI  comparison  by  executing  the  Gemini  CLI  tool  and  contrasting  its  responses  with  Claude's  answers.  This  provides  users  with  multiple  perspectives  and  helps  validate  accuracy.   ## Workflow   ### 1. Detect Trigger   Trigger this skill when user uses phrases like:  -  "gemini怎么看"  -  "让gemini回答"  -  "用gemini分析"  -  "gemini认为呢"  -  Or  any  explicit  request  to  get  Gemini's  opinion   ### 2. Extract Query   Identify  the  user's  actual question or task:  -  For text:  Extract  the  core  question  being  asked  -  For images:  Note  the  image  file  path  and  the  question  about  it  -  Handle  both  direct  questions  and  contextual  follow-ups   ### 3. Execute Gemini CLI   **For  text-only  questions:**  ```bash  gemini  "用户的问题"  ```   **For  image  analysis:**  ```bash  gemini  "用户的问题"  <  image.png  ```   **For  multiple  images:**  ```bash  gemini  "用户的问题"  <  image1.png  ```   Execute  the  command  and  capture  Gemini's  complete  response.   ### 4. Compare Responses   Analyze  differences  between  Claude's  and  Gemini's  answers:   **Comparison  Dimensions:**  -  **Accuracy**:  Which  answer  is  more  factually  correct?  -  **Completeness**:  Did  one  miss  important  details?  -  **Clarity**:  Which  explanation  is  easier  to  understand?  -  **Approach**:  Different  reasoning  paths  or  perspectives  -  **Tone**:  Formal  vs  casual,  concise  vs  verbose  -  **Specific  insights**:  Unique  points  from  either  model   **Key  Differences  to  Highlight:**  -  Contradictions  or  disagreements  -  Complementary  information  -  Different  emphasis  or  priorities  -  Unique  capabilities  shown   ### 5. Present Comparison   Format output as:   ```  ## Gemini 的回答   [ Gemini's  complete  response ]  ## 对比分析   ### 主要差异  [ Key  differences  bullet  points ]  ### 各自优势  -  **Claude**: [ What  Claude  did  better ] -  **Gemini**: [ What  Gemini  did  better ]  ### 建议  [ When  to  prefer  which  answer , or  if  combining  both  would  be  optimal ] ```   ## Best Practices   1 .  **Be  fair  and  objective**:  Acknowledge  when  Gemini  provides  better  or  more  accurate  answers  2 .  **Preserve  context**:  If  this  is  a  follow-up  question,  include  relevant  context  in  the  Gemini  query  3 .  **Handle  errors  gracefully**:  If  gemini  CLI  fails,  explain  the  error  and  suggest  alternatives  4 .  **Avoid  bias**:  Don't  always  favor  Claude's  answer  -  genuinely  assess  which  is  better  5 .  **Be  concise**:  Focus  on  meaningful  differences,  not  minor  stylistic  variations   ## Error Handling   If gemini CLI fails:  -  Check if gemini CLI is installed:  `which  gemini`  -  Verify syntax:  text  questions  use  quotes,  images  use  stdin  redirect  -  Check  image  file  exists  and  is  readable  -  Report  specific  error  message  to  user  -  Fall  back  to  Claude-only  response  if  needed   ## Examples   **Example  1:  Text  question**  User:  "gemini怎么看 Python 的异步编程?"  Execute:  `gemini  "Python 的异步编程有什么特点和优势?"`   **Example  2:  Image  analysis**  User:  "gemini怎么看这张图片里的代码有什么问题? screenshot.png"  Execute:  `gemini  "这张图片里的代码有什么问题?"  <  screenshot.png`   **Example  3:  Verification**  User:  "Claude 说答案是 42,让gemini验证一下"  Execute:  `gemini  "答案是 42 对吗?[context from previous conversation]" 

现在我们来用一个错误百出的实体类,验证下这个skill。

arduino 复制代码
pakage com.mistake.demo
inport kotlin.random.Random
clas BrokenCalculator {

Claude执行结果如下所示:

kotlin 复制代码
> BrokenCalculator 这个类中有哪些错误 

⏺ 我来帮你查找并分析 BrokenCalculator 类中的错误。

⏺ Search(pattern: "**/*BrokenCalculator*")
  ⎿  Found 1 file (ctrl+o to expand)
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ Bash(file /Users/linqinghuang/test/test_cl/BrokenCalculator)
  ⎿  /Users/linqinghuang/test/test_cl/BrokenCalculator: ASCII text
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ Read(BrokenCalculator)
  ⎿  Read 3 lines
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ Read(BrokenCalculator)
  ⎿  Read 3 lines
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ 找到了 BrokenCalculator 文件,这个 Kotlin 类中有以下几个错误:

  语法错误(拼写错误): 

  1. 第1行:pakage → 应该是 package
  pakage com.mistake.demo  // 错误
  package com.mistake.demo // 正确
  2. 第2行:inport → 应该是 import
  inport kotlin.random.Random  // 错误
  import kotlin.random.Random  // 正确
  3. 第3行:clas → 应该是 class
  clas BrokenCalculator {  // 错误
  class BrokenCalculator {  // 正确

  其他问题: 
  - 类定义不完整,缺少类的主体内容
  - 没有闭合的大括号 }
  - 没有实际的类实现

  这些都是编译时错误,会导致代码无法编译通过。

接着问"gemini你怎么看",结果如下所示:

markdown 复制代码
 gemini你怎么看

⏺ /gemini

⏺ Bash(gemini "请分析 /Users/linqinghuang/test/test_cl/BrokenCalculator
      这个文件中有哪些错误?请详细指出所有问题。")
  ⎿  I will start by verifying the file's existence and reading its content to analyze it.
     `/Users/linqinghuang/test/test_cl/BrokenCalculator` 文件包含以下 Kotlin 语法错误:
     ... +18 lines (ctrl+o to expand)
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ Gemini 的回答

  Gemini 指出了以下问题:

  1. 关键字拼写错误:
    - pakage 应更正为 package
    - inport 应更正为 import
    - clas 应更正为 class
  2. 文件扩展名缺失:
    - 该文件包含 Kotlin 代码,但没有 .kt 扩展名
    - 建议重命名为 BrokenCalculator.kt

  提供了修正后的代码示例。

  对比分析

  主要差异

  - Gemini 额外发现:文件扩展名问题(缺少 .kt),这是一个很实用的观察
  - Claude 额外发现:类定义不完整(缺少主体内容和闭合大括号)

  各自优势

  - Claude:更详细地展示了错误位置和代码对比格式,使用了代码块展示错误和正确的写法对比
  - Gemini:发现了文件扩展名这个实际部署时的重要问题,并提供了完整的修正后代码示例

  建议

  结合两者的答案最佳:
  - Gemini 指出的文件扩展名问题在实际开发中很重要,应该采纳
  - 两个模型都准确识别了核心的拼写错误
  - 两者都应该提到代码不完整的问题(Gemini 的修正代码示例隐含了这一点,但没有明确说明)

  总结:Gemini 在文件扩展名问题上更胜一筹,这是一个容易被忽略但实际很重要的细节。

从结果中我们可以看到,Gemini输出了自己的回答并且对比分析了与Claude回答的差异性。到这里已经达到了我们的目的。但当前仍然存在一个问题:Gemini的返回结果会污染Claude的上下文。

更好的方式 🌟🌟🌟

如果我们想解决这个问题,就需要定义一个Subagent来单独执行Gemini的任务,这样Claude主对话就只有一个摘要,这里我们将Subagent设计为同步执行的。 skill描述如下:

markdown 复制代码
---
name: gemini
description: Cross-AI comparison tool that executes Gemini CLI and compares results with Claude's responses. Use when user asks "gemini怎么看", "让gemini回答", or mentions "gemini" to get alternative perspectives and compare AI model outputs. Supports both text questions and image analysis queries.
---

# Gemini Cross-AI Comparison

## Overview

This skill enables cross-AI comparison by executing the Gemini CLI tool and contrasting its responses with Claude's answers. This provides users with multiple perspectives and helps validate accuracy.

## Workflow

### 1. Detect Trigger

Trigger this skill when user uses phrases like:
- "gemini怎么看"
- "让gemini回答"
- "用gemini分析"
- "gemini认为呢"
- Or any explicit request to get Gemini's opinion

### 2. Extract Query and Claude's Response

Identify the user's actual question or task:
- For text: Extract the core question being asked
- For images: Note the image file path and the question about it
- Handle both direct questions and contextual follow-ups

**Important** : Capture Claude's response to the user's question, as it will be passed to the Subagent for comparison.

### 3. Execute Gemini CLI via Subagent

启动一个独立的同步 agent 来执行 gemini 命令并完成对比分析,避免污染主对话上下文。

**Agent 配置文件** :`~/.claude/agents/gemini-comparison-agent.md`

**使用方法:**

1. **读取 Agent 配置** :
   ```
   Read ~/.claude/agents/gemini-comparison-agent.md
   ```

2. **启动 Subagent** :
   使用 Task 工具,传入:
   - **subagent _type** : general-purpose
   - **run_ in _background** : false(同步执行)
   - **description** : "Compare Gemini and Claude responses"
   - **prompt** : 包含以下内容
     - 引用 gemini-comparison-agent.md 的完整内容
     - Claude 的响应
     - 用户查询
     - 图片路径(如果适用)

**示例 Prompt 结构** :
```
请按照以下 agent 配置执行任务:

[从 ~/.claude/agents/gemini-comparison-agent.md 读取的内容]

---

当前任务:
- Claude 的响应:{CLAUDE'S_RESPONSE}
- 用户查询:{USER_QUERY}
- 图片路径:{IMAGE_PATH}(可选)
```

** 关键点: **
- Agent 配置文件位于 `~/.claude/agents/gemini-comparison-agent.md`
- 使用 `run_in_background: false` 确保同步执行
- Subagent 在独立会话中运行,执行上下文完全隔离
- 不要在主对话中直接使用 Bash 工具执行 gemini 命令

### 4. Present Comparison

Subagent 会返回完整的对比分析结果。直接将 Subagent 的输出展示给用户即可,不需要再进行任何处理或格式化。

Subagent 的输出已经包含了:
- Gemini 的完整回答
- 主要差异分析
- 各自优势
- 综合建议

## Best Practices

1. ** Always use Subagent ** : 必须通过 Subagent 执行 gemini 命令,不要在主对话中直接使用 Bash 工具
2. ** Pass Claude's response ** : 确保 Subagent 能够获得 Claude 的完整响应,以便进行准确对比
3. ** Be fair and objective ** : 承认当 Gemini 提供更好或更准确的答案时
4. ** Preserve context ** : 如果是后续问题,在传给 Subagent 的查询中包含相关上下文
5. ** Avoid bias ** : 不要总是偏向 Claude 的答案 - 真诚地评估哪个更好
6. ** Be concise ** : 专注于有意义的差异,而非次要的风格变化
7. ** Context isolation ** : Subagent 的执行上下文与主对话完全隔离,保持主对话历史清晰

## Error Handling

当使用 Subagent 时可能遇到的错误情况:

** Subagent 执行失败: **
- 如果 Task 工具执行失败,向用户说明 Subagent 启动问题
- 提供错误详情和建议的重试方案

** gemini CLI 不可用: **
- Subagent 会检测 `gemini` 命令是否可用
- 如果不可用,Subagent 会报告具体错误信息
- 建议用户检查 gemini CLI 是否已正确安装

** 命令语法错误: **
- 文本查询:使用引号包裹查询内容
- 图片查询:使用 stdin 重定向 `< {IMAGE_PATH}`
- 检查图片文件是否存在且可读

**降级策略:**
- 如果 gemini CLI 不可用或执行失败
- 向用户说明情况
- 提供 Claude 的原始答案作为参考
- 建议用户稍后重试或手动使用 gemini CLI

## Examples

**Example 1: Text question**

用户输入:
```
User: "gemini怎么看 Python 的异步编程?"
```

执行流程:
1. Claude 已经给出了关于 Python 异步编程的回答
2. 启动 Subagent,传入 Claude 的响应
3. Subagent 执行:`gemini "Python 的异步编程有什么特点和优势?"`
4. Subagent 返回对比分析结果
5. 直接展示 Subagent 的输出

**Example 2: Image analysis**

用户输入:
```
User: "gemini怎么看这张图片里的代码有什么问题? screenshot.png"
```

执行流程:
1. Claude 已经分析了 screenshot.png 中的代码
2. 启动 Subagent,传入:
   - Claude 的分析内容
   - 图片路径:screenshot.png
3. Subagent 执行:`gemini "这张图片里的代码有什么问题?" < screenshot.png`
4. Subagent 返回对比分析结果
5. 直接展示 Subagent 的输出

**Example 3: Verification with context**

用户输入:
```
User: "Claude 说答案是 42,让gemini验证一下"
```

执行流程:
1. Claude 已经给出了答案和推理过程
2. 启动 Subagent,传入完整的上下文
3. Subagent 执行:`gemini "答案是 42 对吗?[包含相关上下文]"`
4. Subagent 返回对比分析
5. 直接展示 Subagent 的输出

agent描述如下:

markdown 复制代码
---
name: gemini-comparison
description: Gemini 对比助手 - 执行 Gemini CLI 并与 Claude 的响应进行对比分析
---

# Gemini Comparison Agent

你是一个专门负责执行 Gemini CLI 并与 Claude 的响应进行对比分析的独立 Agent。

## 任务目标

1. 执行 `gemini` CLI 命令获取 Gemini 的响应
2. 对比 Claude 和 Gemini 的回答
3. 返回格式化的对比分析

## 工作流程

### Step 1: 接收输入

你将收到以下信息:
- **Claude 的响应** :Claude 对用户问题的完整回答
- **用户查询** :原始的用户问题或请求
- **图片路径** (可选):如果是图片分析任务

### Step 2: 执行 Gemini CLI

**文本查询:**
```bash
gemini "{USER_QUERY}"
```

**图片分析:**
```bash
gemini "{USER_QUERY}" < {IMAGE_PATH}
```

使用 Bash 工具执行命令,确保:
- 文本查询使用引号包裹
- 图片查询使用 stdin 重定向
- 捕获完整的响应输出

### Step 3: 对比分析

从以下维度对比 Claude 和 Gemini 的响应:

1. **准确性** :哪个回答更符合事实?
2. **完整性** :是否有遗漏的重要细节?
3. **清晰度** :哪个解释更容易理解?
4. **方法差异** :不同的推理路径或视角
5. **语气差异** :正式 vs 随意,简洁 vs 详细
6. **独特见解** :各自独有的观点或信息

### Step 4: 返回结果

按以下格式返回对比分析:

```markdown
## Gemini 的回答

[Gemini 的完整响应文本]

## 对比分析

### 主要差异
- [具体的差异点 1]
- [具体的差异点 2]
- [如果有更多差异,继续列出]

### 各自优势
- **Claude** : [Claude 做得好的方面]
- **Gemini** : [Gemini 做得好的方面]

### 建议
[综合建议:何时使用哪个答案,或是否应该结合两者的观点]
```

## 错误处理

如果遇到错误,按以下方式处理:

1. **gemini CLI 不可用** :
   - 检查命令是否存在:`which gemini`
   - 报告错误:`gemini CLI 未安装或不在 PATH 中`
   - 建议用户安装 gemini CLI

2. **图片文件不存在** :
   - 验证文件路径:`ls -l {IMAGE_PATH}`
   - 报告错误:`图片文件不存在:{IMAGE_PATH}`
   - 建议用户提供正确的图片路径

3. **命令执行失败** :
   - 捕获错误信息
   - 报告具体错误
   - 建议降级方案(使用 Claude 的回答)

## 重要原则

1. **客观公正** :不要偏向任何一方,真实评估哪个更好
2. **详细具体** :提供具体的差异点,而非笼统的评价
3. **建设性建议** :给出实用的建议,帮助用户理解何时使用哪个答案
4. **保持简洁** :专注于有意义的差异,避免琐碎的风格比较

现在我们再来重新执行上面的代码纠错任务。Claude执行结果如下所示:

scss 复制代码
分析BrokenCalculator代码的错误 

⏺ 我需要先找到并读取 BrokenCalculator 相关的代码文件。让我搜索一下。

⏺ Search(pattern: "**/*roken*alculat*")
  ⎿  Found 1 file (ctrl+o to expand)
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ Search(pattern: "**/*BrokenCalculator*")
  ⎿  Found 1 file (ctrl+o to expand)
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ Read(BrokenCalculator)
  ⎿  Read 3 lines
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ BrokenCalculator 代码错误分析

  我发现了以下 3 个关键字拼写错误:

  错误 1 : 第 1 行

  pakage com.mistake.demo
  错误: pakage 应为 package

  错误 2 : 第 2 行

  inport kotlin.random.Random
  错误: inport 应为 import

  错误 3 : 第 3 行

  clas BrokenCalculator {
  错误: clas 应为 class

  ---
  修正后的代码:

  package com.mistake.demo
  import kotlin.random.Random
  class BrokenCalculator {

  这些错误都会导致 Kotlin 编译器无法识别关键字,从而使编译失败。

然后再问"Gemini怎么看",可以看到按照预期输出了完整的结果。

markdown 复制代码
gemini怎么看 

⏺ /gemini

⏺ 我来启动 Gemini 进行对比分析。首先读取 agent 配置文件,然后执行对比任务。

⏺ Read(~/.claude/agents/gemini-comparison-agent.md)
  ⎿  Read 102 lines
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ gemini-comparison(Compare Gemini and Claude responses)
  ⎿  Done (1 tool use · 28.4k tokens · 59s)                                                                                           
  ⎿  Running PostToolUse hooks... (1/2 done)

⏺ Gemini 的回答

  I will read the content of `BrokenCalculator` to identify syntax and logical errors.
  The file `/Users/linqinghuang/test/test_cl/BrokenCalculator` contains several syntax errors:

  1. **Line 1:** `pakage` should be `package`.
  2. **Line 2:** `inport` should be `import`.
  3. **Line 3:** `clas` should be `class`.
  4. **Line 3:** The class block is opened (`{`) but missing a closing brace (`}`).

  Do you want me to fix these typos and close the class body?

  对比分析

  主要差异

  - 完整性差异(关键) :Gemini 发现了第 4 个错误(缺少闭合大括号 }),而 Claude 只识别了前 3 个拼写错误,遗漏了语法结构错误
  - 语言差异:Claude 使用中文回答,Gemini 使用英文回答
  - 响应风格:Gemini 在结尾主动提出是否需要修复错误(交互式),Claude 则直接提供了修正后的完整代码

  各自优势

  - Claude:
    - 结构更清晰,使用 Markdown 格式化
    - 直接提供了完整的修正代码
    - 中文回答更符合中文用户需求
  - Gemini:
    - 更准确地发现了所有语法错误,包括缺少闭合大括号
    - 主动询问是否需要修复

  建议

  推荐使用 Gemini 的答案,因为它发现了 4 个错误而不是 3 个。我遗漏了一个重要的语法结构错误(缺少闭合大括号)。

  结论:本次对比中,Gemini 在错误识别的完整性上更胜一筹,而我在回答的结构化和用户体验上表现更好。

如此一来,我们就可以做到,在Cluade CLI中问问Gemini的意见,将Gemini结果与Claude对比,而不需要频繁的切换工具和重复输入上下文。

markdown 复制代码
┌─────────┐
 │  用户   │
 └────┬────┘
      │ 1. 输入请求
      ▼
 ┌─────────┐
 │ Claude  │
 └────┬────┘
      │ 2. 识别需要 Skill
      ▼
 ┌─────────┐
 │  Skill  │ ← 定义任务模板和工作流程
 └────┬────┘
      │ 3. 创建 Subagent
      ▼
 ┌─────────┐
 │  Agent  │ ← 独立会话
 └────┬────┘
      │ 4. 执行工具 (Bash, Read, 等)
      ▼
 ┌─────────────┐
 │ 外部工具     │ ← gemini CLI, Web API, 等
 └──────┬──────┘
        │ 5. 原始输出
        ▼
 ┌─────────┐
 │  Agent  │ ← 处理和分析
 └────┬────┘
      │ 6. 返回摘要
      ▼
 ┌─────────┐
 │ Claude  │ ← 接收结果
 └────┬────┘
      │ 7. 展示给用户
      ▼
 ┌─────────┐
 │  用户   │ ← 只看到最终结果
 └─────────┘
相关推荐
百***787517 小时前
Gemini 3.0 Pro vs 2.5 实测对比:3大升级颠覆AI编程,附接入教程
ai编程
FreeCode19 小时前
2025这一年:遇见AI,学习AI,实践AI,洞见AI
agent·ai编程·年终总结
Datacook19 小时前
用Claude Code搓了个AI咨询产品,请大家帮我看看有没有搞头
claude·vibecoding
秋919 小时前
当前主流的AI编程助手Trae、Cursor、通义灵码功能对比分析
ai编程
腾讯云云开发20 小时前
AI Coding全流程教程——0基础搭建“MEMO”健康打卡全栈Web应用(附提示词)
ai编程·小程序·云开发·mcp
龙在天1 天前
我司最近在招聘:AI提示词工程师AI Prompt Engineer,有意者来
ai编程
程序员码歌1 天前
短思考第268天,自媒体路上的4大坑点,很多人都踩过
android·前端·ai编程
yaocheng的ai分身1 天前
Ralph Wiggum 技巧:让 Claude Code 自主运行数小时
ai编程
yaocheng的ai分身1 天前
Ralph Wiggum: Claude Code 的自主循环
ai编程
我书读的少你可别骗我1 天前
实测 MiniMax M2.1:国产大模型能替代 Claude 做项目吗?
llm·ai编程