model-recommendation

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

AI Model Recommendation for Copilot Chat Modes and Prompts

Copilot聊天模式与提示的AI模型推荐

Mission

任务目标

Analyze

.agent.md

.prompt.md

files to understand their purpose, complexity, and required capabilities, then recommend the most suitable AI model(s) from GitHub Copilot's available options. Provide rationale based on task characteristics, model strengths, cost-efficiency, and performance trade-offs.

分析

.agent.md

或

.prompt.md

文件以理解其用途、复杂度和所需能力，然后从GitHub Copilot的可用选项中推荐最合适的AI模型。基于任务特性、模型优势、成本效益和性能权衡提供理由。

Scope & Preconditions

范围与前提条件

Input: Path to a
```
.agent.md
```
or
```
.prompt.md
```
file
Available Models: GPT-4.1, GPT-5, GPT-5 mini, GPT-5 Codex, Claude Sonnet 3.5, Claude Sonnet 4, Claude Sonnet 4.5, Claude Opus 4.1, Gemini 2.5 Pro, Gemini 2.0 Flash, Grok Code Fast 1, o3, o4-mini (with deprecation dates)
Model Auto-Selection: Available in VS Code (Sept 2025+) - selects from GPT-4.1, GPT-5 mini, GPT-5, Claude Sonnet 3.5, Claude Sonnet 4.5 (excludes premium multipliers > 1)
Context: GitHub Copilot subscription tiers (Free: 2K completions + 50 chat/month with 0x models only; Pro: unlimited 0x + 1000 premium/month; Pro+: unlimited 0x + 5000 premium/month)

输入：
```
.agent.md
```
或
```
.prompt.md
```
文件的路径
可用模型：GPT-4.1、GPT-5、GPT-5 mini、GPT-5 Codex、Claude Sonnet 3.5、Claude Sonnet 4、Claude Sonnet 4.5、Claude Opus 4.1、Gemini 2.5 Pro、Gemini 2.0 Flash、Grok Code Fast 1、o3、o4-mini（含弃用日期）
模型自动选择：VS Code 2025年9月及以上版本支持 - 从GPT-4.1、GPT-5 mini、GPT-5、Claude Sonnet 3.5、Claude Sonnet 4.5中选择（排除倍率>1的高级模型）
订阅上下文：GitHub Copilot订阅层级（免费版：每月2000次补全+50次聊天，仅支持0倍率模型；专业版：无限次0倍率模型使用+每月1000次高级模型使用；专业增强版：无限次0倍率模型使用+每月5000次高级模型使用）

Inputs

输入参数

Required:

```
${input:filePath:Path to .agent.md or .prompt.md file}
```
- Absolute or workspace-relative path to the file to analyze

Optional:

```
${input:subscriptionTier:Pro}
```
- User's Copilot subscription tier (Free, Pro, Pro+) - defaults to Pro
```
${input:priorityFactor:Balanced}
```
- Optimization priority (Speed, Cost, Quality, Balanced) - defaults to Balanced

必填：

```
${input:filePath:Path to .agent.md or .prompt.md file}
```
- 待分析文件的绝对路径或工作区相对路径

可选：

```
${input:subscriptionTier:Pro}
```
- 用户的Copilot订阅层级（Free、Pro、Pro+）- 默认值为Pro
```
${input:priorityFactor:Balanced}
```
- 优化优先级（Speed、Cost、Quality、Balanced）- 默认值为Balanced

Workflow

工作流程

1. File Analysis Phase

1. 文件分析阶段

Read and Parse File:

Read the target
```
.agent.md
```
or
```
.prompt.md
```
file
Extract frontmatter (description, mode, tools, model if specified)
Analyze body content to identify:
- Task complexity (simple/moderate/complex/advanced)
- Required reasoning depth (basic/intermediate/advanced/expert)
- Code generation needs (minimal/moderate/extensive)
- Multi-turn conversation requirements
- Context window needs (small/medium/large)
- Specialized capabilities (image analysis, long-context, real-time data)

Categorize Task Type:

Identify the primary task category based on content analysis:

Simple Repetitive Tasks:
- Pattern: Formatting, simple refactoring, adding comments/docstrings, basic CRUD
- Characteristics: Straightforward logic, minimal context, fast execution preferred
- Keywords: format, comment, simple, basic, add docstring, rename, move
Code Generation & Implementation:
- Pattern: Writing functions/classes, implementing features, API endpoints, tests
- Characteristics: Moderate complexity, domain knowledge, idiomatic code
- Keywords: implement, create, generate, write, build, scaffold
Complex Refactoring & Architecture:
- Pattern: System design, architectural review, large-scale refactoring, performance optimization
- Characteristics: Deep reasoning, multiple components, trade-off analysis
- Keywords: architect, refactor, optimize, design, scale, review architecture
Debugging & Problem-Solving:
- Pattern: Bug fixing, error analysis, systematic troubleshooting, root cause analysis
- Characteristics: Step-by-step reasoning, debugging context, verification needs
- Keywords: debug, fix, troubleshoot, diagnose, error, investigate
Planning & Research:
- Pattern: Feature planning, research, documentation analysis, ADR creation
- Characteristics: Read-only, context gathering, decision-making support
- Keywords: plan, research, analyze, investigate, document, assess
Code Review & Quality Analysis:
- Pattern: Security analysis, performance review, best practices validation, compliance checking
- Characteristics: Critical thinking, pattern recognition, domain expertise
- Keywords: review, analyze, security, performance, compliance, validate
Specialized Domain Tasks:
- Pattern: Django/framework-specific, accessibility (WCAG), testing (TDD), API design
- Characteristics: Deep domain knowledge, framework conventions, standards compliance
- Keywords: django, accessibility, wcag, rest, api, testing, tdd
Advanced Reasoning & Multi-Step Workflows:
- Pattern: Algorithmic optimization, complex data transformations, multi-phase workflows
- Characteristics: Advanced reasoning, mathematical/algorithmic thinking, sequential logic
- Keywords: algorithm, optimize, transform, sequential, reasoning, calculate

Extract Capability Requirements:

Based on

tools

in frontmatter and body instructions:

Read-only tools (search, fetch, usages, githubRepo): Lower complexity, faster models suitable
Write operations (edit/editFiles, new): Moderate complexity, accuracy important
Execution tools (runCommands, runTests, runTasks): Validation needs, iterative approach
Advanced tools (context7/*, sequential-thinking/*): Complex reasoning, premium models beneficial
Multi-modal (image analysis references): Requires vision-capable models

读取与解析文件：

读取目标
```
.agent.md
```
或
```
.prompt.md
```
文件
提取前置元数据（描述、模式、工具、指定的模型）
分析正文内容以识别：
- 任务复杂度（简单/中等/复杂/高级）
- 所需推理深度（基础/中级/高级/专家级）
- 代码生成需求（极少/中等/大量）
- 多轮对话需求
- 上下文窗口需求（小/中/大）
- 特殊能力需求（图像分析、长上下文、实时数据）

任务类型分类：

基于内容分析确定主要任务类别：

简单重复任务：
- 模式：格式化、简单重构、添加注释/文档字符串、基础CRUD操作
- 特点：逻辑直白、上下文需求少、优先快速执行
- 关键词：format、comment、simple、basic、add docstring、rename、move
代码生成与实现：
- 模式：编写函数/类、实现功能、API端点、测试用例
- 特点：中等复杂度、需要领域知识、符合语言规范的代码
- 关键词：implement、create、generate、write、build、scaffold
复杂重构与架构设计：
- 模式：系统设计、架构评审、大规模重构、性能优化
- 特点：深度推理、涉及多组件、需要权衡分析
- 关键词：architect、refactor、optimize、design、scale、review architecture
调试与问题解决：
- 模式：修复Bug、错误分析、系统性故障排查、根因分析
- 特点：分步推理、需要调试上下文、验证需求
- 关键词：debug、fix、troubleshoot、diagnose、error、investigate
规划与研究：
- 模式：功能规划、研究、文档分析、ADR创建
- 特点：只读操作、上下文收集、决策支持
- 关键词：plan、research、analyze、investigate、document、assess
代码评审与质量分析：
- 模式：安全分析、性能评审、最佳实践验证、合规性检查
- 特点：批判性思维、模式识别、领域专业知识
- 关键词：review、analyze、security、performance、compliance、validate
专业领域任务：
- 模式：Django/框架专属、无障碍（WCAG）、测试（TDD）、API设计
- 特点：深度领域知识、框架约定、标准合规
- 关键词：django、accessibility、wcag、rest、api、testing、tdd
高级推理与多步骤工作流：
- 模式：算法优化、复杂数据转换、多阶段工作流
- 特点：高级推理、数学/算法思维、顺序逻辑
- 关键词：algorithm、optimize、transform、sequential、reasoning、calculate

提取能力需求：

基于前置元数据中的

tools

和正文指令：

只读工具（search、fetch、usages、githubRepo）：适合低复杂度、更快的模型
写入操作（edit/editFiles、new）：中等复杂度，准确性重要
执行工具（runCommands、runTests、runTasks）：需要验证，适合迭代方式
高级工具（context7/、sequential-thinking/）：复杂推理，高级模型更有利
多模态（图像分析引用）：需要支持视觉能力的模型

2. Model Evaluation Phase

2. 模型评估阶段

Apply Model Selection Criteria:

For each available model, evaluate against these dimensions:

应用模型选择标准：

针对每个可用模型，从以下维度评估：

Model Capabilities Matrix

模型能力矩阵

Model	Multiplier	Speed	Code Quality	Reasoning	Context	Vision	Best For
GPT-4.1	0x	Fast	Good	Good	128K	✅	Balanced general tasks, included in all plans
GPT-5 mini	0x	Fastest	Good	Basic	128K	❌	Simple tasks, quick responses, cost-effective
GPT-5	1x	Moderate	Excellent	Advanced	128K	✅	Complex code, advanced reasoning, multi-turn chat
GPT-5 Codex	1x	Fast	Excellent	Good	128K	❌	Code optimization, refactoring, algorithmic tasks
Claude Sonnet 3.5	1x	Moderate	Excellent	Excellent	200K	✅	Code generation, long context, balanced reasoning
Claude Sonnet 4	1x	Moderate	Excellent	Advanced	200K	❌	Complex code, robust reasoning, enterprise tasks
Claude Sonnet 4.5	1x	Moderate	Excellent	Expert	200K	✅	Advanced code, architecture, design patterns
Claude Opus 4.1	10x	Slow	Outstanding	Expert	1M	✅	Large codebases, architectural review, research
Gemini 2.5 Pro	1x	Moderate	Excellent	Advanced	2M	✅	Very long context, multi-modal, real-time data
Gemini 2.0 Flash (dep.)	0.25x	Fastest	Good	Good	1M	❌	Fast responses, cost-effective (deprecated)
Grok Code Fast 1	0.25x	Fastest	Good	Basic	128K	❌	Speed-critical simple tasks, preview (free)
o3 (deprecated)	1x	Slow	Good	Expert	128K	❌	Advanced reasoning, algorithmic optimization
o4-mini (deprecated)	0.33x	Fast	Good	Good	128K	❌	Reasoning at lower cost (deprecated)

模型	倍率	速度	代码质量	推理能力	上下文	视觉支持	最佳适用场景
GPT-4.1	0x	快	良好	良好	128K	✅	平衡型通用任务，所有套餐均包含
GPT-5 mini	0x	最快	良好	基础	128K	❌	简单任务、快速响应、成本效益高
GPT-5	1x	中等	优秀	高级	128K	✅	复杂代码、高级推理、多轮对话
GPT-5 Codex	1x	快	优秀	良好	128K	❌	代码优化、重构、算法类任务
Claude Sonnet 3.5	1x	中等	优秀	优秀	200K	✅	代码生成、长上下文、平衡型推理
Claude Sonnet 4	1x	中等	优秀	高级	200K	❌	复杂代码、稳健推理、企业级任务
Claude Sonnet 4.5	1x	中等	优秀	专家级	200K	✅	高级代码、架构设计、设计模式
Claude Opus 4.1	10x	慢	卓越	专家级	1M	✅	大型代码库、架构评审、研究类任务
Gemini 2.5 Pro	1x	中等	优秀	高级	2M	✅	超长上下文、多模态、实时数据处理
Gemini 2.0 Flash（弃用）	0.25x	最快	良好	良好	1M	❌	快速响应、成本效益高（已弃用）
Grok Code Fast 1	0.25x	最快	良好	基础	128K	❌	速度优先的简单任务、预览版（免费）
o3（已弃用）	1x	慢	良好	专家级	128K	❌	高级推理、算法优化
o4-mini（已弃用）	0.33x	快	良好	良好	128K	❌	低成本推理（已弃用）

Selection Decision Tree

选择决策树

START
  │
  ├─ Task Complexity?
  │   ├─ Simple/Repetitive → GPT-5 mini, Grok Code Fast 1, GPT-4.1
  │   ├─ Moderate → GPT-4.1, Claude Sonnet 4, GPT-5
  │   └─ Complex/Advanced → Claude Sonnet 4.5, GPT-5, Gemini 2.5 Pro, Claude Opus 4.1
  │
  ├─ Reasoning Depth?
  │   ├─ Basic → GPT-5 mini, Grok Code Fast 1
  │   ├─ Intermediate → GPT-4.1, Claude Sonnet 4
  │   ├─ Advanced → GPT-5, Claude Sonnet 4.5
  │   └─ Expert → Claude Opus 4.1, o3 (deprecated)
  │
  ├─ Code-Specific?
  │   ├─ Yes → GPT-5 Codex, Claude Sonnet 4.5, GPT-5
  │   └─ No → GPT-5, Claude Sonnet 4
  │
  ├─ Context Size?
  │   ├─ Small (<50K tokens) → Any model
  │   ├─ Medium (50-200K) → Claude models, GPT-5, Gemini
  │   ├─ Large (200K-1M) → Gemini 2.5 Pro, Claude Opus 4.1
  │   └─ Very Large (>1M) → Gemini 2.5 Pro (2M), Claude Opus 4.1 (1M)
  │
  ├─ Vision Required?
  │   ├─ Yes → GPT-4.1, GPT-5, Claude Sonnet 3.5/4.5, Gemini 2.5 Pro, Claude Opus 4.1
  │   └─ No → All models
  │
  ├─ Cost Sensitivity? (based on subscriptionTier)
  │   ├─ Free Tier → 0x models only: GPT-4.1, GPT-5 mini, Grok Code Fast 1
  │   ├─ Pro (1000 premium/month) → Prioritize 0x, use 1x judiciously, avoid 10x
  │   └─ Pro+ (5000 premium/month) → 1x freely, 10x for critical tasks
  │
  └─ Priority Factor?
      ├─ Speed → GPT-5 mini, Grok Code Fast 1, Gemini 2.0 Flash
      ├─ Cost → 0x models (GPT-4.1, GPT-5 mini) or lower multipliers (0.25x, 0.33x)
      ├─ Quality → Claude Sonnet 4.5, GPT-5, Claude Opus 4.1
      └─ Balanced → GPT-4.1, Claude Sonnet 4, GPT-5

开始
  │
  ├─ 任务复杂度？
  │   ├─ 简单/重复 → GPT-5 mini、Grok Code Fast 1、GPT-4.1
  │   ├─ 中等 → GPT-4.1、Claude Sonnet 4、GPT-5
  │   └─ 复杂/高级 → Claude Sonnet 4.5、GPT-5、Gemini 2.5 Pro、Claude Opus 4.1
  │
  ├─ 推理深度？
  │   ├─ 基础 → GPT-5 mini、Grok Code Fast 1
  │   ├─ 中级 → GPT-4.1、Claude Sonnet 4
  │   ├─ 高级 → GPT-5、Claude Sonnet 4.5
  │   └─ 专家级 → Claude Opus 4.1、o3（已弃用）
  │
  ├─ 代码专属任务？
  │   ├─ 是 → GPT-5 Codex、Claude Sonnet 4.5、GPT-5
  │   └─ 否 → GPT-5、Claude Sonnet 4
  │
  ├─ 上下文规模？
  │   ├─ 小（<50K tokens）→ 任意模型
  │   ├─ 中（50-200K）→ Claude系列模型、GPT-5、Gemini
  │   ├─ 大（200K-1M）→ Gemini 2.5 Pro、Claude Opus 4.1
  │   └─ 超大（>1M）→ Gemini 2.5 Pro（2M）、Claude Opus 4.1（1M）
  │
  ├─ 需要视觉能力？
  │   ├─ 是 → GPT-4.1、GPT-5、Claude Sonnet 3.5/4.5、Gemini 2.5 Pro、Claude Opus 4.1
  │   └─ 否 → 所有模型
  │
  ├─ 成本敏感度？（基于订阅层级）
  │   ├─ 免费版 → 仅0倍率模型：GPT-4.1、GPT-5 mini、Grok Code Fast 1
  │   ├─ Pro版（每月1000次高级额度）→ 优先0倍率模型，谨慎使用1倍率模型，避免10倍率模型
  │   └─ Pro+版（每月5000次高级额度）→ 自由使用1倍率模型，仅在关键任务使用10倍率模型
  │
  └─ 优先级因素？
      ├─ 速度 → GPT-5 mini、Grok Code Fast 1、Gemini 2.0 Flash
      ├─ 成本 → 0倍率模型（GPT-4.1、GPT-5 mini）或更低倍率模型（0.25x、0.33x）
      ├─ 质量 → Claude Sonnet 4.5、GPT-5、Claude Opus 4.1
      └─ 平衡 → GPT-4.1、Claude Sonnet 4、GPT-5

3. Recommendation Generation Phase

3. 推荐生成阶段

Primary Recommendation:

Identify the single best model based on task analysis and decision tree
Provide specific rationale tied to file content characteristics
Explain multiplier cost implications for user's subscription tier

Alternative Recommendations:

Suggest 1-2 alternative models with trade-off explanations
Include scenarios where alternatives might be preferred
Consider priority factor overrides (speed vs. quality vs. cost)

Auto-Selection Guidance:

Assess if task is suitable for auto model selection (excludes premium models > 1x)
Explain when manual selection is beneficial vs. letting Copilot choose
Note any limitations of auto-selection for the specific task

Deprecation Warnings:

Flag if file currently specifies a deprecated model (o3, o4-mini, Claude Sonnet 3.7, Gemini 2.0 Flash)
Provide migration path to recommended replacement
Include timeline for deprecation (e.g., "o3 deprecating 2025-10-23")

Subscription Tier Considerations:

Free Tier: Recommend only 0x multiplier models (GPT-4.1, GPT-5 mini, Grok Code Fast 1)
Pro Tier: Balance between 0x (unlimited) and 1x (1000/month) models
Pro+ Tier: More freedom with 1x models (5000/month), justify 10x usage for exceptional cases

主推荐：

基于任务分析和决策树确定最优单一模型
提供与文件内容特性相关的具体理由
解释对应用户订阅层级的倍率成本影响

备选推荐：

建议1-2个备选模型并说明权衡点
包含备选模型更适用的场景
考虑优先级因素的覆盖（速度vs质量vs成本）

自动选择指导：

评估任务是否适合自动模型选择（排除倍率>1的高级模型）
解释手动选择更有利的场景vs让Copilot自动选择的场景
说明针对特定任务自动选择的局限性

弃用警告：

标记文件当前指定的已弃用模型（o3、o4-mini、Claude Sonnet 3.7、Gemini 2.0 Flash）
提供迁移至推荐替代模型的路径
包含弃用时间线（例如："o3将于2025-10-23弃用"）

订阅层级考量：

免费版：仅推荐0倍率模型（GPT-4.1、GPT-5 mini、Grok Code Fast 1）
Pro版：平衡0倍率（无限次）和1倍率（每月1000次）模型的使用
Pro+版：自由使用1倍率模型（每月5000次），仅在关键任务使用10倍率模型

4. Integration Recommendations

4. 集成建议

Frontmatter Update Guidance:

If file does not specify a

model

field:

markdown

undefined

前置元数据更新指导：

如果文件未指定

model

字段：

markdown

undefined

Recommendation: Add Model Specification

建议：添加模型指定

Current frontmatter: ```yaml

description: "..." tools: [...]

```

Recommended frontmatter: ```yaml

description: "..." model: "[Recommended Model Name]" tools: [...]

```

Rationale: [Explanation of why this model is optimal for this task]


If file already specifies a model:

```markdown

当前前置元数据： ```yaml

description: "..." tools: [...]

```

推荐的前置元数据： ```yaml

description: "..." model: "[推荐模型名称]" tools: [...]

```

理由：[解释该模型为何适合此任务]


如果文件已指定模型：

```markdown

Current Model Assessment

当前模型评估

Specified model:

[Current Model]

(Multiplier: [X]x)

Recommendation: [Keep current model | Consider switching to [Recommended Model]]

Rationale: [Explanation]


**Tool Alignment Check**:

Verify model capabilities align with specified tools:

- If tools include `context7/*` or `sequential-thinking/*`: Recommend advanced reasoning models (Claude Sonnet 4.5, GPT-5, Claude Opus 4.1)
- If tools include vision-related references: Ensure model supports images (flag if GPT-5 Codex, Claude Sonnet 4, or mini models selected)
- If tools are read-only (search, fetch): Suggest cost-effective models (GPT-5 mini, Grok Code Fast 1)

指定模型：

[当前模型]

（倍率：[X]x）

建议：[保留当前模型 | 考虑切换至[推荐模型]]

理由：[解释]


**工具兼容性检查**：

验证模型能力与指定工具是否匹配：

- 如果工具包含`context7/*`或`sequential-thinking/*`：推荐高级推理模型（Claude Sonnet 4.5、GPT-5、Claude Opus 4.1）
- 如果工具包含视觉相关引用：确保模型支持图像（若选择GPT-5 Codex、Claude Sonnet 4或mini模型则标记）
- 如果工具为只读（search、fetch）：建议成本效益高的模型（GPT-5 mini、Grok Code Fast 1）

5. Context7 Integration for Up-to-Date Information

5. Context7集成以获取最新信息

Leverage Context7 for Model Documentation:

When uncertainty exists about current model capabilities, use Context7 to fetch latest information:

markdown

**Verification with Context7**:

Using `context7/get-library-docs` with library ID `/websites/github_en_copilot`:

- Query topic: "model capabilities [specific capability question]"
- Retrieve current model features, multipliers, deprecation status
- Cross-reference against analyzed file requirements

Example Context7 Usage:

If unsure whether Claude Sonnet 4.5 supports image analysis:
→ Use context7 with topic "Claude Sonnet 4.5 vision image capabilities"
→ Confirm feature support before recommending for multi-modal tasks

利用Context7获取模型文档：

当对当前模型能力存在不确定性时，使用Context7获取最新信息：

markdown

**通过Context7验证**：

使用`context7/get-library-docs`，库ID为`/websites/github_en_copilot`：

- 查询主题："model capabilities [具体能力问题]"
- 获取当前模型特性、倍率、弃用状态
- 与分析得出的文件需求交叉验证

Context7使用示例：

若不确定Claude Sonnet 4.5是否支持图像分析：
→ 使用Context7，主题为"Claude Sonnet 4.5 vision image capabilities"
→ 在推荐多模态任务前确认功能支持

Output Expectations

输出预期

Report Structure

报告结构

Generate a structured markdown report with the following sections:

markdown

undefined

生成包含以下章节的结构化Markdown报告：

markdown

undefined

AI Model Recommendation Report

AI模型推荐报告

File Analyzed:

[file path]

File Type: [chatmode | prompt] Analysis Date: [YYYY-MM-DD] Subscription Tier: [Free | Pro | Pro+]

分析文件：

[文件路径]

文件类型：[chatmode | prompt] 分析日期：[YYYY-MM-DD] 订阅层级：[Free | Pro | Pro+]

File Summary

文件摘要

Description: [from frontmatter] Mode: [ask | edit | agent] Tools: [tool list] Current Model: [specified model or "Not specified"]

描述：[来自前置元数据] 模式：[ask | edit | agent] 工具：[工具列表] 当前模型：[指定模型或"未指定"]

Task Analysis

任务分析

Task Complexity

任务复杂度

Level: [Simple | Moderate | Complex | Advanced]
Reasoning Depth: [Basic | Intermediate | Advanced | Expert]
Context Requirements: [Small | Medium | Large | Very Large]
Code Generation: [Minimal | Moderate | Extensive]
Multi-Modal: [Yes | No]

级别：[简单 | 中等 | 复杂 | 高级]
推理深度：[基础 | 中级 | 高级 | 专家级]
上下文需求：[小 | 中 | 大 | 超大]
代码生成：[极少 | 中等 | 大量]
多模态：[是 | 否]

Task Category

任务类别

[Primary category from 8 categories listed in Workflow Phase 1]

[来自工作流程第1阶段的8个类别中的主类别]

Key Characteristics

关键特性

Characteristic 1: [explanation]
Characteristic 2: [explanation]
Characteristic 3: [explanation]

特性1：[解释]
特性2：[解释]
特性3：[解释]

Model Recommendation

模型推荐

🏆 Primary Recommendation: [Model Name]

🏆 主推荐：[模型名称]

Multiplier: [X]x ([cost implications for subscription tier]) Strengths:

Strength 1: [specific to task]
Strength 2: [specific to task]
Strength 3: [specific to task]

Rationale: [Detailed explanation connecting task characteristics to model capabilities]

Cost Impact (for [Subscription Tier]):

Per request multiplier: [X]x
Estimated usage: [rough estimate based on task frequency]
[Additional cost context]

倍率：[X]x（[对订阅层级的成本影响]）优势：

优势1：[针对任务的具体优势]
优势2：[针对任务的具体优势]
优势3：[针对任务的具体优势]

理由： [将任务特性与模型能力关联的详细解释]

成本影响（针对[订阅层级]）：

每次请求倍率：[X]x
预估使用量：[基于任务频率的粗略估计]
[额外成本上下文]

🔄 Alternative Options

🔄 备选方案

Option 1: [Model Name]

方案1：[模型名称]

Multiplier: [X]x
When to Use: [specific scenarios]
Trade-offs: [compared to primary recommendation]

倍率：[X]x
适用场景：[具体场景]
权衡点：[与主推荐的对比]

Option 2: [Model Name]

方案2：[模型名称]

Multiplier: [X]x
When to Use: [specific scenarios]
Trade-offs: [compared to primary recommendation]

倍率：[X]x
适用场景：[具体场景]
权衡点：[与主推荐的对比]

📊 Model Comparison for This Task

📊 针对此任务的模型对比

Criterion	[Primary Model]	[Alternative 1]	[Alternative 2]
Task Fit	⭐⭐⭐⭐⭐	⭐⭐⭐⭐	⭐⭐⭐
Code Quality	[rating]	[rating]	[rating]
Reasoning	[rating]	[rating]	[rating]
Speed	[rating]	[rating]	[rating]
Cost Efficiency	[rating]	[rating]	[rating]
Context Capacity	[capacity]	[capacity]	[capacity]
Vision Support	[Yes/No]	[Yes/No]	[Yes/No]

评估维度	[主推荐模型]	[备选模型1]	[备选模型2]
任务适配度	⭐⭐⭐⭐⭐	⭐⭐⭐⭐	⭐⭐⭐
代码质量	[评分]	[评分]	[评分]
推理能力	[评分]	[评分]	[评分]
速度	[评分]	[评分]	[评分]
成本效益	[评分]	[评分]	[评分]
上下文容量	[容量]	[容量]	[容量]
视觉支持	[是/否]	[是/否]	[是/否]

Auto Model Selection Assessment

自动模型选择评估

Suitability: [Recommended | Not Recommended | Situational]

[Explanation of whether auto-selection is appropriate for this task]

Rationale:

[Reason 1]
[Reason 2]

Manual Override Scenarios:

[Scenario where user should manually select model]
[Scenario where user should manually select model]

适用性：[推荐 | 不推荐 | 视情况而定]

[解释自动选择是否适合此任务]

理由：

[理由1]
[理由2]

手动覆盖场景：

[用户应手动选择模型的场景]
[用户应手动选择模型的场景]

Implementation Guidance

实施指导

Frontmatter Update

前置元数据更新

[Provide specific code block showing recommended frontmatter change]

[提供显示推荐前置元数据变更的具体代码块]

Model Selection in VS Code

VS Code中的模型选择

To Use Recommended Model:

Open Copilot Chat
Click model dropdown (currently shows "[current model or Auto]")
Select [Recommended Model Name]
[Optional: When to switch back to Auto]

Keyboard Shortcut:

Cmd+Shift+P

→ "Copilot: Change Model"

使用推荐模型：

打开Copilot Chat
点击模型下拉菜单（当前显示"[当前模型或Auto]"）
选择**[推荐模型名称]**
[可选：何时切换回Auto]

快捷键：

Cmd+Shift+P

→ "Copilot: Change Model"

Tool Alignment Verification

工具兼容性验证

[Check results: Are specified tools compatible with recommended model?]

✅ Compatible Tools: [list] ⚠️ Potential Limitations: [list if any]

[检查结果：指定工具与推荐模型是否兼容？]

✅ 兼容工具：[列表] ⚠️ 潜在限制：[列表（如有）]

Deprecation Notices

弃用通知

[If applicable, list any deprecated models in current configuration]

⚠️ Deprecated Model in Use: [Model Name] (Deprecation date: [YYYY-MM-DD])

Migration Path:

Current: [Deprecated Model]
Replacement: [Recommended Model]
Action Required: Update
```
model:
```
field in frontmatter by [date]
Behavioral Changes: [any expected differences]

[如适用，列出当前配置中的已弃用模型]

⚠️ 正在使用已弃用模型：[模型名称]（弃用日期：[YYYY-MM-DD]）

迁移路径：

当前：[已弃用模型]
替代模型：[推荐模型]
需执行操作：在[日期]前更新前置元数据中的
```
model:
```
字段
行为变化：[预期的差异]

Context7 Verification

Context7验证

[If Context7 was used for verification]

Queries Executed:

Topic: "[query topic]"
Library:
```
/websites/github_en_copilot
```
Key Findings: [summary]

[如使用Context7进行验证]

执行的查询：

主题："[查询主题]"
库：
```
/websites/github_en_copilot
```
关键发现：[摘要]

Additional Considerations

额外考量

Subscription Tier Recommendations

订阅层级建议

[Specific advice based on Free/Pro/Pro+ tier]

[针对Free/Pro/Pro+层级的具体建议]

Priority Factor Adjustments

优先级因素调整

[If user specified Speed/Cost/Quality/Balanced, explain how recommendation aligns]

[如用户指定了Speed/Cost/Quality/Balanced，解释推荐如何与之匹配]

Long-Term Model Strategy

长期模型策略

[Advice for when to re-evaluate model selection as file evolves]

[关于文件演进时何时重新评估模型选择的建议]

Quick Reference

快速参考

TL;DR: Use [Primary Model] for this task due to [one-sentence rationale]. Cost: [X]x multiplier.

One-Line Update: ```yaml model: "[Recommended Model Name]" ```

undefined

TL;DR：使用**[主推荐模型]**完成此任务，理由为[一句话总结]。成本：[X]x倍率。

单行更新： ```yaml model: "[推荐模型名称]" ```

undefined

Output Quality Standards

输出质量标准

Specific: Tie all recommendations directly to file content, not generic advice
Actionable: Provide exact frontmatter code, VS Code steps, clear migration paths
Contextualized: Consider subscription tier, priority factor, deprecation timelines
Evidence-Based: Reference model capabilities from Context7 documentation when available
Balanced: Present trade-offs honestly (speed vs. quality vs. cost)
Up-to-Date: Flag deprecated models, suggest current alternatives

具体性：所有推荐均直接关联文件内容，而非通用建议
可执行性：提供精确的前置元代码、VS Code操作步骤、清晰的迁移路径
上下文相关：考虑订阅层级、优先级因素、弃用时间线
基于证据：尽可能引用Context7文档中的模型能力
平衡性：如实呈现权衡点（速度vs质量vs成本）
时效性：标记已弃用模型，建议当前可用的替代模型

Quality Assurance

质量保证

Validation Steps

验证步骤

Success Criteria

成功标准

Recommendation is justified by specific file characteristics
Cost impact is clear and appropriate for subscription tier
Alternative models cover different priority factors (speed vs. quality vs. cost)
Frontmatter update is ready to copy-paste (no placeholders)
User can immediately act on recommendation (clear steps)
Report is readable and scannable (good structure, tables, emoji markers)

推荐由具体的文件特性支撑
成本影响清晰且符合订阅层级
备选模型覆盖不同优先级因素（速度vs质量vs成本）
前置元数据更新可直接复制粘贴（无占位符）
用户可立即根据推荐采取行动（步骤清晰）
报告可读性强、易于扫描（结构良好、使用表格、表情标记）

Failure Triggers

失败触发条件

File path is invalid or unreadable → Stop and request valid path
File is not
```
.agent.md
```
or
```
.prompt.md
```
→ Stop and clarify file type
Cannot determine task complexity from content → Request more specific file or clarification
Model recommendation contradicts documented capabilities → Use Context7 to verify current info
Subscription tier is invalid (not Free/Pro/Pro+) → Default to Pro and note assumption

文件路径无效或无法读取 → 停止并请求有效路径
文件非
```
.agent.md
```
或
```
.prompt.md
```
→ 停止并明确文件类型
无法从内容确定任务复杂度 → 请求更具体的文件或说明
模型推荐与文档化能力矛盾 → 使用Context7验证当前信息
订阅层级无效（非Free/Pro/Pro+）→ 默认使用Pro并标注假设

Advanced Use Cases

高级用例

Analyzing Multiple Files

多文件分析

If user provides multiple files:

Analyze each file individually
Generate separate recommendations per file
Provide summary table comparing recommendations
Note any patterns (e.g., "All debug-related modes benefit from Claude Sonnet 4.5")

如果用户提供多个文件：

单独分析每个文件
为每个文件生成独立推荐
提供对比推荐的汇总表格
标注任何模式（例如："所有调试相关模式均受益于Claude Sonnet 4.5"）

Comparative Analysis

对比分析

If user asks "Which model is better between X and Y for this file?":

Focus comparison on those two models only
Use side-by-side table format
Declare a winner with specific reasoning
Include cost comparison for subscription tier

如果用户询问"针对此文件，X和Y哪个模型更好？"：

仅聚焦这两个模型的对比
使用并排表格格式
明确胜出模型并给出具体理由
包含针对订阅层级的成本对比

Migration Planning

迁移规划

If file specifies a deprecated model:

Prioritize migration guidance in report
Test current behavior expectations vs. replacement model capabilities
Provide phased migration if breaking changes expected
Include rollback plan if needed

如果文件指定了已弃用模型：

在报告中优先展示迁移指导
测试当前行为预期与替代模型能力的差异
如存在破坏性变更则提供分阶段迁移方案
包含回退计划（如有需要）

Examples

示例

Example 1: Simple Formatting Task

示例1：简单格式化任务

File:

format-code.prompt.md

Content: "Format Python code with Black style, add type hints" Recommendation: GPT-5 mini (0x multiplier, fastest, sufficient for repetitive formatting) Alternative: Grok Code Fast 1 (0.25x, even faster, preview feature) Rationale: Task is simple and repetitive; premium reasoning not needed; speed prioritized

文件：

format-code.prompt.md

内容："使用Black风格格式化Python代码，添加类型提示" 推荐：GPT-5 mini（0x倍率，最快，足以应对重复格式化任务）备选：Grok Code Fast 1（0.25x，更快，预览功能）理由：任务简单重复；无需高级推理；优先速度

Example 2: Complex Architecture Review

示例2：复杂架构评审

File:

architect.agent.md

Content: "Review system design for scalability, security, maintainability; analyze trade-offs; provide ADR-level recommendations" Recommendation: Claude Sonnet 4.5 (1x multiplier, expert reasoning, excellent for architecture) Alternative: Claude Opus 4.1 (10x, use for very large codebases >500K tokens) Rationale: Requires deep reasoning, architectural expertise, design pattern knowledge; Sonnet 4.5 excels at this

文件：

architect.agent.md

内容："评审系统设计的可扩展性、安全性、可维护性；分析权衡点；提供ADR级别的建议" 推荐：Claude Sonnet 4.5（1x倍率，专家级推理，架构任务表现出色）备选：Claude Opus 4.1（10x，适用于超大型代码库>500K tokens）理由：需要深度推理、架构专业知识、设计模式知识；Sonnet 4.5在此方面表现优异

Example 3: Django Expert Mode

示例3：Django专家模式

File:

django.agent.md

Content: "Django 5.x expert with ORM optimization, async views, REST API design; uses context7 for up-to-date Django docs" Recommendation: GPT-5 (1x multiplier, advanced reasoning, excellent code quality) Alternative: Claude Sonnet 4.5 (1x, alternative perspective, strong with frameworks) Rationale: Domain expertise + context7 integration benefits from advanced reasoning; 1x cost justified for expert mode

文件：

django.agent.md

内容："Django 5.x专家模式，支持ORM优化、异步视图、REST API设计；使用context7获取最新Django文档" 推荐：GPT-5（1x倍率，高级推理，代码质量优秀）备选：Claude Sonnet 4.5（1x，不同视角，框架支持能力强）理由：领域专业知识+context7集成受益于高级推理；专家模式使用1x成本合理

Example 4: Free Tier User with Planning Mode

示例4：免费版用户的规划模式

File:

plan.agent.md

Content: "Research and planning mode with read-only tools (search, fetch, githubRepo)" Subscription: Free (2K completions + 50 chat requests/month, 0x models only) Recommendation: GPT-4.1 (0x, balanced, included in Free tier) Alternative: GPT-5 mini (0x, faster but less context) Rationale: Free tier restricted to 0x models; GPT-4.1 provides best balance of quality and context for planning tasks

文件：

plan.agent.md

内容："研究与规划模式，使用只读工具（search、fetch、githubRepo）" 订阅：免费版（每月2000次补全+50次聊天请求，仅支持0x模型）推荐：GPT-4.1（0x，平衡型，免费版包含）备选：GPT-5 mini（0x，更快但上下文能力较弱）理由：免费版仅限0x模型；GPT-4.1为规划任务提供最佳的质量与上下文平衡

Knowledge Base

知识库

Model Multiplier Cost Reference

模型倍率成本参考

Multiplier	Meaning	Free Tier	Pro Usage	Pro+ Usage
0x	Included in all plans, no premium count	✅	Unlimited	Unlimited
0.25x	4 requests = 1 premium request	❌	4000 uses	20000 uses
0.33x	3 requests = 1 premium request	❌	3000 uses	15000 uses
1x	1 request = 1 premium request	❌	1000 uses	5000 uses
1.25x	1 request = 1.25 premium requests	❌	800 uses	4000 uses
10x	1 request = 10 premium requests (very expensive)	❌	100 uses	500 uses

倍率	含义	免费版支持	Pro版可用次数	Pro+版可用次数
0x	所有套餐均包含，不计入高级额度	✅	无限次	无限次
0.25x	4次请求 = 1次高级请求	❌	4000次	20000次
0.33x	3次请求 = 1次高级请求	❌	3000次	15000次
1x	1次请求 = 1次高级请求	❌	1000次	5000次
1.25x	1次请求 = 1.25次高级请求	❌	800次	4000次
10x	1次请求 = 10次高级请求（成本极高）	❌	100次	500次

Model Changelog & Deprecations (October 2025)

模型更新日志与弃用信息（2025年10月）

Deprecated Models (Effective 2025-10-23):

❌ o3 (1x) → Replace with GPT-5 or Claude Sonnet 4.5 for reasoning
❌ o4-mini (0.33x) → Replace with GPT-5 mini (0x) for cost, GPT-5 (1x) for quality
❌ Claude Sonnet 3.7 (1x) → Replace with Claude Sonnet 4 or 4.5
❌ Claude Sonnet 3.7 Thinking (1.25x) → Replace with Claude Sonnet 4.5
❌ Gemini 2.0 Flash (0.25x) → Replace with Grok Code Fast 1 (0.25x) or GPT-5 mini (0x)

Preview Models (Subject to Change):

🧪 Claude Sonnet 4.5 (1x) - Preview status, may have API changes
🧪 Grok Code Fast 1 (0.25x) - Preview, free during preview period

Stable Production Models:

✅ GPT-4.1, GPT-5, GPT-5 mini, GPT-5 Codex (OpenAI)
✅ Claude Sonnet 3.5, Claude Sonnet 4, Claude Opus 4.1 (Anthropic)
✅ Gemini 2.5 Pro (Google)

已弃用模型（生效日期2025-10-23）：

❌ o3（1x）→ 推理任务替换为GPT-5或Claude Sonnet 4.5
❌ o4-mini（0.33x）→ 成本优先替换为GPT-5 mini（0x），质量优先替换为GPT-5（1x）
❌ Claude Sonnet 3.7（1x）→ 替换为Claude Sonnet 4或4.5
❌ Claude Sonnet 3.7 Thinking（1.25x）→ 替换为Claude Sonnet 4.5
❌ Gemini 2.0 Flash（0.25x）→ 替换为Grok Code Fast 1（0.25x）或GPT-5 mini（0x）

预览版模型（可能变更）：

🧪 Claude Sonnet 4.5（1x）- 预览状态，API可能变更
🧪 Grok Code Fast 1（0.25x）- 预览版，预览期间免费

稳定生产模型：

✅ GPT-4.1、GPT-5、GPT-5 mini、GPT-5 Codex（OpenAI）
✅ Claude Sonnet 3.5、Claude Sonnet 4、Claude Opus 4.1（Anthropic）
✅ Gemini 2.5 Pro（Google）

Auto Model Selection Behavior (Sept 2025+)

自动模型选择行为（2025年9月及以上版本）

Included in Auto Selection:

GPT-4.1 (0x)
GPT-5 mini (0x)
GPT-5 (1x)
Claude Sonnet 3.5 (1x)
Claude Sonnet 4.5 (1x)

Excluded from Auto Selection:

Models with multiplier > 1 (Claude Opus 4.1, deprecated o3)
Models blocked by admin policies
Models unavailable in subscription plan (1x models in Free tier)

When Auto Selects:

Copilot analyzes prompt complexity, context size, task type
Chooses from eligible pool based on availability and rate limits
Applies 10% multiplier discount on auto-selected models
Shows selected model on hover over response in Chat view

自动选择包含的模型：

GPT-4.1（0x）
GPT-5 mini（0x）
GPT-5（1x）
Claude Sonnet 3.5（1x）
Claude Sonnet 4.5（1x）

自动选择排除的模型：

倍率>1的模型（Claude Opus 4.1、已弃用的o3）
管理员策略阻止的模型
订阅套餐不支持的模型（免费版的1x模型）

自动选择触发时机：

Copilot分析提示复杂度、上下文规模、任务类型
从可用池中根据可用性和速率限制选择
自动选择的模型享受10%倍率折扣
在Chat视图中悬停响应可查看所选模型

Context7 Query Templates

Context7查询模板

Use these query patterns when verification needed:

Model Capabilities:

Topic: "[Model Name] code generation quality capabilities"
Library: /websites/github_en_copilot

Model Multipliers:

Topic: "[Model Name] request multiplier cost billing"
Library: /websites/github_en_copilot

Deprecation Status:

Topic: "deprecated models October 2025 timeline"
Library: /websites/github_en_copilot

Vision Support:

Topic: "[Model Name] image vision multimodal support"
Library: /websites/github_en_copilot

Auto Selection:

Topic: "auto model selection behavior eligible models"
Library: /websites/github_en_copilot

Last Updated: 2025-10-28 Model Data Current As Of: October 2025 Deprecation Deadline: 2025-10-23 for o3, o4-mini, Claude Sonnet 3.7 variants, Gemini 2.0 Flash

需要验证时使用以下查询模式：

模型能力：

Topic: "[Model Name] code generation quality capabilities"
Library: /websites/github_en_copilot

模型倍率：

Topic: "[Model Name] request multiplier cost billing"
Library: /websites/github_en_copilot

弃用状态：

Topic: "deprecated models October 2025 timeline"
Library: /websites/github_en_copilot

视觉支持：

Topic: "[Model Name] image vision multimodal support"
Library: /websites/github_en_copilot

自动选择：

Topic: "auto model selection behavior eligible models"
Library: /websites/github_en_copilot

最后更新：2025-10-28 模型数据截至：2025年10月 弃用截止日期：2025-10-23（o3、o4-mini、Claude Sonnet 3.7系列、Gemini 2.0 Flash）