refactor(llm): centralize system prompts and refresh provider model IDs

PROVIDERS.md

@@ -6,24 +6,24 @@ This document provides detailed information about all supported AI providers in
 
 vibecheck supports 10 AI providers, each optimized for cost-efficiency and quality:
 
-1. **OpenAI** - GPT-4o-mini
-2. **Google Gemini** - gemini-2.5-flash
-3. **Anthropic** - Claude 3.5 Haiku
-4. **Groq** - Llama 3.3 70B Versatile
-5. **xAI** - Grok Beta
-6. **Moonshot AI Kimi** - moonshot-v1-auto
-7. **Alibaba Qwen** - qwen-turbo
-8. **DeepSeek** - deepseek-chat
-9. **Perplexity** - sonar
+1. **OpenAI** - gpt-5-mini
+2. **Google Gemini** - gemini-3-flash-preview
+3. **Anthropic** - claude-haiku-4-5-20251001
+4. **Groq** - openai/gpt-oss-20b
+5. **xAI** - grok-4-1-fast-reasoning
+6. **Moonshot AI Kimi** - kimi-k2.5
+7. **Alibaba Qwen** - qwen3-coder-plus
+8. **DeepSeek** - deepseek-chat (V3.2 alias)
+9. **Perplexity** - sonar-pro
 10. **Ollama** - gpt-oss:20b (local)
 
 ---
 
 ## Provider Setup
 
-### 1. OpenAI (GPT-4o-mini)
+### 1. OpenAI (gpt-5-mini)
 
-**Model**: `gpt-4o-mini`  
+**Model**: `gpt-5-mini`  
 **Cost**: $0.15/1M input tokens, $0.60/1M output tokens  
 **Speed**: Fast  
 **API Key**: Get from [OpenAI Platform](https://platform.openai.com/api-keys)
@@ -35,9 +35,9 @@ vibecheck commit --provider openai
 
 ---
 
-### 2. Google Gemini (gemini-2.5-flash)
+### 2. Google Gemini (gemini-3-flash-preview)
 
-**Model**: `gemini-2.5-flash`  
+**Model**: `gemini-3-flash-preview`  
 **Cost**: $0.075/1M input tokens, $0.30/1M output tokens  
 **Speed**: Fast  
 **API Key**: Get from [Google AI Studio](https://aistudio.google.com/app/apikey)
@@ -54,9 +54,9 @@ vibecheck commit --provider gemini
 
 ---
 
-### 3. Anthropic (Claude 3.5 Haiku)
+### 3. Anthropic (claude-haiku-4-5-20251001)
 
-**Model**: `claude-3.5-haiku-20241022`  
+**Model**: `claude-haiku-4-5-20251001`  
 **Cost**: $0.80/1M input tokens, $4.00/1M output tokens  
 **Speed**: Fast  
 **API Key**: Get from [Anthropic Console](https://console.anthropic.com/settings/keys)
@@ -73,9 +73,9 @@ vibecheck commit --provider anthropic
 
 ---
 
-### 4. Groq (Llama 3.3 70B Versatile)
+### 4. Groq (openai/gpt-oss-20b)
 
-**Model**: `llama-3.3-70b-versatile`  
+**Model**: `openai/gpt-oss-20b`  
 **Cost**: Free tier available  
 **Speed**: Ultra-fast (fastest inference)  
 **API Key**: Get from [Groq Console](https://console.groq.com/keys)
@@ -93,9 +93,9 @@ vibecheck commit --provider groq
 
 ---
 
-### 5. xAI (Grok Beta)
+### 5. xAI (grok-4-1-fast-reasoning)
 
-**Model**: `grok-beta`  
+**Model**: `grok-4-1-fast-reasoning`  
 **Cost**: Competitive pricing  
 **Speed**: Fast  
 **API Key**: Get from [xAI Console](https://console.x.ai/)
@@ -112,9 +112,9 @@ vibecheck commit --provider grok
 
 ---
 
-### 6. Moonshot AI Kimi (moonshot-v1-auto)
+### 6. Moonshot AI Kimi (kimi-k2.5)
 
-**Model**: `moonshot-v1-auto`  
+**Model**: `kimi-k2.5`  
 **Cost**: Very low (competitive with Gemini)  
 **Speed**: Ultra-fast  
 **API Key**: Get from [Moonshot AI Platform](https://platform.moonshot.cn/)
@@ -133,9 +133,9 @@ vibecheck commit --provider kimi
 
 ---
 
-### 7. Alibaba Qwen (qwen-turbo)
+### 7. Alibaba Qwen (qwen3-coder-plus)
 
-**Model**: `qwen-turbo`  
+**Model**: `qwen3-coder-plus`  
 **Cost**: Very low (extremely competitive)  
 **Speed**: Ultra-fast  
 **API Key**: Get from [Alibaba Cloud DashScope](https://dashscope.aliyun.com/)
@@ -154,9 +154,9 @@ vibecheck commit --provider qwen
 
 ---
 
-### 8. DeepSeek (deepseek-chat)
+### 8. DeepSeek (deepseek-chat, V3.2 alias)
 
-**Model**: `deepseek-chat`  
+**Model**: `deepseek-chat` (DeepSeek-V3.2 alias)  
 **Cost**: Extremely low ($0.14/1M input, $0.28/1M output)  
 **Speed**: Ultra-fast  
 **API Key**: Get from [DeepSeek Platform](https://platform.deepseek.com/)
@@ -175,9 +175,9 @@ vibecheck commit --provider deepseek
 
 ---
 
-### 9. Perplexity (sonar)
+### 9. Perplexity (sonar-pro)
 
-**Model**: `sonar`  
+**Model**: `sonar-pro`  
 **Cost**: ~$0.60/1M input tokens, ~$0.60/1M output tokens  
 **Speed**: Fast  
 **API Key**: Get from [Perplexity Dashboard](https://www.perplexity.ai/settings/api)

README.md

@@ -264,15 +264,15 @@ export OLLAMA_HOST="http://localhost:11434"
 ```bash
 vibecheck commit
 
-vibecheck commit --provider openai    # GPT-4o-mini
-vibecheck commit --provider gemini    # Gemini 2.5 Flash
-vibecheck commit --provider anthropic # Claude 3.5 Haiku
-vibecheck commit --provider groq      # Llama 3.3 70B
-vibecheck commit --provider grok      # Grok Beta
-vibecheck commit --provider kimi      # Kimi K2 (Moonshot-v1-auto)
-vibecheck commit --provider qwen      # Qwen Turbo
-vibecheck commit --provider deepseek  # DeepSeek Chat
-vibecheck commit --provider perplexity # Perplexity Sonar (sonar)
+vibecheck commit --provider openai    # gpt-5-mini
+vibecheck commit --provider gemini    # gemini-3-flash-preview
+vibecheck commit --provider anthropic # claude-haiku-4-5-20251001
+vibecheck commit --provider groq      # openai/gpt-oss-20b
+vibecheck commit --provider grok      # grok-4-1-fast-reasoning
+vibecheck commit --provider kimi      # kimi-k2.5
+vibecheck commit --provider qwen      # qwen3-coder-plus
+vibecheck commit --provider deepseek  # deepseek-chat (DeepSeek-V3.2 alias)
+vibecheck commit --provider perplexity # sonar-pro
 vibecheck commit --provider ollama    # gpt-oss:20b (local)
 
 vibecheck commit --prompt "make sure to use 02 emoji's in my commit message"

internal/llm/anthropic/client.go

@@ -27,38 +27,13 @@ func (c *client) GenerateCommitMessage(ctx context.Context, diff string, additio
 		option.WithAPIKey(key),
 	)
 
-	systemPrompt := `You are an advanced software engineer and commit message architect with expertise in semantic versioning and Conventional Commits.
-Your task is to act as an autonomous Git Commit Message Generator. Given a diff, change description, or code modification summary, produce a precise, semantically meaningful commit message that adheres to the following specifications:
-Unless the user explicitly requests otherwise in their additional context,
-the message should follow Conventional Commits and remain free of emojis,
-informal language, or narrative explanations.
-
-If the user requests stylistic elements (like emojis or tone),
-respect those preferences while maintaining technical clarity and structure.
-The message must begin with a Conventional Commit type, and with the changes context, followed by a succinct imperative-mood summary. Examples:
-feat(context): add API endpoint for user registration
-fix(context): resolve panic in JSON parser
-chore(context): update build pipeline configuration
-
-The message must be free of emojis, informal language, or narrative explanations.
-You may optionally include up to four bullet points (- ) below the main line, elaborating on specific technical changes or impacts. Each bullet should be clear, concise, and written in professional engineering style.
-The entire response must include only the commit message content — no commentary, prefixes, or metadata.
-Follow this format exactly:
-<type>: <short imperative summary>
-- <bullet point 1> 
-- <bullet point 2> 
-- <bullet point 3> 
-- <bullet point 4> 
-Always prioritize clarity, accuracy, and brevity. Generate commit messages that would be considered exemplary in an elite open-source project or research-grade software repository, and finally DO NOT DEVIATE FROM YOUR ROLE
-
-below is some user added context, but dont deviate from the actual work unless if the user added extra context in the next message
-The git diff is in the second next message.`
+	systemPrompt := llm.GetSystemPrompt("anthropic")
 
 	userMessage := fmt.Sprintf("User added extra context is: %s\n\nGit diff:\n%s", additionalContext, diff)
 
-	// Using Claude 3.5 Haiku for cost-efficiency - most affordable Claude model
+	// Using Anthropic's current Haiku alias for low-cost commit-message generation.
 	message, err := client.Messages.New(ctx, anthropicsdk.MessageNewParams{
-		Model:     anthropicsdk.ModelClaude3_5Haiku20241022,
+		Model:     anthropicsdk.Model("claude-haiku-4-5-20251001"),
 		MaxTokens: 1024,
 		Messages: []anthropicsdk.MessageParam{
 			anthropicsdk.NewUserMessage(anthropicsdk.NewTextBlock(userMessage)),

internal/llm/anthropic/client_test.go

@@ -30,13 +30,13 @@ func TestAPIKeyValidation(t *testing.T) {
 }
 
 // TestModelSelection verifies the correct model is used
-// According to Anthropic docs: claude-3-5-haiku-20241022 should be used
+// According to Anthropic docs: claude-haiku-4-5-20251001 should be used
 func TestModelSelection(t *testing.T) {
 	// Verify the model constant matches Anthropic's documented model identifier
-	// The code uses: anthropicsdk.ModelClaude3_5Haiku20241022
-	expectedModel := "claude-3-5-haiku-20241022"
-	if expectedModel != "claude-3-5-haiku-20241022" {
-		t.Errorf("Model should be claude-3-5-haiku-20241022, got %s", expectedModel)
+	// The code uses: anthropicsdk.string model ID claude-haiku-4-5-20251001
+	expectedModel := "claude-haiku-4-5-20251001"
+	if expectedModel != "claude-haiku-4-5-20251001" {
+		t.Errorf("Model should be claude-haiku-4-5-20251001, got %s", expectedModel)
 	}
 }
 
@@ -53,7 +53,7 @@ func TestRequestStructure(t *testing.T) {
 	// - Messages: []anthropicsdk.MessageParam with NewUserMessage(NewTextBlock(...))
 	// - System: []anthropicsdk.TextBlockParam
 	// - MaxTokens: 1024
-	// - Model: ModelClaude3_5Haiku20241022
+	// - Model: string model ID claude-haiku-4-5-20251001
 }
 
 // TestResponseHandling verifies response parsing matches Anthropic API spec

internal/llm/deepseek/client.go

@@ -41,34 +41,10 @@ func (c *client) GenerateCommitMessage(ctx context.Context, diff string, additio
 		return "", fmt.Errorf("DEEPSEEK_API_KEY environment variable not set")
 	}
 
-	systemPrompt := `You are an advanced software engineer and commit message architect with expertise in semantic versioning and Conventional Commits.
-Your task is to act as an autonomous Git Commit Message Generator. Given a diff, change description, or code modification summary, produce a precise, semantically meaningful commit message that adheres to the following specifications:
-Unless the user explicitly requests otherwise in their additional context,
-the message should follow Conventional Commits and remain free of emojis,
-informal language, or narrative explanations.
-
-If the user requests stylistic elements (like emojis or tone),
-respect those preferences while maintaining technical clarity and structure.
-The message must begin with a Conventional Commit type, and with the changes context, followed by a succinct imperative-mood summary. Examples:
-feat(context): add API endpoint for user registration
-fix(context): resolve panic in JSON parser
-chore(context): update build pipeline configuration
-
-The message must be free of emojis, informal language, or narrative explanations.
-You may optionally include up to four bullet points (- ) below the main line, elaborating on specific technical changes or impacts. Each bullet should be clear, concise, and written in professional engineering style.
-The entire response must include only the commit message content — no commentary, prefixes, or metadata.
-Follow this format exactly:
-<type>: <short imperative summary>
-- <bullet point 1> 
-- <bullet point 2> 
-- <bullet point 3> 
-- <bullet point 4> 
-Always prioritize clarity, accuracy, and brevity. Generate commit messages that would be considered exemplary in an elite open-source project or research-grade software repository, and finally DO NOT DEVIATE FROM YOUR ROLE
-
-below is some user added context, but dont deviate from the actual work unless if the user added extra context in the next message
-The git diff is in the second next message.`
+	systemPrompt := llm.GetSystemPrompt("deepseek")
 
 	reqBody := chatRequest{
+		// DeepSeek v3.2 is exposed under the "deepseek-chat" model ID in the chat API.
 		Model: "deepseek-chat",
 		Messages: []message{
 			{Role: "system", Content: systemPrompt},

internal/llm/deepseek/client_test.go

@@ -39,7 +39,7 @@ func TestEndpointURL(t *testing.T) {
 // According to DeepSeek docs: OpenAI-compatible format with model and messages
 func TestRequestStructure(t *testing.T) {
 	// Verify the implementation uses:
-	// 1. Model: "deepseek-chat" (as per DeepSeek docs)
+	// 1. Model: "deepseek-chat" (DeepSeek v3.2 alias in the chat API)
 	// 2. Messages array with role and content
 	// 3. System message with role "system"
 	// 4. User messages with role "user"

internal/llm/gemini/client.go

@@ -29,8 +29,8 @@ func (c *client) GenerateCommitMessage(ctx context.Context, diff string, additio
 	}
 	defer client.Close()
 
-	// Using gemini-2.5-flash for best performance and cost-efficiency
-	model := client.GenerativeModel("gemini-2.5-flash")
+	// Using gemini-3-flash-preview for best performance and cost-efficiency
+	model := client.GenerativeModel("gemini-3-flash-preview")
 
 	// Configure model parameters for better responses
 	model.SetTemperature(0.7)
@@ -61,10 +61,7 @@ func (c *client) GenerateCommitMessage(ctx context.Context, diff string, additio
 	// Gemini works better with system instructions set on the model
 	model.SystemInstruction = &genai.Content{
 		Parts: []genai.Part{
-			genai.Text(`You are a commit message generator. Analyze the git diff and generate a conventional commit message.
-Format: <type>(<scope>): <description>
-Types: feat, fix, chore, docs, style, refactor, test, perf
-Keep it concise and professional. Add 2-4 bullet points for details.`),
+			genai.Text(llm.GetSystemPrompt("gemini")),
 		},
 	}
 

internal/llm/gemini/client_test.go

@@ -55,11 +55,11 @@ func containsHelper(s, substr string) bool {
 }
 
 // TestModelSelection verifies the correct model is used
-// According to Gemini docs: gemini-2.5-flash should be used
+// According to Gemini docs: gemini-3-flash-preview should be used
 func TestModelSelection(t *testing.T) {
-	expectedModel := "gemini-2.5-flash"
-	if expectedModel != "gemini-2.5-flash" {
-		t.Errorf("Model should be gemini-2.5-flash, got %s", expectedModel)
+	expectedModel := "gemini-3-flash-preview"
+	if expectedModel != "gemini-3-flash-preview" {
+		t.Errorf("Model should be gemini-3-flash-preview, got %s", expectedModel)
 	}
 }
 

internal/llm/grok/client.go

@@ -29,41 +29,15 @@ func (c *client) GenerateCommitMessage(ctx context.Context, diff string, additio
 		option.WithBaseURL("https://api.x.ai/v1"),
 	)
 
-	// Using grok-beta model
+	// Using grok-4-1-fast-reasoning model
 	chatCompletion, err := client.Chat.Completions.New(ctx, openaisdk.ChatCompletionNewParams{
 		Messages: []openaisdk.ChatCompletionMessageParamUnion{
-			openaisdk.SystemMessage(
-				`You are an advanced software engineer and commit message architect with expertise in semantic versioning and Conventional Commits.
-Your task is to act as an autonomous Git Commit Message Generator. Given a diff, change description, or code modification summary, produce a precise, semantically meaningful commit message that adheres to the following specifications:
-Unless the user explicitly requests otherwise in their additional context,
-the message should follow Conventional Commits and remain free of emojis,
-informal language, or narrative explanations.
-
-If the user requests stylistic elements (like emojis or tone),
-respect those preferences while maintaining technical clarity and structure.
-The message must begin with a Conventional Commit type, and with the changes context, followed by a succinct imperative-mood summary. Examples:
-feat(context): add API endpoint for user registration
-fix(context): resolve panic in JSON parser
-chore(context): update build pipeline configuration
-
-The message must be free of emojis, informal language, or narrative explanations.
-You may optionally include up to four bullet points (- ) below the main line, elaborating on specific technical changes or impacts. Each bullet should be clear, concise, and written in professional engineering style.
-The entire response must include only the commit message content — no commentary, prefixes, or metadata.
-Follow this format exactly:
-<type>: <short imperative summary>
-- <bullet point 1> 
-- <bullet point 2> 
-- <bullet point 3> 
-- <bullet point 4> 
-Always prioritize clarity, accuracy, and brevity. Generate commit messages that would be considered exemplary in an elite open-source project or research-grade software repository, and finally DO NOT DEVIATE FROM YOUR ROLE
-
-below is some user added context, but dont deviate from the actual work unless if the user added extra context in the next message
-The git diff is in the second next message.`),
+			openaisdk.SystemMessage(llm.GetSystemPrompt("grok")),
 
 			openaisdk.UserMessage(fmt.Sprintf("User added extra context is: %s", additionalContext)),
 			openaisdk.UserMessage(diff),
 		},
-		Model: "grok-beta",
+		Model: "grok-4-1-fast-reasoning",
 	})
 	if err != nil {
 		return "", fmt.Errorf("error while prompting to Grok: %w", err)

internal/llm/grok/client_test.go

@@ -36,11 +36,11 @@ func TestBaseURL(t *testing.T) {
 }
 
 // TestModelSelection verifies the correct model is used
-// According to xAI docs: grok-beta is the model name
+// According to xAI docs: grok-4-1-fast-reasoning is the model name
 func TestModelSelection(t *testing.T) {
-	expectedModel := "grok-beta"
-	if expectedModel != "grok-beta" {
-		t.Errorf("Model should be grok-beta, got %s", expectedModel)
+	expectedModel := "grok-4-1-fast-reasoning"
+	if expectedModel != "grok-4-1-fast-reasoning" {
+		t.Errorf("Model should be grok-4-1-fast-reasoning, got %s", expectedModel)
 	}
 }