feat(llm): add prompt caching configuration

cmd/opencodereview/config_cmd.go

@@ -176,7 +176,7 @@ func unsetActiveProvider(configPath string) error {
 }
 
 func legacyLLMShadowWarning(provider, key string) string {
-	if provider == "" || !strings.HasPrefix(key, "llm.") {
+	if provider == "" || !strings.HasPrefix(key, "llm.") || key == "llm.prompt_caching" || key == "llm.PromptCaching" {
 		return ""
 	}
 	section := "custom_providers"
@@ -309,6 +309,8 @@ type LlmConfig struct {
 	TimeoutSec   int               `json:"timeout_sec,omitempty"`   // per-request HTTP timeout in seconds
 	ExtraBody    map[string]any    `json:"extra_body,omitempty"`
 	ExtraHeaders map[string]string `json:"extra_headers,omitempty"`
+	// PromptCaching is the global Anthropic caching toggle for the active provider.
+	PromptCaching *bool `json:"prompt_caching,omitempty"`
 }
 
 // TelemetryConfig holds telemetry-specific settings.
@@ -365,6 +367,7 @@ var supportedConfigKeys = []string{
 	"llm.model",
 	"llm.protocol",
 	"llm.use_anthropic",
+	"llm.prompt_caching",
 	"llm.extra_body",
 	"llm.extra_headers",
 	"language",
@@ -476,6 +479,12 @@ func setConfigValue(cfg *Config, key, value string) error {
 		} else if cfg.Llm.Protocol == "" || cfg.Llm.Protocol == llm.ProtocolAnthropic || cfg.Llm.Protocol == llm.ProtocolOpenAIChatCompletions {
 			cfg.Llm.Protocol = llm.ProtocolOpenAIChatCompletions
 		}
+	case "llm.prompt_caching", "llm.PromptCaching":
+		b, err := strconv.ParseBool(value)
+		if err != nil {
+			return fmt.Errorf("invalid boolean for llm.prompt_caching: %w", err)
+		}
+		cfg.Llm.PromptCaching = &b
 	case "language", "Language":
 		cfg.Language = value
 	case "telemetry.enabled", "telemetry.Enabled":

cmd/opencodereview/config_cmd_test.go

@@ -830,6 +830,22 @@ func TestSetConfigValueLlmUseAnthropicInvalid(t *testing.T) {
 	}
 }
 
+func TestSetConfigValueLlmPromptCaching(t *testing.T) {
+	cfg := &Config{}
+	if err := setConfigValue(cfg, "llm.prompt_caching", "false"); err != nil {
+		t.Fatalf("setConfigValue: %v", err)
+	}
+	if cfg.Llm.PromptCaching == nil || *cfg.Llm.PromptCaching {
+		t.Fatalf("PromptCaching = %v, want false", cfg.Llm.PromptCaching)
+	}
+}
+
+func TestSetConfigValueLlmPromptCachingInvalid(t *testing.T) {
+	if err := setConfigValue(&Config{}, "llm.prompt_caching", "sometimes"); err == nil {
+		t.Fatal("expected error for invalid boolean")
+	}
+}
+
 func TestSetConfigValueLanguage(t *testing.T) {
 	cfg := &Config{}
 	if err := setConfigValue(cfg, "language", "English"); err != nil {
@@ -929,7 +945,7 @@ func TestSetConfigValueUnknownKeyMessage(t *testing.T) {
 		t.Fatal("expected error for unknown key")
 	}
 	want := "unknown config key: bogus.key\n" +
-		"Supported keys: provider, model, providers.<name>.<field>, custom_providers.<name>.<field>, mcp_servers.<name>.<field>, llm.url, llm.auth_token, llm.auth_header, llm.model, llm.protocol, llm.use_anthropic, llm.extra_body, llm.extra_headers, language, telemetry.enabled, telemetry.exporter, telemetry.otlp_endpoint, telemetry.content_logging\n" +
+		"Supported keys: provider, model, providers.<name>.<field>, custom_providers.<name>.<field>, mcp_servers.<name>.<field>, llm.url, llm.auth_token, llm.auth_header, llm.model, llm.protocol, llm.use_anthropic, llm.prompt_caching, llm.extra_body, llm.extra_headers, language, telemetry.enabled, telemetry.exporter, telemetry.otlp_endpoint, telemetry.content_logging\n" +
 		"Provider fields: api_key, url, protocol, model, models, auth_header, extra_body, extra_headers\n" +
 		"Protocol values: anthropic, openai, openai-responses\n" +
 		"MCP server fields: type, command, args, env, url, headers, tools, setup"
@@ -998,6 +1014,9 @@ func TestLegacyLLMShadowWarning(t *testing.T) {
 	if got := legacyLLMShadowWarning("dashscope", "Llm.model"); got != "" {
 		t.Errorf("warning for invalid mixed-case legacy key = %q", got)
 	}
+	if got := legacyLLMShadowWarning("dashscope", "llm.prompt_caching"); got != "" {
+		t.Errorf("warning for global prompt caching setting = %q", got)
+	}
 	if got := legacyLLMShadowWarning("dashscope", "llm.model"); !strings.Contains(got, "providers.dashscope.<field>") {
 		t.Errorf("preset-provider warning = %q", got)
 	}

cmd/opencodereview/provider_cmd.go

@@ -95,6 +95,16 @@ func removeModels(existing, toRemove []string) []string {
 	return result
 }
 
+func applyPromptCachingConfig(cfg *Config, protocol string, promptCaching *bool) {
+	if llm.NormalizeProtocol(protocol) != llm.ProtocolAnthropic {
+		cfg.Llm.PromptCaching = nil
+		return
+	}
+	if promptCaching != nil {
+		cfg.Llm.PromptCaching = promptCaching
+	}
+}
+
 func applyManualConfig(configPath string, cfg *Config, result providerTUIResult) error {
 	if result.url == "" {
 		return fmt.Errorf("URL is required for manual configuration")
@@ -120,6 +130,7 @@ func applyManualConfig(configPath string, cfg *Config, result providerTUIResult)
 	// older binaries pick the OpenAI auth header/endpoint instead of wrongly
 	// defaulting to anthropic.
 	protocol := llm.NormalizeProtocol(result.protocol)
+	applyPromptCachingConfig(cfg, protocol, result.promptCaching)
 	cfg.Llm.Protocol = protocol
 	switch protocol {
 	case llm.ProtocolAnthropic:
@@ -186,6 +197,9 @@ func applyCustomProviderConfig(configPath string, cfg *Config, result providerTU
 	} else {
 		entry.APIKey = ""
 	}
+	if !result.isEdit || cfg.Provider == result.provider || result.promptCaching != nil {
+		applyPromptCachingConfig(cfg, entry.Protocol, result.promptCaching)
+	}
 	cfg.CustomProviders[result.provider] = entry
 
 	if !result.isEdit {
@@ -234,6 +248,10 @@ func applyOfficialProviderConfig(configPath string, cfg *Config, result provider
 	}
 
 	preset, isPreset := llm.LookupProvider(result.provider)
+	protocol := result.protocol
+	if isPreset {
+		protocol = preset.Protocol
+	}
 
 	if result.apiKey == "" {
 		if isPreset && preset.EnvVar != "" {
@@ -260,6 +278,7 @@ func applyOfficialProviderConfig(configPath string, cfg *Config, result provider
 		// Confirmed empty key: clear saved api_key so resolver falls back to $ENV_VAR.
 		entry.APIKey = ""
 	}
+	applyPromptCachingConfig(cfg, protocol, result.promptCaching)
 	cfg.Providers[result.provider] = entry
 
 	if cfg.Provider != result.provider {

cmd/opencodereview/provider_cmd_test.go

@@ -6,6 +6,8 @@ import (
 	"os"
 	"path/filepath"
 	"testing"
+
+	"github.com/alibaba/open-code-review/internal/llm"
 )
 
 func TestMaskKey(t *testing.T) {
@@ -274,6 +276,45 @@ func TestApplyCustomProviderConfig_EmptyKeyClearsSavedAPIKey(t *testing.T) {
 	}
 }
 
+func TestApplyCustomProviderConfig_PersistsPromptCaching(t *testing.T) {
+	disabled := false
+	configPath := filepath.Join(t.TempDir(), "config.json")
+	cfg := &Config{
+		Provider: "gateway",
+		Model:    "claude-test",
+		CustomProviders: map[string]ProviderEntry{
+			"gateway": {
+				URL:      "https://gateway.example/v1",
+				Protocol: llm.ProtocolAnthropic,
+				APIKey:   "test-key",
+				Model:    "claude-test",
+				Models:   []string{"claude-test"},
+			},
+		},
+	}
+
+	err := applyCustomProviderConfig(configPath, cfg, providerTUIResult{
+		provider:      "gateway",
+		model:         "claude-test",
+		models:        []string{"claude-test"},
+		apiKey:        "test-key",
+		isCustom:      true,
+		url:           "https://gateway.example/v1",
+		protocol:      llm.ProtocolAnthropic,
+		promptCaching: &disabled,
+	})
+	if err != nil {
+		t.Fatalf("applyCustomProviderConfig: %v", err)
+	}
+	diskCfg, err := loadOrCreateConfig(configPath)
+	if err != nil {
+		t.Fatalf("load config: %v", err)
+	}
+	if diskCfg.Llm.PromptCaching == nil || *diskCfg.Llm.PromptCaching {
+		t.Fatalf("PromptCaching = %v, want false", diskCfg.Llm.PromptCaching)
+	}
+}
+
 func TestProviderTUIResult_ResolvedModel(t *testing.T) {
 	r := providerTUIResult{
 		provider: "baidu-qianfan",