feat(cache): add provider prompt cache planner

src/main/presenter/deepchatAgentPresenter/accumulator.ts

@@ -168,6 +168,7 @@ export function accumulate(state: StreamState, event: LLMCoreStreamEvent): void
       state.metadata.outputTokens = event.usage.completion_tokens
       state.metadata.totalTokens = event.usage.total_tokens
       state.metadata.cachedInputTokens = event.usage.cached_tokens
+      state.metadata.cacheWriteInputTokens = event.usage.cache_write_tokens
       break
     }
     case 'stop': {

src/main/presenter/deepchatAgentPresenter/index.ts

@@ -1405,7 +1405,8 @@ export class DeepChatAgentPresenter implements IAgentImplementation {
       maxTokens: generationSettings.maxTokens,
       thinkingBudget: generationSettings.thinkingBudget,
       reasoningEffort: generationSettings.reasoningEffort,
-      verbosity: generationSettings.verbosity
+      verbosity: generationSettings.verbosity,
+      conversationId: sessionId
     }
 
     const traceEnabled = this.configPresenter.getSetting<boolean>('traceDebugEnabled') === true

src/main/presenter/deepchatAgentPresenter/process.ts

@@ -308,5 +308,8 @@ function buildUsageSnapshot(state: StreamState): Record<string, number> {
   if (typeof state.metadata.cachedInputTokens === 'number') {
     usage.cachedInputTokens = state.metadata.cachedInputTokens
   }
+  if (typeof state.metadata.cacheWriteInputTokens === 'number') {
+    usage.cacheWriteInputTokens = state.metadata.cacheWriteInputTokens
+  }
   return usage
 }

src/main/presenter/llmProviderPresenter/promptCacheCapabilities.ts

@@ -0,0 +1,50 @@
+export type PromptCacheMode =
+  | 'disabled'
+  | 'openai_implicit'
+  | 'anthropic_auto'
+  | 'anthropic_explicit'
+
+function normalizeId(value: string | undefined): string {
+  return value?.trim().toLowerCase() ?? ''
+}
+
+function isClaudeModel(modelId: string): boolean {
+  return modelId.includes('claude')
+}
+
+export function resolvePromptCacheMode(providerId: string, modelId: string): PromptCacheMode {
+  const normalizedProviderId = normalizeId(providerId)
+  const normalizedModelId = normalizeId(modelId)
+
+  if (normalizedProviderId === 'openai') {
+    return 'openai_implicit'
+  }
+
+  if (normalizedProviderId === 'anthropic' && isClaudeModel(normalizedModelId)) {
+    return 'anthropic_auto'
+  }
+
+  if (
+    normalizedProviderId === 'zenmux' &&
+    normalizedModelId.startsWith('anthropic/') &&
+    isClaudeModel(normalizedModelId)
+  ) {
+    return 'anthropic_explicit'
+  }
+
+  if (
+    normalizedProviderId === 'aws-bedrock' &&
+    (normalizedModelId.includes('anthropic.claude') || isClaudeModel(normalizedModelId))
+  ) {
+    return 'anthropic_explicit'
+  }
+
+  if (
+    normalizedProviderId === 'openrouter' &&
+    (normalizedModelId.startsWith('anthropic/') || isClaudeModel(normalizedModelId))
+  ) {
+    return 'anthropic_explicit'
+  }
+
+  return 'disabled'
+}