@kenkaiiii/gg-ai 4.3.166 → 4.3.168

package/dist/index.cjs

@@ -438,9 +438,18 @@ function toAnthropicToolResultContent(content) {
     };
   });
 }
+function remapAnthropicToolCallId(id, idMap) {
+  if (/^[a-zA-Z0-9_-]+$/.test(id)) return id;
+  const existing = idMap.get(id);
+  if (existing) return existing;
+  const mapped = id.replace(/[^a-zA-Z0-9_-]/g, "_");
+  idMap.set(id, mapped);
+  return mapped;
+}
 function toAnthropicMessages(messages, cacheControl) {
   let systemText;
   const out = [];
+  const idMap = /* @__PURE__ */ new Map();
   for (const msg of messages) {
     if (msg.role === "system") {
       systemText = msg.content;
@@ -475,7 +484,7 @@ function toAnthropicMessages(messages, cacheControl) {
         if (part.type === "tool_call")
           return {
             type: "tool_use",
-            id: part.id,
+            id: remapAnthropicToolCallId(part.id, idMap),
             name: part.name,
             input: part.args
           };
@@ -500,7 +509,7 @@ function toAnthropicMessages(messages, cacheControl) {
         role: "user",
         content: msg.content.map((result) => ({
           type: "tool_result",
-          tool_use_id: result.toolCallId,
+          tool_use_id: remapAnthropicToolCallId(result.toolCallId, idMap),
           content: toAnthropicToolResultContent(result.content),
           is_error: result.isError
         }))
@@ -556,12 +565,19 @@ function toAnthropicMessages(messages, cacheControl) {
   }
   return { system, messages: out };
 }
-function toAnthropicTools(tools) {
-  return tools.map((tool) => ({
-    name: tool.name,
-    description: tool.description,
-    input_schema: tool.rawInputSchema ?? zodToJsonSchema(tool.parameters)
-  }));
+function toAnthropicTools(tools, options) {
+  return tools.map((tool, index) => {
+    const anthropicTool = {
+      name: tool.name,
+      description: tool.description,
+      input_schema: tool.rawInputSchema ?? zodToJsonSchema(tool.parameters),
+      ...options?.enableFineGrainedToolStreaming ? { eager_input_streaming: true } : {}
+    };
+    if (options?.cacheControl && index === tools.length - 1) {
+      anthropicTool.cache_control = options.cacheControl;
+    }
+    return anthropicTool;
+  });
 }
 function toAnthropicToolChoice(choice) {
   if (choice === "auto") return { type: "auto" };
@@ -780,6 +796,7 @@ async function* runStream(options) {
   const isOAuth = options.apiKey?.startsWith("sk-ant-oat");
   const useStreaming = options.streaming !== false;
   const cacheControl = toAnthropicCacheControl(options.cacheRetention, options.baseUrl);
+  const supportsFirstPartyToolExtras = !options.baseUrl || options.baseUrl.includes("api.anthropic.com");
   const downgradedMessages = downgradeUnsupportedImages(options.messages, options.supportsImages);
   const { system: rawSystem, messages } = toAnthropicMessages(downgradedMessages, cacheControl);
   const system = isOAuth ? [
@@ -812,7 +829,10 @@ async function* runStream(options) {
     ...options.stop ? { stop_sequences: options.stop } : {},
     ...options.tools?.length || options.serverTools?.length || options.webSearch ? {
       tools: [
-        ...options.tools?.length ? toAnthropicTools(options.tools) : [],
+        ...options.tools?.length ? toAnthropicTools(options.tools, {
+          ...supportsFirstPartyToolExtras && cacheControl ? { cacheControl } : {},
+          ...supportsFirstPartyToolExtras ? { enableFineGrainedToolStreaming: true } : {}
+        }) : [],
         ...options.serverTools ?? [],
         ...options.webSearch ? [{ type: "web_search_20250305", name: "web_search" }] : []
       ]
@@ -1198,12 +1218,15 @@ async function* runStream2(options) {
   };
   if (options.provider === "openai" || options.provider === "moonshot") {
     const paramsAny = params;
-    paramsAny.prompt_cache_key = "ggcoder";
+    paramsAny.prompt_cache_key = options.promptCacheKey ?? "ggcoder";
     const retention = options.cacheRetention ?? "short";
     if (retention === "long") {
       paramsAny.prompt_cache_retention = "24h";
     }
   }
+  if (options.provider === "openai" && options.serviceTier) {
+    params.service_tier = options.serviceTier;
+  }
   if (usesThinkingParam) {
     if (options.thinking) {
       params.thinking = { type: "enabled" };
@@ -1486,6 +1509,9 @@ async function* runStream3(options) {
   if (options.tools?.length) {
     body.tools = toCodexTools(options.tools);
   }
+  if (options.promptCacheKey) {
+    body.prompt_cache_key = options.promptCacheKey;
+  }
   if (options.temperature != null && !options.thinking) {
     body.temperature = options.temperature;
   }
@@ -1537,6 +1563,7 @@ async function* runStream3(options) {
   const toolCalls = /* @__PURE__ */ new Map();
   let inputTokens = 0;
   let outputTokens = 0;
+  let cacheRead = 0;
   for await (const event of parseSSE(response.body)) {
     const type = event.type;
     if (!type) continue;
@@ -1629,7 +1656,8 @@ async function* runStream3(options) {
       const resp = event.response;
       const usage = resp?.usage;
       if (usage) {
-        inputTokens = usage.input_tokens ?? 0;
+        cacheRead = usage.input_tokens_details?.cached_tokens ?? 0;
+        inputTokens = (usage.input_tokens ?? 0) - cacheRead;
         outputTokens = usage.output_tokens ?? 0;
       }
     }
@@ -1659,7 +1687,7 @@ async function* runStream3(options) {
       content: contentParts.length > 0 ? contentParts : textAccum || ""
     },
     stopReason,
-    usage: { inputTokens, outputTokens }
+    usage: { inputTokens, outputTokens, ...cacheRead > 0 && { cacheRead } }
   };
   yield { type: "done", stopReason };
   return streamResponse;