@kenkaiiii/gg-ai 5.13.2 → 5.13.3

package/dist/index.d.cts

@@ -175,6 +175,10 @@ interface StreamOptions {
     serviceTier?: "auto" | "default" | "flex" | "priority";
     /** OpenAI ChatGPT account ID (from OAuth JWT) for codex endpoint */
     accountId?: string;
+    /** Stable conversation identity for Codex transport headers. This is distinct from
+     *  promptCacheKey: sessions with matching prefixes may share a cache key, but must
+     *  retain independent session/thread identities. */
+    transportSessionId?: string;
     /** Google Cloud/Code Assist project ID used by Gemini OAuth transport. */
     projectId?: string;
     /** Enable provider-native web search. Each provider uses its own format:

package/dist/index.d.ts

@@ -175,6 +175,10 @@ interface StreamOptions {
     serviceTier?: "auto" | "default" | "flex" | "priority";
     /** OpenAI ChatGPT account ID (from OAuth JWT) for codex endpoint */
     accountId?: string;
+    /** Stable conversation identity for Codex transport headers. This is distinct from
+     *  promptCacheKey: sessions with matching prefixes may share a cache key, but must
+     *  retain independent session/thread identities. */
+    transportSessionId?: string;
     /** Google Cloud/Code Assist project ID used by Gemini OAuth transport. */
     projectId?: string;
     /** Enable provider-native web search. Each provider uses its own format:

package/dist/index.js

@@ -1711,11 +1711,16 @@ function getEnvironment() {
 // src/providers/openai.ts
 function extractOpenAIUsage(usage) {
   let cacheRead = 0;
+  let cacheWrite = 0;
   const details = usage.prompt_tokens_details;
   if (details?.cached_tokens) {
     cacheRead = details.cached_tokens;
   }
   const usageAny = usage;
+  const detailsAny = details;
+  if (typeof detailsAny?.cache_write_tokens === "number") {
+    cacheWrite = detailsAny.cache_write_tokens;
+  }
   if (!cacheRead && typeof usageAny.cached_tokens === "number" && usageAny.cached_tokens > 0) {
     cacheRead = usageAny.cached_tokens;
   }
@@ -1723,9 +1728,10 @@ function extractOpenAIUsage(usage) {
     cacheRead = usageAny.prompt_cache_hit_tokens;
   }
   return {
-    inputTokens: usage.prompt_tokens - cacheRead,
+    inputTokens: usage.prompt_tokens - cacheRead - cacheWrite,
     outputTokens: usage.completion_tokens,
-    cacheRead
+    cacheRead,
+    cacheWrite
   };
 }
 var openaiClientCache = /* @__PURE__ */ new Map();
@@ -1790,8 +1796,9 @@ async function* runStream2(options) {
   if (options.provider === "openai" || options.provider === "moonshot") {
     const paramsAny = params;
     paramsAny.prompt_cache_key = normalizePromptCacheKey(options.promptCacheKey ?? "ggcoder");
-    const retention = options.cacheRetention ?? "short";
-    if (retention === "long") {
+    if (options.provider === "openai" && options.model.startsWith("gpt-5.6")) {
+      paramsAny.prompt_cache_options = { mode: "implicit", ttl: "30m" };
+    } else if ((options.cacheRetention ?? "short") === "long") {
       paramsAny.prompt_cache_retention = "24h";
     }
   }
@@ -1842,6 +1849,7 @@ async function* runStream2(options) {
   let inputTokens = 0;
   let outputTokens = 0;
   let cacheRead = 0;
+  let cacheWrite = 0;
   let finishReason = null;
   let receivedAnyChunk = false;
   try {
@@ -1849,7 +1857,7 @@ async function* runStream2(options) {
       receivedAnyChunk = true;
       const choice = chunk.choices?.[0];
       if (chunk.usage) {
-        ({ inputTokens, outputTokens, cacheRead } = extractOpenAIUsage(chunk.usage));
+        ({ inputTokens, outputTokens, cacheRead, cacheWrite } = extractOpenAIUsage(chunk.usage));
       }
       if (!choice) continue;
       if (choice.finish_reason) {
@@ -1929,7 +1937,12 @@ async function* runStream2(options) {
       content: contentParts.length > 0 ? contentParts : textAccum || ""
     },
     stopReason,
-    usage: { inputTokens, outputTokens, ...cacheRead > 0 && { cacheRead } }
+    usage: {
+      inputTokens,
+      outputTokens,
+      ...cacheRead > 0 && { cacheRead },
+      ...cacheWrite > 0 && { cacheWrite }
+    }
   };
   yield { type: "done", stopReason };
   return response;
@@ -2002,8 +2015,9 @@ function completionToResponse(completion) {
   let inputTokens = 0;
   let outputTokens = 0;
   let cacheRead = 0;
+  let cacheWrite = 0;
   if (completion.usage) {
-    ({ inputTokens, outputTokens, cacheRead } = extractOpenAIUsage(completion.usage));
+    ({ inputTokens, outputTokens, cacheRead, cacheWrite } = extractOpenAIUsage(completion.usage));
   }
   const stopReason = normalizeOpenAIStopReason(choice?.finish_reason ?? null);
   return {
@@ -2012,7 +2026,12 @@ function completionToResponse(completion) {
       content: contentParts.length > 0 ? contentParts : textAccum
     },
     stopReason,
-    usage: { inputTokens, outputTokens, ...cacheRead > 0 && { cacheRead } }
+    usage: {
+      inputTokens,
+      outputTokens,
+      ...cacheRead > 0 && { cacheRead },
+      ...cacheWrite > 0 && { cacheWrite }
+    }
   };
 }
 function classifyOpenAICompatLimit(args) {
@@ -2192,10 +2211,9 @@ async function* runStream3(options) {
   if (options.accountId) {
     headers["chatgpt-account-id"] = options.accountId;
   }
-  const cacheScopeId = body.prompt_cache_key;
-  if (cacheScopeId) {
-    headers["session_id"] = cacheScopeId;
-    headers["x-client-request-id"] = cacheScopeId;
+  if (options.transportSessionId) {
+    headers["session_id"] = options.transportSessionId;
+    headers["x-client-request-id"] = options.transportSessionId;
   }
   const response = await fetch(url, {
     method: "POST",
@@ -2239,6 +2257,7 @@ async function* runStream3(options) {
   let inputTokens = 0;
   let outputTokens = 0;
   let cacheRead = 0;
+  let cacheWrite = 0;
   const diagStart = Date.now();
   const diagSeen = /* @__PURE__ */ new Set();
   for await (const event of parseSSE(response.body)) {
@@ -2415,7 +2434,8 @@ async function* runStream3(options) {
       const usage = resp?.usage;
       if (usage) {
         cacheRead = usage.input_tokens_details?.cached_tokens ?? 0;
-        inputTokens = (usage.input_tokens ?? 0) - cacheRead;
+        cacheWrite = usage.input_tokens_details?.cache_write_tokens ?? 0;
+        inputTokens = (usage.input_tokens ?? 0) - cacheRead - cacheWrite;
         outputTokens = usage.output_tokens ?? 0;
       }
     }
@@ -2463,7 +2483,12 @@ async function* runStream3(options) {
       content: contentParts.length > 0 ? contentParts : textAccum || ""
     },
     stopReason,
-    usage: { inputTokens, outputTokens, ...cacheRead > 0 && { cacheRead } }
+    usage: {
+      inputTokens,
+      outputTokens,
+      ...cacheRead > 0 && { cacheRead },
+      ...cacheWrite > 0 && { cacheWrite }
+    }
   };
   yield { type: "done", stopReason };
   return streamResponse;