From 55879396c85f96fab0262da51e944bebcc9a511c Mon Sep 17 00:00:00 2001
From: hex2077 <justlikemaki@qq.com>
Date: Sun, 15 Feb 2026 11:21:26 +0800
Subject: [PATCH] =?UTF-8?q?feat(converters):=20=E4=B8=BA=E8=AF=B7=E6=B1=82?=
 =?UTF-8?q?=E6=B7=BB=E5=8A=A0=20metadata=20=E6=94=AF=E6=8C=81=E5=B9=B6?=
 =?UTF-8?q?=E4=BC=98=E5=8C=96=E4=BC=9A=E8=AF=9D=E7=BC=93=E5=AD=98?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 在 CodexConverter 和 ClaudeConverter 中增加 metadata 字段，透传上游元数据
- 重构 CodexApiService 的缓存键生成逻辑，优先使用 session_id 或 conversation_id
- 从上游请求体中移除 metadata 避免透传，仅用于生成缓存键
---
 src/converters/strategies/ClaudeConverter.js |  1 +
 src/converters/strategies/CodexConverter.js  |  1 +
 src/providers/openai/codex-core.js           | 26 ++++++++++++++++----
 3 files changed, 23 insertions(+), 5 deletions(-)

diff --git a/src/converters/strategies/ClaudeConverter.js b/src/converters/strategies/ClaudeConverter.js
index ecf72cf..d654d5d 100644
--- a/src/converters/strategies/ClaudeConverter.js
+++ b/src/converters/strategies/ClaudeConverter.js
@@ -1871,6 +1871,7 @@ export class ClaudeConverter extends BaseConverter {
             stream: true,
             store: false,
             parallel_tool_calls: true,
+            metadata: claudeRequest.metadata || {},
             reasoning: {
                 effort: claudeRequest.reasoning?.effort || 'medium',
                 summary: 'auto'
diff --git a/src/converters/strategies/CodexConverter.js b/src/converters/strategies/CodexConverter.js
index c0fa990..31ad6be 100644
--- a/src/converters/strategies/CodexConverter.js
+++ b/src/converters/strategies/CodexConverter.js
@@ -153,6 +153,7 @@ export class CodexConverter extends BaseConverter {
             input: this.convertMessages(data.messages || []),
             stream: true,
             store: false,
+            metadata: data.metadata || {},
             reasoning: {
                 effort: data.reasoning_effort || 'medium',
                 summary: 'auto'
diff --git a/src/providers/openai/codex-core.js b/src/providers/openai/codex-core.js
index 73222ca..3160484 100644
--- a/src/providers/openai/codex-core.js
+++ b/src/providers/openai/codex-core.js
@@ -300,8 +300,25 @@ export class CodexApiService {
      * 准备请求体
      */
     prepareRequestBody(model, requestBody, stream) {
-        // 添加会话缓存 ID
-        const cacheKey = `${model}-${requestBody.metadata?.user_id || 'default'}`;
+        // 提取 metadata 并从请求体中移除，避免透传到上游
+        const metadata = requestBody.metadata || {};
+        
+        // 明确会话维度：优先使用 session_id 或 conversation_id，其次 user_id
+        const sessionId = metadata.session_id || metadata.conversation_id || metadata.user_id || 'default';
+        
+        const cleanedBody = { ...requestBody };
+        delete cleanedBody.metadata;
+
+        // 生成会话缓存键
+        // 默认弱化 model 依赖，以提升同会话跨模型的缓存命中率
+        // 如果 sessionId 为 'default'，则必须加上 model 以提供基础隔离
+        let cacheKey = sessionId;
+        if (sessionId === 'default') {
+            cacheKey = `${model}-default`;
+        } else {
+            cacheKey = `${model}-${sessionId}`;
+        }
+        
         let cache = this.conversationCache.get(cacheKey);
 
         if (!cache || cache.expire < Date.now()) {
@@ -312,10 +329,9 @@ export class CodexApiService {
             this.conversationCache.set(cacheKey, cache);
         }
 
-        // 注意：requestBody 已经是转换后的 Codex 格式
-        // 只需要添加 cache key 和 stream 参数
+        // 注意：requestBody 已经去除了 metadata
         return {
-            ...requestBody,
+            ...cleanedBody,
             stream,
             prompt_cache_key: cache.id
         };