feat: display actual model names from API responses

Updates `call_llm` to return both the content and the actual model name returned by the API. This ensures that the review summary and notifications display the real model used (e.g., "codex" when using a local proxy) rather than static configuration values. Updated `main.py` to propagate and display these dynamic model names in the final report.

README.md

@@ -137,7 +137,8 @@ git diff → 文件过滤 → LLM 分流判定 → 审查 → 飞书通知
 |------|------|
 | `AI_REVIEW_API_URL` | `http://<你Mac内网IP>:8088/v1` |
 | `AI_REVIEW_API_KEY` | 上面的 `CODEX_PROXY_TOKEN` |
-| `AI_REVIEW_MODEL` | 任意值（代理忽略，例如 `codex`） |
+
+仅此两项。飞书通知展示的模型名走 OpenAI 协议响应里的 `model` 字段——代理会回报 `codex`，代码自动如实展示，无需配 `AI_REVIEW_MODEL` / `AI_REVIEW_TRIAGE_MODEL`。
 
 > 已有项目检查项目级 Variables 里有没有 `AI_REVIEW_API_KEY` —— 项目级会覆盖 group 级，要么删掉让它继承，要么改成新 token。
 

ai_review/llm.py

@@ -170,8 +170,11 @@ def build_deep_prompt(diff_text):
 
 
 def call_llm(system_msg, user_msg, max_tokens=4096, timeout=300, model=None, enable_thinking=False):
+    """返回 (content, response_model)。response_model 来自服务端响应（OpenAI 协议字段），
+    本地代理会回报 "codex"，真 LLM 会回报真实模型名；用于飞书通知里如实展示。"""
+    request_model = model or MODEL
     payload = {
-        "model": model or MODEL,
+        "model": request_model,
         "messages": [
             {"role": "system", "content": system_msg},
             {"role": "user", "content": user_msg}
@@ -199,23 +202,24 @@ def call_llm(system_msg, user_msg, max_tokens=4096, timeout=300, model=None, ena
     try:
         with urllib.request.urlopen(req, timeout=timeout) as resp:
             result = json.loads(resp.read().decode("utf-8"))
+            response_model = result.get("model") or request_model
             choices = result.get("choices", [])
             if not choices:
-                return "⚠️ API 返回了空结果，可能触发了内容安全策略。"
+                return "⚠️ API 返回了空结果，可能触发了内容安全策略。", response_model
             msg = choices[0].get("message", {})
             content = msg.get("content", "⚠️ API 返回格式异常")
             # 优先使用 reasoning_content（部分 OpenAI 兼容网关将 thinking 分离到此字段）
             # 如果存在 reasoning_content，说明 content 已经是干净的正文
             if msg.get("reasoning_content"):
-                return content
+                return content, response_model
             # 过滤 thinking 内容，避免飞书通知等场景内容过长
             if content:
                 content = _strip_thinking(content)
-            return content
+            return content, response_model
     except urllib.error.HTTPError as e:
         body = e.read().decode("utf-8") if e.fp else ""
-        return f"❌ API 请求失败 (HTTP {e.code}): {body}"
+        return f"❌ API 请求失败 (HTTP {e.code}): {body}", request_model
     except urllib.error.URLError as e:
-        return f"❌ 网络错误: {e.reason}"
+        return f"❌ 网络错误: {e.reason}", request_model
     except Exception as e:
-        return f"❌ 调用失败: {e}"
+        return f"❌ 调用失败: {e}", request_model

ai_review/main.py

@@ -45,7 +45,6 @@ def main():
             [
                 f"提交者: {committer}",
                 f"提交: {commit_sha_short} {commit_msg}",
-                f"分流模型: {TRIAGE_MODEL}",
                 "",
                 "✅ 没有需要审查的代码变更。",
             ]
@@ -65,7 +64,7 @@ def main():
     print("   ⏳ 正在调用大模型分流判定...")
     t_triage = time.time()
     triage_system_msg, triage_user_msg = build_triage_prompt(triage_diff)
-    triage_result = call_llm(
+    triage_result, triage_model_used = call_llm(
         triage_system_msg,
         triage_user_msg,
         max_tokens=256,
@@ -89,6 +88,8 @@ def main():
     review_level = triage_decision  # skip / standard / deep
     print(f"   📌 分流决策: {review_level} | 类型: {triage_category} | 原因: {triage_reason}")
 
+    review_model_used = triage_model_used  # skip 路径下不会再发起调用，沿用分流模型名
+
     if triage_decision == "skip":
         review_result = f"✅ 代码审查通过，LLM 判定可跳过完整审查：{triage_reason}"
         print(f"   ✅ 分流判定：跳过完整审查（{triage_reason}）")
@@ -100,7 +101,7 @@ def main():
         print(f"   ⏳ 正在调用标准审查 ({TRIAGE_MODEL})...")
         t_review = time.time()
         system_msg, user_msg = build_standard_prompt(diff)
-        review_result = call_llm(
+        review_result, review_model_used = call_llm(
             system_msg, user_msg,
             max_tokens=STANDARD_MAX_TOKENS,
             timeout=STANDARD_TIMEOUT,
@@ -116,7 +117,7 @@ def main():
         print(f"   ⏳ 正在调用深度审查 ({MODEL}, thinking mode)...")
         t_review = time.time()
         system_msg, user_msg = build_deep_prompt(diff)
-        review_result = call_llm(
+        review_result, review_model_used = call_llm(
             system_msg, user_msg,
             max_tokens=DEEP_MAX_TOKENS,
             timeout=DEEP_TIMEOUT,
@@ -147,8 +148,8 @@ def main():
 
     _level_labels = {
         "skip": "跳过",
-        "standard": f"标准审查 ({TRIAGE_MODEL})",
-        "deep": f"深度审查 ({MODEL}, thinking)",
+        "standard": f"标准审查 ({review_model_used})",
+        "deep": f"深度审查 ({review_model_used}, thinking)",
     }
     review_level_label = _level_labels.get(review_level, review_level)
 
@@ -158,7 +159,7 @@ def main():
         f"分支: {branch}",
         f"提交: {commit_sha_short} {commit_msg}",
         f"变更: {files} 个文件, +{added} -{removed} 行",
-        f"分流模型: {TRIAGE_MODEL}",
+        f"分流模型: {triage_model_used}",
         f"审查层级: {review_level_label}",
         f"变更类型: {triage_category}",
         f"分析耗时: {elapsed_str}",