From 85250c26942eb1c760e1ffe15292e1a1302775a6 Mon Sep 17 00:00:00 2001
From: Liang Jiaqing <l.j.q.light@gmail.com>
Date: Wed, 25 Mar 2026 09:45:43 +0800
Subject: [PATCH] fix: unify assistant content format for prompt caching to
 work

---
 llmcore.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/llmcore.py b/llmcore.py
index 3abc151..8fdb068 100644
--- a/llmcore.py
+++ b/llmcore.py
@@ -120,10 +120,10 @@ class ClaudeSession:
         except Exception as e: yield f"Error: {str(e)}"
     def make_messages(self, raw_list):
         trimmed = self._trim_messages(raw_list)
-        msgs = [{"role": m['role'], "content": m['prompt']} for m in trimmed]
+        msgs = [{"role": m['role'], "content": [{"type": "text", "text": m['prompt']}] if m['role'] == "assistant" else m['prompt']} for m in trimmed]
         for i in range(len(msgs)-1, -1, -1):
             if msgs[i]["role"] == "assistant":
-                msgs[i]["content"] = [{"type": "text", "text": msgs[i]["content"], "cache_control": {"type": "ephemeral"}}]
+                msgs[i]["content"][-1]["cache_control"] = {"type": "ephemeral"}
                 break
         return msgs
     def ask(self, prompt, model=None, stream=False):