From c86f5bc304e53ac82378f86bb13c1d9cb85d6c53 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=B0=8F=E5=94=AF?= <xiaoxue@xiaoxue.ai>
Date: Thu, 2 Jul 2026 01:00:23 +0800
Subject: [PATCH] fix: NewAPI key strip sk- prefix + LLM wiki graceful fallback

---
 scripts/wiki_curator.py | 40 ++++++++++++++++++++++++++++------------
 1 file changed, 28 insertions(+), 12 deletions(-)

diff --git a/scripts/wiki_curator.py b/scripts/wiki_curator.py
index 1b4098a..4526642 100644
--- a/scripts/wiki_curator.py
+++ b/scripts/wiki_curator.py
@@ -40,7 +40,7 @@ STATE_FILE = os.path.expanduser("~/.hermes/wiki_curator_state.json")
 
 # LLM 配置
 LLM_API = "http://127.0.0.1:3000/v1/chat/completions"
-LLM_MODEL = "minimaxai/minimax-m3"
+LLM_MODEL = "minimaxai/minimax-m2.7"  # m3 sometimes returns empty, use m2.7
 
 # 扫描时排除的目录名称（大小写不敏感）
 EXCLUDE_DIRS = {
@@ -259,7 +259,11 @@ def _get_llm_key() -> str:
             cfg_path = os.path.expanduser("~/.hermes/config.yaml")
             with open(cfg_path) as f:
                 cfg = yaml.safe_load(f)
-            return cfg.get("providers", {}).get("newapi-local", {}).get("api_key", "")
+            raw = cfg.get("providers", {}).get("newapi-local", {}).get("api_key", "")
+            # NewAPI 的 key 不需要 sk- 前缀
+            if raw.startswith("sk-"):
+                raw = raw[3:]
+            return raw
         except Exception:
             pass
     # 回退到环境变量
@@ -273,27 +277,39 @@ def _extract_with_llm(content: str, filepath: str) -> dict | None:
         print("  ⚠️ No LLM API key found (check config.yaml or NEWAPI_API_KEY env)")
         return None
 
-    prompt = f'''Analyze this technical document. Extract concepts (what things are), entities (specific instances), and relations (how they connect).
-Return JSON ONLY:
-{{"concepts": [{{"name":"...","summary":"..."}}],
- "entities": [{{"name":"...","attributes":{{}}}}],
- "relations": [{{"source":"...","relation":"uses|contains|depends_on|part_of|implements","target":"..."}}]}}
+    prompt = f'''Extract concepts, entities, and relations from this document.
+Return ONLY valid JSON:
+{{"concepts":[{{"name":"...","summary":"..."}}],"entities":[{{"name":"...","attributes":{{}}}}],"relations":[{{"source":"...","relation":"uses|contains|depends_on|part_of|implements","target":"..."}}]}}
 
-Document: {content[:3000]}
+Document:
+{content[:2000]}
 '''
     try:
         resp = requests.post(LLM_API,
             headers={"Authorization": f"Bearer {llm_key}", "Content-Type": "application/json"},
-            json={"model": LLM_MODEL, "messages": [{"role": "user", "content": prompt}], "temperature": 0.1},
+            json={
+                "model": LLM_MODEL,
+                "messages": [
+                    {"role": "system", "content": "You are a knowledge extraction assistant. Always respond with valid JSON only."},
+                    {"role": "user", "content": prompt}
+                ],
+                "temperature": 0.1,
+                "max_tokens": 1000,
+            },
             timeout=30)
         data = resp.json()
+        if "error" in data and data["error"].get("message"):
+            print(f"  ⚠️ LLM API error: {data['error']['message'][:60]}")
+            return None
         choices = data.get("choices", [])
         if not choices:
             print("  ⚠️ LLM returned empty choices (API/model may be unavailable)")
             return None
-        text = choices[0].get("message", {}).get("content", "")
-        if not text:
-            print("  ⚠️ LLM returned empty content")
+        msg = choices[0].get("message", {})
+        text = msg.get("content", "") or ""
+        if not text.strip():
+            finish = choices[0].get("finish_reason", "")
+            print(f"  ⚠️ LLM returned empty content (finish={finish})")
             return None
         # Parse JSON from response
         json_match = re.search(r'\{[\s\S]*\}', text)