Upload run.py with huggingface_hub

run.py

@@ -436,7 +436,11 @@ def run(arxiv_url: str, top_n: int = 5, progress=None) -> dict:
 
     broad_queries = _extract_broad_queries(title, abstract)
     # 合并对比算法搜索词，去重
-    all_queries = list(dict.fromkeys(s2_extra_queries + comparison_queries + broad_queries))[:15]
+    has_github_token = bool(os.getenv("GITHUB_TOKEN", ""))
+    max_queries = 15 if has_github_token else 8
+    all_queries = list(dict.fromkeys(s2_extra_queries + comparison_queries + broad_queries))[:max_queries]
+    if not has_github_token:
+        print(f"  ⚠️ 未设置 GITHUB_TOKEN，搜索词限制为 {max_queries} 个以避免限速")
     print(f"  LLM: {len(broad_queries)} + S2引用: {len(comparison_queries)} + S2论文: {len(s2_extra_queries)} = {len(all_queries)} 个总搜索词:")
     for q in all_queries:
         print(f"    - {q}")
@@ -495,10 +499,8 @@ def run(arxiv_url: str, top_n: int = 5, progress=None) -> dict:
     try:
         direction = analyze_direction(title, abstract, filtered_results, domain_context)
     except Exception as e:
-        return {
-            "paper": paper,
-            "error": f"研究方向解析失败: {e}",
-        }
+        print(f"  [WARN] Agent 1 方向解析失败，降级为基本分析: {e}")
+        direction = _make_fallback_direction(title, abstract, filtered_results)
 
     subfield = direction.get("subfield", "未知")
     families = direction.get("method_families", [])
@@ -845,6 +847,31 @@ def _sanity_check_direction(title: str, subfield: str) -> None:
         print(f"     这可能是 LLM 混淆了论文，请人工核实分析结果。")
 
 
+def _make_fallback_direction(title: str, abstract: str, repos: list[dict]) -> dict:
+    """Agent 1 失败时的降级方向分析：基于搜索到的仓库名称推断子领域。"""
+    # 从仓库 topic/description 提取高频词作为子领域
+    all_words = []
+    for r in repos[:10]:
+        desc = (r.get("description") or "")
+        topics = " ".join(r.get("topics", []))
+        all_words.extend((desc + " " + topics).lower().split())
+
+    stops = {'a', 'an', 'the', 'of', 'for', 'in', 'on', 'to', 'and', 'or', 'is', 'are',
+             'we', 'our', 'that', 'this', 'with', 'from', 'by', 'as', 'at', 'be', 'it'}
+    meaningful = [w for w in all_words if w not in stops and len(w) >= 4]
+    word_freq = {}
+    for w in meaningful:
+        word_freq[w] = word_freq.get(w, 0) + 1
+    top_words = sorted(word_freq, key=word_freq.get, reverse=True)[:6]
+
+    return {
+        "subfield": f"基于仓库数据推断: {', '.join(top_words[:3])}" if top_words else "未知领域",
+        "subfield_trend": "（Agent 1 暂不可用，趋势分析跳过。后续版本将自动恢复。）",
+        "method_families": [],
+        "broad_queries": [],
+    }
+
+
 def _make_error_evaluation(error_msg: str) -> dict:
     """构造一个表示评估失败的 evaluation dict"""
     return {