diff --git a/.claude/skills/review-cards/scripts/review_cards.py b/.claude/skills/review-cards/scripts/review_cards.py
index 6da0dac..4f85acc 100644
--- a/.claude/skills/review-cards/scripts/review_cards.py
+++ b/.claude/skills/review-cards/scripts/review_cards.py
@@ -88,17 +88,22 @@ def review(work_id, batch, model, dry_run):
         if not rows:
             click.echo("无待审卡")
             return
-        cards_json = json.dumps([{"idx": i + 1, **r[1]} for i, r in enumerate(rows)],
-                                ensure_ascii=False, indent=1)
-        all_reviews = {}
-        for role, desc in ROLES.items():
-            content, usage = chat(
-                REVIEW_PROMPT.format(criteria=CRITERIA, n=len(rows), cards=cards_json),
-                model=model, system=SYSTEM_TMPL.format(role_desc=desc))
-            data = extract_json(content)
-            all_reviews[role] = {int(r["idx"]): r for r in data.get("reviews", []) if r.get("idx")}
-            click.echo(f"[{role}] 评 {len(data.get('reviews', []))} 卡 "
-                       f"(in={usage.get('prompt_tokens')}) 总评: {data.get('overall', '')[:80]}")
+        # 分页 ≤30 卡/批：放量后单调用塞全库必爆；同批互见保 family 检出
+        PAGE = 30
+        all_reviews = {role: {} for role in ROLES}
+        for p0 in range(0, len(rows), PAGE):
+            page = rows[p0:p0 + PAGE]
+            cards_json = json.dumps([{"idx": p0 + i + 1, **r[1]} for i, r in enumerate(page)],
+                                    ensure_ascii=False, indent=1)
+            for role, desc in ROLES.items():
+                content, usage = chat(
+                    REVIEW_PROMPT.format(criteria=CRITERIA, n=len(page), cards=cards_json),
+                    model=model, system=SYSTEM_TMPL.format(role_desc=desc))
+                data = extract_json(content)
+                all_reviews[role].update(
+                    {int(r["idx"]): r for r in data.get("reviews", []) if r.get("idx")})
+                click.echo(f"[{role}] 批{p0 // PAGE + 1} 评 {len(data.get('reviews', []))} 卡 "
+                           f"(in={usage.get('prompt_tokens')}) 总评: {data.get('overall', '')[:80]}")
         # 汇总判定并写库（按序号匹配）
         stats = {"pass": 0, "revise": 0, "reject": 0}
         for i, (cid, payload) in enumerate(rows):