fix: 三平台内容差异化 + admin敏感词管理表格化

- writer.py: _expand_section() 去除 <100字阈值，始终调用 LLM 平台专属扩写
- prompt_loader.py: 新增 section_expansion_zhihu/wechat/xiaohongshu 三个独立 prompt
- admin.html: 配置管理标签页 + 敏感词/清理规则子标签 + 敏感词表格化管理(编辑/删除)
- config_items.py: PUT /sensitive-words/{id} 支持更新 word/category
- compliance_checker.py: AI 套话从 DB 加载 + 人称规则修正
- initial_data.py: PlatformConfig 字数迁移 + 新种子
- 各前端页面: LLM 配置 rate_limit 字段 + 供应商列表排序

AGENTS.md

@@ -59,16 +59,40 @@ yu-zhi-ran/
 - `USE_POSTGRES=false` falls back to SQLite (used in tests)
 - Models have timezone-aware `DateTime(timezone=True)` columns
 
+### Content quality architecture (三平台差异化)
+
+**核心原则：三平台不再共享同一篇 markdown，各自独立展开。**
+- `writer.py` 对 zhihu/wechat/xiaohongshu 分别调用 `generate_platform_markdown(platform)`，各走不同的 `section_expansion_{platform}` prompt
+- 知乎：数据分析深度（400-800字/节），用「你」称呼读者
+- 公众号：个人叙事对话感（300-500字/节），用「我」口吻
+- 小红书：精炼干货（100-200字/节），直接给方法，可用 emoji
+
+**字数配置：**
+- 知乎 min 3000 / max 8000
+- 公众号 min 2000 / max 4000
+- 小红书 min 400 / max 1000
+- 存于 `platform_configs` 表，`min_words` / `max_words` 字段，后台「平台管理」可改
+
+**AI 套话检测：**
+- DB 存储：`content_clean_rules` 表，`rule_type='ai_telltale'`
+- 由 `config_items.py` 中的 `DEFAULT_CONTENT_CLEAN_RULES` 种子数据
+- `compliance_checker.py` 运行时从 DB 加载（`_load_ai_telltales()`），DB 不可用时回退代码硬编码列表
+- 后台「配置管理→敏感词/清理规则」可增删改（操作 `ContentCleanRule` 表）
+
 ### Prompts (`prompt_configs` table)
 - **DB 是唯一来源**，修改 prompt 直接 `UPDATE prompt_configs SET content = '...' WHERE key = '...';`
 - 代码 `scripts/prompt_loader.py` 中的 `_PROMPT_DEFAULTS` **仅作种子数据**，第一次写入后就不再生效
 - 新增 prompt：在 `_PROMPT_DEFAULTS` 添加定义 → 重启后自动补入 DB（仅当该 key 不存在时）
 - 修改 prompt：**直接改 DB，不要改代码**（除非要更新种子供新环境用）
 - DB 不可用时回退代码默认值（仅紧急模式）
+- 三平台独立 section_expansion 提示词 key: `section_expansion_zhihu` / `section_expansion_wechat` / `section_expansion_xiaohongshu`
+- 标题提示词去套路化，使用自然语言（不像 AI 写的 prompt）
 
 ### Prompt quality checks (`compliance_checker.py`)
 - 软质量问题（AI套话/人称混用/阅读体验）只降分、不挡流程（`passed=true`）
 - 硬合规问题（敏感词/法律/品牌）扣分多且阻塞流程
+- AI 套话从 DB `content_clean_rules(rule_type='ai_telltale')` 加载，后台可动态管理
+- 人称检查修正：去掉「大家」误报，仅检查「你们」和「你」混用
 
 ### LLM
 - `call_llm()` in `core/nvidia_client.py` — reads active provider from DB `LLMConfig.is_active`, API key from env

PROGRESS.md

@@ -3,7 +3,7 @@
 > 本文件为项目进度唯一真理源，所有进度信息以此为准。
 > 其他文档中的进度描述一律以本文为准。
 
-**最后更新**：2026-06-02 (v18)
+**最后更新**：2026-06-05 (v19)
 
 ---
 
@@ -154,6 +154,9 @@
 | 搜索提供商重构 | 2026-05-28 | 移除 opencode MCP（配额耗尽），新增 360/搜狗/微信搜索（免 Key），百度千帆日限 50→200 |
 | opencode 冗余代码清理 | 2026-06-02 | 删除 opencode_search.py / mcp_search_server.py / _call_mcp / 前后端所有 opencode/MCP 引用；禁用搜索缓存定时任务 |
 | 任务抽屉产出记录优化 | 2026-06-03 | drawer 产出结果 tab 优先显示日志消息（last_message），过滤 None/空值；修复 tasks.py 因 opencode 清理导致的 SyntaxError；清理 DB 中旧搜索缓存的 56 条 TaskLog + 1 条 TaskConfig |
+| Hermes gateway 用户级服务恢复 | 2026-06-04 | 系统级服务 → 用户级（MemoryMax=1G 生效）；删除 system 级文件；linger + default.target.wants 开机自启 |
+| v-cloak 修复 Vue 模板闪烁 | 2026-06-04 | 9 页 `<div id="app">` + `theme-modern.css` 的 `[v-cloak]` 规则，消除原始模板代码闪烁 |
+| **sensenova 多模型 + 模型级 rate limit** | **2026-06-04** | `LLMConfig` 加 `rate_limit`/`rate_limit_window_minutes`；seed 3 模型：deepseek-v4-flash(500次/5h)、6.7-flash-lite(1500次/5h)、u1-fast(1500次/5h)；进程内 rate limiter；`LLM_TASK_MODEL` 环境变量实现任务级模型选择；`ai_image_generator.py` 已固定用 `sensenova-u1-fast` |
 
 ### ⏳ 待办
 
@@ -180,7 +183,28 @@
 
 ---
 
-## 七、规范说明
+## 七、内容质量升级（v19 · 2026-06-05）
+
+### 三平台差异化内容架构
+
+**背景**：此前三平台共用同一篇 markdown，仅做浅层适配（加 emoji/换人称），导致知乎缺深度、公众号缺情感、小红书过长。
+
+**核心改动：**
+1. **writer.py** — `generate_full_markdown()` → `generate_platform_markdown(platform)`，每平台调用独立的 `section_expansion_{platform}` prompt
+   - 知乎：`section_expansion_zhihu` — 数据分析深度，400-800字/节，「你」视角
+   - 公众号：`section_expansion_wechat` — 个人叙事对话感，300-500字/节，「我」口吻
+   - 小红书：`section_expansion_xiaohongshu` — 精炼干货，100-200字/节，直接给方法+可 emoji
+   - 去掉公众号中「我们→我→你」的强制替换
+2. **prompt_loader.py** — 新增 3 个平台独立 section_expansion 提示词，标题提示词去套路化（去除「直接输出」「不要其他文字」等 AI 味指令）
+3. **平台字数提升** — min/max 大幅上调（知乎 3000-8000, 公众号 2000-4000, 小红书 400-1000），存于 `platform_configs` 表，后台可改
+4. **compliance_checker.py** — AI 套话从 DB `content_clean_rules(rule_type='ai_telltale')` 加载，后台可动态管理；人称规则去掉「大家」误报
+5. **config_items.py** — DEFAULT_CONTENT_CLEAN_RULES 新增 30+ 条 ai_telltale 规则
+6. **initial_data.py** — PlatformConfig 字数迁移（已有库也会更新）
+7. **AGENTS.md** — 文档化三平台差异化架构
+
+**验证**：33/33 测试通过
+
+## 八、规范说明
 
 ### 规划文档
 

platform/backend/app/api/articles.py

@@ -161,7 +161,6 @@ def update_article_content(
         )
         db.add(article)
     article.html_content = html_content
-    article.updated_at = datetime.now()
     db.commit()
     return {"ok": True, "id": article_id}
 

platform/backend/app/api/config_items.py

@@ -100,6 +100,32 @@ DEFAULT_CONTENT_CLEAN_RULES = [
     {"rule_type": "verbosity", "pattern": r"^综上所述$", "description": "AI废话-综上所述", "sort_order": 14},
     {"rule_type": "verbosity", "pattern": r"^通过以上", "description": "AI废话-通过以上", "sort_order": 15},
     {"rule_type": "html_thinking", "pattern": r"<p[^>]*>(好的|好的，|好[的，]|我来|让我|我将|我这就)", "description": "AI思考-HTML模式", "sort_order": 16},
+    {"rule_type": "ai_telltale", "pattern": r"值得注意的是", "description": "AI套话—值得注意的是", "sort_order": 20},
+    {"rule_type": "ai_telltale", "pattern": r"首先.*其次.*最后", "description": "AI套话—首先其次最后", "sort_order": 21},
+    {"rule_type": "ai_telltale", "pattern": r"综上所述", "description": "AI套话—综上所述", "sort_order": 22},
+    {"rule_type": "ai_telltale", "pattern": r"总的来说", "description": "AI套话—总的来说", "sort_order": 23},
+    {"rule_type": "ai_telltale", "pattern": r"总而言之", "description": "AI套话—总而言之", "sort_order": 24},
+    {"rule_type": "ai_telltale", "pattern": r"无可否认", "description": "AI套话—无可否认", "sort_order": 25},
+    {"rule_type": "ai_telltale", "pattern": r"毋庸置疑", "description": "AI套话—毋庸置疑", "sort_order": 26},
+    {"rule_type": "ai_telltale", "pattern": r"众所周知", "description": "AI套话—众所周知", "sort_order": 27},
+    {"rule_type": "ai_telltale", "pattern": r"不可否认", "description": "AI套话—不可否认", "sort_order": 28},
+    {"rule_type": "ai_telltale", "pattern": r"从某种意义上", "description": "AI套话—从某种意义上", "sort_order": 29},
+    {"rule_type": "ai_telltale", "pattern": r"从某种程度上", "description": "AI套话—从某种程度上", "sort_order": 30},
+    {"rule_type": "ai_telltale", "pattern": r"在一定程度上", "description": "AI套话—在一定程度上", "sort_order": 31},
+    {"rule_type": "ai_telltale", "pattern": r"换而言之", "description": "AI套话—换而言之", "sort_order": 32},
+    {"rule_type": "ai_telltale", "pattern": r"换言之", "description": "AI套话—换言之", "sort_order": 33},
+    {"rule_type": "ai_telltale", "pattern": r"归根结底", "description": "AI套话—归根结底", "sort_order": 34},
+    {"rule_type": "ai_telltale", "pattern": r"说到底", "description": "AI套话—说到底", "sort_order": 35},
+    {"rule_type": "ai_telltale", "pattern": r"这为我们提供了", "description": "AI套话—这为我们提供了", "sort_order": 36},
+    {"rule_type": "ai_telltale", "pattern": r"引人深思", "description": "AI套话—引人深思", "sort_order": 37},
+    {"rule_type": "ai_telltale", "pattern": r"毫无悬念", "description": "AI套话—毫无悬念", "sort_order": 38},
+    {"rule_type": "ai_telltale", "pattern": r"不知大家", "description": "AI套话—不知大家", "sort_order": 39},
+    {"rule_type": "ai_telltale", "pattern": r"说回到", "description": "AI套话—说回到", "sort_order": 40},
+    {"rule_type": "ai_telltale", "pattern": r"如果你也", "description": "AI套话—如果你也", "sort_order": 41},
+    {"rule_type": "ai_telltale", "pattern": r"我们不难发现", "description": "AI套话—我们不难发现", "sort_order": 42},
+    {"rule_type": "ai_telltale", "pattern": r"我们可以看出", "description": "AI套话—我们可以看出", "sort_order": 43},
+    {"rule_type": "ai_telltale", "pattern": r"从以上分析可以看出", "description": "AI套话—从以上分析可以看出", "sort_order": 44},
+    {"rule_type": "ai_telltale", "pattern": r"来说说到底", "description": "AI套话—来说说到底", "sort_order": 45},
 ]
 
 class KeywordDomainMapResponse(BaseModel):
@@ -193,6 +219,14 @@ def _ensure_defaults(db: Session):
     if db.query(ContentCleanRule).count() == 0:
         for item in DEFAULT_CONTENT_CLEAN_RULES:
             db.add(ContentCleanRule(**item))
+    else:
+        # 补入缺失的 ai_telltale 规则（不覆盖已存在的内容）
+        existing_patterns = {r.pattern for r in db.query(ContentCleanRule).filter(ContentCleanRule.rule_type == 'ai_telltale').all()}
+        for item in DEFAULT_CONTENT_CLEAN_RULES:
+            if item['rule_type'] == 'ai_telltale' and item['pattern'] not in existing_patterns:
+                db.add(ContentCleanRule(**item))
+        if any(item['rule_type'] == 'ai_telltale' and item['pattern'] not in existing_patterns for item in DEFAULT_CONTENT_CLEAN_RULES):
+            db.flush()
     if db.query(TrendFieldMapping).count() == 0:
         for item in DEFAULT_TREND_FIELD_MAP:
             db.add(TrendFieldMapping(**item))
@@ -253,12 +287,12 @@ def create_sensitive_word(data: SensitiveWordCreate, db: Session = Depends(get_d
     return item
 
 @router.put("/sensitive-words/{item_id}", response_model=SensitiveWordResponse)
-def update_sensitive_word(item_id: int, enabled: bool = None, db: Session = Depends(get_db), admin_user=Depends(get_current_admin)):
+def update_sensitive_word(item_id: int, body: SensitiveWordCreate, db: Session = Depends(get_db), admin_user=Depends(get_current_admin)):
     item = db.query(SensitiveWord).filter(SensitiveWord.id == item_id).first()
     if not item:
         raise HTTPException(status_code=404, detail="未找到")
-    if enabled is not None:
+    item.word = body.word
-        item.is_active = enabled
+    item.category = body.category
     db.commit()
     db.refresh(item)
     return item

platform/backend/app/api/topics.py

@@ -304,10 +304,17 @@ def batch_update_status(
     of = org_filter(current_user, Topic)
     if of is not True:
         q = q.filter(of)
-    updated = q.update(
+    now = datetime.now()
-        {Topic.status: status, Topic.updated_at: datetime.now()},
+    update_dict = {Topic.status: status, Topic.updated_at: now}
-        synchronize_session=False
+    if status in ('review', '待审查'):
-    )
+        update_dict[Topic.generated_at] = now
+        update_dict[Topic.reviewed_at] = now
+    elif status in ('ready', '待发布'):
+        update_dict[Topic.ready_at] = now.date()
+        update_dict[Topic.reviewed_at] = now
+    elif status in ('published', '已发布'):
+        update_dict[Topic.published_at] = now.date()
+    updated = q.update(update_dict, synchronize_session=False)
     db.commit()
     return {"ok": True, "updated": updated}
 

platform/backend/app/core/generator.py

@@ -65,6 +65,7 @@ def run_creator_blocking(topic_id: str = None, timeout: int = 1800):
             topic = db.query(Topic).filter(Topic.id == topic_id).first()
             if topic:
                 topic.generated_at = datetime.now(timezone.utc)
+                topic.reviewed_at = datetime.now(timezone.utc)
                 if topic.status in ('pending', '待处理'):
                     topic.status = 'review'
                 db.commit()

platform/backend/app/core/nvidia_client.py

@@ -1,9 +1,11 @@
 """
 Unified LLM Client
-支持 NVIDIA / opencode-go / 兼容 OpenAI 格式的 API,配置从环境变量读取
+支持 NVIDIA / sensenova / 兼容 OpenAI 格式的 API,配置从环境变量读取
+支持模型级 rate limit 和任务级模型选择（LLM_TASK_MODEL）
 """
 
 import os
+import time
 import requests
 import json
 import logging
@@ -30,10 +32,28 @@ _API_KEYS = {
 # 代码级回退默认值（实际配置优先从 DB 读取）
 _FALLBACK = {
     "opencode-go": {"base_url": "https://opencode.ai/zen/go/v1", "model": "deepseek-v4-flash"},
-    "nvidia": {"base_url": "https://integrate.api.nvidia.com/v1", "model": "stepfun-ai/step-3.7-flash"},
+    "nvidia": {"base_url": "https://integrate.api.nvidia.com/v1", "model": "stepfun-ai/step-3.5-flash"},
     "sensenova": {"base_url": "https://token.sensenova.cn/v1", "model": "deepseek-v4-flash"},
 }
 
+# 模型级 rate limiter（进程内，重启重置）
+_RATE_LIMITER: Dict[str, List[float]] = {}
+
+def _check_rate_limit(provider: str, model: str, limit: int, window_minutes: int) -> bool:
+    if limit <= 0:
+        return True
+    key = f"{provider}/{model}"
+    now = time.time()
+    window_sec = window_minutes * 60
+    timestamps = _RATE_LIMITER.get(key, [])
+    timestamps = [t for t in timestamps if now - t < window_sec]
+    _RATE_LIMITER[key] = timestamps
+    return len(timestamps) < limit
+
+def _record_usage(provider: str, model: str):
+    key = f"{provider}/{model}"
+    _RATE_LIMITER.setdefault(key, []).append(time.time())
+
 def _get_active_provider() -> str:
     """从 DB 读取活跃供应商，优先取 is_default=True；DB 不可用时回退环境变量"""
     try:
@@ -61,6 +81,8 @@ def _get_provider_config(provider: Optional[str] = None) -> dict:
     db_model = None
     db_base_url = None
     db_api_key = None
+    db_rate_limit = 0
+    db_rate_window = 300
     try:
         from ..database import SessionLocal
         from ..models import LLMConfig
@@ -70,6 +92,8 @@ def _get_provider_config(provider: Optional[str] = None) -> dict:
             db_model = cfg.model
             db_base_url = cfg.base_url
             db_api_key = cfg.api_key
+            db_rate_limit = cfg.rate_limit or 0
+            db_rate_window = cfg.rate_limit_window_minutes or 300
         db.close()
     except Exception:
         pass
@@ -82,6 +106,8 @@ def _get_provider_config(provider: Optional[str] = None) -> dict:
         "api_key": api_key,
         "model": db_model or fb.get("model", ""),
         "base_url": db_base_url or fb.get("base_url", ""),
+        "rate_limit": db_rate_limit,
+        "rate_limit_window_minutes": db_rate_window,
     }
 
 def _get_db_defaults(provider: Optional[str] = None) -> dict:
@@ -139,14 +165,25 @@ def call_llm(
     if not provider and os.getenv("LLM_TASK_PROVIDER"):
         task_provider = os.getenv("LLM_TASK_PROVIDER")
         providers_to_try = [task_provider] + [p for p in providers_to_try if p != task_provider]
+    # LLM_TASK_MODEL 环境变量可覆盖任务级别的模型选择
+    if not model and os.getenv("LLM_TASK_MODEL"):
+        model = os.getenv("LLM_TASK_MODEL")
     last_error = None
     for p in providers_to_try:
         try:
             cfg = _get_provider_config(p)
+            actual_model = model or cfg["model"]
+            # Rate limit check
+            rl = cfg.get("rate_limit", 0)
+            rw = cfg.get("rate_limit_window_minutes", 300)
+            if not _check_rate_limit(p, actual_model, rl, rw):
+                logger.warning(f"[LLM] {p}/{actual_model} rate limit reached ({rl}/{rw}min) → 尝试下一个")
+                last_error = LLMError(f"{p}/{actual_model} rate limit reached")
+                continue
             endpoint = f"{cfg['base_url'].rstrip('/')}/chat/completions"
             headers = {"Authorization": f"Bearer {cfg['api_key']}", "Content-Type": "application/json"}
             payload = {
-                "model": model or cfg["model"],
+                "model": actual_model,
                 "messages": [{"role": "system", "content": system_prompt}, {"role": "user", "content": prompt}],
                 "temperature": temperature, "max_tokens": max_tokens, "top_p": top_p,
                 "frequency_penalty": frequency_penalty, "presence_penalty": presence_penalty,
@@ -176,6 +213,7 @@ def call_llm(
                             if delta.get('content'): content_parts.append(delta['content'])
                             if delta.get('reasoning_content'): reasoning_parts.append(delta['reasoning_content'])
                         except Exception: continue
+                _record_usage(p, actual_model)
                 return "".join(content_parts) or "".join(reasoning_parts)
             else:
                 data = resp.json()
@@ -186,6 +224,7 @@ def call_llm(
                     if rc:
                         parts = [p.strip() for p in rc.replace('\n', '。').split('。') if p.strip()]
                         content = parts[-1] if parts else rc
+                _record_usage(p, actual_model)
                 return content.strip() or ''
         except LLMError as e:
             if '429' in str(e):

platform/backend/app/core/scheduler.py

@@ -20,7 +20,7 @@ from .collector import run_collector_blocking
 logger = logging.getLogger(__name__)
 
 def _set_task_llm_provider(module_id: str):
-    """从 TaskConfig 读取 llm_provider 并设为环境变量，供子进程和 call_llm 读取"""
+    """从 TaskConfig 读取 llm_provider 和 llm_model 并设为环境变量，供子进程和 call_llm 读取"""
     try:
         from ..database import SessionLocal
         from ..models import TaskConfig
@@ -32,10 +32,20 @@ def _set_task_llm_provider(module_id: str):
             if provider:
                 os.environ["LLM_TASK_PROVIDER"] = provider
                 logger.debug("[%s] LLM provider set to %s", module_id, provider)
+            else:
+                os.environ.pop("LLM_TASK_PROVIDER", None)
+
+            model = cfg.params.get("llm_model")
+            if model:
+                os.environ["LLM_TASK_MODEL"] = model
+                logger.debug("[%s] LLM model set to %s", module_id, model)
+            else:
+                os.environ.pop("LLM_TASK_MODEL", None)
             return
     except Exception:
         pass
     os.environ.pop("LLM_TASK_PROVIDER", None)
+    os.environ.pop("LLM_TASK_MODEL", None)
 
 MODULES = {
     "scheduled_fetch_trends": {"name": "🔥 热点趋势", "cron": "01:10"},

platform/backend/app/database.py

@@ -48,12 +48,15 @@ def init_db():
             conn.execute(text("ALTER TABLE llm_configs ADD COLUMN IF NOT EXISTS base_url VARCHAR"))
             conn.execute(text("ALTER TABLE llm_configs ADD COLUMN IF NOT EXISTS api_key VARCHAR"))
             conn.execute(text("ALTER TABLE llm_configs ADD COLUMN IF NOT EXISTS is_default BOOLEAN DEFAULT FALSE"))
+            conn.execute(text("ALTER TABLE llm_configs ADD COLUMN IF NOT EXISTS rate_limit INTEGER DEFAULT 0"))
+            conn.execute(text("ALTER TABLE llm_configs ADD COLUMN IF NOT EXISTS rate_limit_window_minutes INTEGER DEFAULT 300"))
             try:
                 conn.execute(text("ALTER TABLE articles ADD COLUMN IF NOT EXISTS images JSON DEFAULT '{}'::json"))
             except Exception:
                 conn.execute(text("ALTER TABLE articles ADD COLUMN IF NOT EXISTS images TEXT DEFAULT '{}'"))
             for table, col, typ in [
                 ("users", "org_id", "VARCHAR DEFAULT 'default'"),
+                ("articles", "updated_at", "TIMESTAMP"),
                 ("topics", "org_id", "VARCHAR DEFAULT 'default'"),
                 ("topics", "reviewed_at", "TIMESTAMP"),
                 ("platform_configs", "requires_image", "BOOLEAN DEFAULT FALSE"),

platform/backend/app/initial_data.py

@@ -35,14 +35,26 @@ def import_initial_data():
             db.commit()
             print(f"✅ 创建默认管理员: {DEFAULT_ADMIN_USERNAME}")
 
-        # 补充或更新 LLM 供应商配置（nvidia 为主，opencode-go 为备）
+        # 补充或更新 LLM 供应商配置（nvidia 为主，sensenova 为备，含多模型）
         expected = {
             "opencode-go": dict(provider="opencode-go", model="deepseek-v4-flash",
                 base_url="https://opencode.ai/zen/go/v1", temperature=0.7, max_tokens=131072, is_active=True,
                 user_prompt_template="将以下内容扩展为完整文章：\n{topic_title}\n{core_concept}"),
-            "nvidia": dict(provider="nvidia", model="stepfun-ai/step-3.5-flash",
+            "nvidia": dict(provider="nvidia", model="qwen/qwen3.5-397b-a17b",
                 base_url="https://integrate.api.nvidia.com/v1", temperature=0.5, max_tokens=131072, is_active=False,
                 user_prompt_template="将以下内容扩展为完整章节：\n{section_content}"),
+            "sensenova-6.7-flash-lite": dict(provider="sensenova", model="sensenova-6.7-flash-lite",
+                base_url="https://token.sensenova.cn/v1", temperature=0.3, max_tokens=16384, is_active=True,
+                rate_limit=1500, rate_limit_window_minutes=300,
+                user_prompt_template="将以下内容扩展为完整章节：\n{section_content}"),
+            "sensenova-u1-fast": dict(provider="sensenova", model="sensenova-u1-fast",
+                base_url="https://token.sensenova.cn/v1", temperature=0.3, max_tokens=16384, is_active=True,
+                rate_limit=1500, rate_limit_window_minutes=300,
+                user_prompt_template="根据以下内容生成信息图：\n{section_content}"),
+            "sensenova-deepseek": dict(provider="sensenova", model="deepseek-v4-flash",
+                base_url="https://token.sensenova.cn/v1", temperature=0.3, max_tokens=16384, is_active=True,
+                rate_limit=500, rate_limit_window_minutes=300,
+                user_prompt_template="将以下内容扩展为完整章节：\n{section_content}"),
         }
         existing = {c.name: c for c in db.query(LLMConfig).all()}
         for name, cfg in expected.items():
@@ -90,7 +102,7 @@ def import_initial_data():
                     "platform": "zhihu",
                     "name": "知乎",
                     "icon": "🔍",
-                    "default_format": "长文深度分析，1500-3000字，有数据支撑",
+                    "default_format": "深度分析 3000-8000字，数据驱动",
                     "compliance_rules": {
                         "max_title_len": 100,
                         "allowed_tags": ["科技", "生活", "职场", "教育", "可持续", "AI", "远程工作", "个人成长"],
@@ -102,14 +114,14 @@ def import_initial_data():
                     "image_count_max": 0,
                     "image_width": 0,
                     "image_height": 0,
-                    "min_words": 1500,
+                    "min_words": 3000,
-                    "max_words": 3000
+                    "max_words": 8000
                 },
                 {
                     "platform": "wechat",
                     "name": "微信公众号",
                     "icon": "💚",
-                    "default_format": "公众号图文，800-1500字，亲切口语化",
+                    "default_format": "个人叙事 2000-4000字，对话感",
                     "compliance_rules": {
                         "max_title_len": 32,
                         "allowed_tags": ["科技", "生活", "职场", "教育", "可持续", "AI", "远程工作", "个人成长"],
@@ -121,14 +133,14 @@ def import_initial_data():
                     "image_count_max": 3,
                     "image_width": 1080,
                     "image_height": 1080,
-                    "min_words": 800,
+                    "min_words": 2000,
-                    "max_words": 1500
+                    "max_words": 4000
                 },
                 {
                     "platform": "xiaohongshu",
                     "name": "小红书",
                     "icon": "📕",
-                    "default_format": "图文笔记，300-800字，emoji+标签",
+                    "default_format": "精炼干货 400-1000字，实用优先",
                     "compliance_rules": {
                         "max_title_len": 50,
                         "allowed_tags": ["生活方式", "可持续", "AI", "个人成长", "极简", "环保"],
@@ -140,14 +152,29 @@ def import_initial_data():
                     "image_count_max": 6,
                     "image_width": 1080,
                     "image_height": 1440,
-                    "min_words": 300,
+                    "min_words": 400,
-                    "max_words": 800
+                    "max_words": 1000
                 }
             ]
             for p in platforms:
                 db.add(PlatformConfig(**p))
             db.commit()
             print("✅ 插入平台配置")
+        else:
+            # 更新已有平台配置的字数要求（迁移：2026-06 内容质量升级）
+            platform_updates = {
+                "zhihu": {"min_words": 3000, "max_words": 8000, "default_format": "深度分析 3000-8000字，数据驱动"},
+                "wechat": {"min_words": 2000, "max_words": 4000, "default_format": "个人叙事 2000-4000字，对话感"},
+                "xiaohongshu": {"min_words": 400, "max_words": 1000, "default_format": "精炼干货 400-1000字，实用优先"},
+            }
+            for p in db.query(PlatformConfig).all():
+                if p.platform in platform_updates:
+                    up = platform_updates[p.platform]
+                    p.min_words = up["min_words"]
+                    p.max_words = up["max_words"]
+                    p.default_format = up["default_format"]
+            db.commit()
+            print("✅ 平台配置已更新（字数/格式）")
 
         if db.query(TopicField).count() == 0:
             fields = [

platform/backend/app/models.py

@@ -288,7 +288,8 @@ class Article(Base):
     html_content = Column(Text)
     word_count = Column(Integer, nullable=True)
     outline = Column(Text, nullable=True)
-    images = Column(JSON, default=dict)  # {"cover": "/path/to/cover.png", "chart": "/path/to/chart.png"}
+    images = Column(JSON, default=dict)
+    updated_at = Column(DateTime(timezone=True), onupdate=func.now())  # {"cover": "/path/to/cover.png", "chart": "/path/to/chart.png"}
 
 
 class PublishRecord(Base):
@@ -628,6 +629,8 @@ class LLMConfig(Base):
     api_key = Column(String, nullable=True)
     is_active = Column(Boolean, default=True)
     is_default = Column(Boolean, default=False)
+    rate_limit = Column(Integer, default=0, comment="每个时间窗口的调用上限，0=不限")
+    rate_limit_window_minutes = Column(Integer, default=300, comment="时间窗口（分钟），默认5小时")
     created_at = Column(DateTime(timezone=True), server_default=func.now())
     updated_at = Column(DateTime(timezone=True), onupdate=func.now())
 
@@ -645,6 +648,8 @@ class LLMConfig(Base):
             "api_key": f"{self.api_key[:8]}..." if self.api_key else None,
             "is_active": self.is_active,
             "is_default": self.is_default,
+            "rate_limit": self.rate_limit,
+            "rate_limit_window_minutes": self.rate_limit_window_minutes,
             "created_at": self.created_at.isoformat() if self.created_at else None,
             "updated_at": self.updated_at.isoformat() if self.updated_at else None,
         }

platform/backend/app/schemas.py

@@ -523,6 +523,8 @@ class LLMConfigBase(BaseModel):
     api_key: Optional[str] = None
     is_active: bool = True
     is_default: bool = False
+    rate_limit: int = 0
+    rate_limit_window_minutes: int = 300
 
 
 class LLMConfigResponse(LLMConfigBase):

platform/frontend/admin.html

@@ -13,7 +13,7 @@
     <script src="uni-nav.js"></script>
 </head>
 <body>
-    <div id="app">
+    <div id="app" v-cloak>
         <uni-nav title="系统管理" :username="currentUser.username" :is-admin="isAdmin" current-page="admin" @navigate="redirectToPage" @logout="logout">
         </uni-nav>
         <div class="main-content">
@@ -32,6 +32,7 @@
                             <el-button size="default" :type="activeTab === 'logs' ? 'primary' : ''" @click="switchTab('logs')">运行日志</el-button>
                             <el-button size="default" :type="activeTab === 'assistant' ? 'primary' : ''" @click="switchTab('assistant')">AI 助手</el-button>
                             <el-button size="default" :type="activeTab === 'searchproviders' ? 'primary' : ''" @click="switchTab('searchproviders')">搜索API</el-button>
+                            <el-button size="default" :type="activeTab === 'configitems' ? 'primary' : ''" @click="switchTab('configitems')">配置管理</el-button>
                         </div>
                     </div>
 
@@ -136,6 +137,9 @@
                                 </el-table-column>
                                 <el-table-column prop="temperature" label="温度" width="70"></el-table-column>
                                  <el-table-column prop="max_tokens" label="最大Token" width="100"></el-table-column>
+                                 <el-table-column label="调用上限" width="110">
+                                     <template #default="scope">{{ scope.row.rate_limit ? scope.row.rate_limit + '次/' + (scope.row.rate_limit_window_minutes || 300) + '分' : '不限' }}</template>
+                                 </el-table-column>
                                  <el-table-column prop="is_active" label="状态" width="80">
                                     <template #default="scope">
                                         <el-tag v-if="scope.row.is_active" type="success" size="small" effect="dark">启用</el-tag>
@@ -164,6 +168,7 @@
                                 <div class="card-row"><span class="card-label">供应商</span><span class="card-value">{{ item.provider }}</span></div>
                                 <div class="card-row"><span class="card-label">模型</span><span class="card-value">{{ item.model }}</span></div>
                                 <div class="card-row"><span class="card-label">温度/最大Token</span><span class="card-value">{{ item.temperature }} / {{ item.max_tokens }}</span></div>
+                                <div class="card-row"><span class="card-label">调用上限</span><span class="card-value">{{ item.rate_limit ? item.rate_limit + '次/' + (item.rate_limit_window_minutes || 300) + '分' : '不限' }}</span></div>
                                 <div class="card-row"><span class="card-label">状态</span><span class="card-value"><el-tag v-if="item.is_active" type="success" size="small" effect="dark">当前使用</el-tag><el-tag v-else type="info" size="small">未启用</el-tag></span></div>
                                 <div class="card-actions">
                                     <el-button size="small" @click="showLLMConfigDialog(item)">编辑</el-button>
@@ -601,19 +606,160 @@
                         </div>
                         <template #footer><el-button @click="searchProviderTestVisible = false">关闭</el-button></template>
                     </el-dialog>
+
+                    <div v-if="activeTab === 'configitems'">
+                        <div class="filter-bar">
+                            <el-button size="default" :type="configSubTab === 'sensitive' ? 'primary' : ''" @click="switchConfigSubTab('sensitive')">🔒 敏感词</el-button>
+                            <el-button size="default" :type="configSubTab === 'cleanrules' ? 'primary' : ''" @click="switchConfigSubTab('cleanrules')">🧹 清理规则</el-button>
+                        </div>
+
+                        <template v-if="configSubTab === 'sensitive'">
+                            <div class="toolbar">
+                                <el-button type="primary" size="small" @click="showAddSensitiveWord">新增敏感词</el-button>
+                                <span style="font-size:13px;color:#909399;margin-left:8px;">共 {{ sensitiveWords.length }} 个</span>
+                            </div>
+                            <div v-if="swLoading" class="card-loading">加载中...</div>
+                            <template v-else-if="sensitiveWords.length === 0">
+                                <div class="empty-state">
+                                    <el-icon style="font-size:48px;color:#c0c4cc;"><IconWarning /></el-icon>
+                                    <div class="empty-text">暂无敏感词</div>
+                                </div>
+                            </template>
+                            <template v-else>
+                                <el-table :data="sensitiveWords" border stripe class="data-table" style="width:100%">
+                                    <el-table-column type="index" label="#" width="50"></el-table-column>
+                                    <el-table-column prop="word" label="敏感词" min-width="160"></el-table-column>
+                                    <el-table-column prop="category" label="类别" width="110">
+                                        <template #default="{ row }">
+                                            <el-tag size="small">{{ row.category || '未分类' }}</el-tag>
+                                        </template>
+                                    </el-table-column>
+                                    <el-table-column label="操作" width="120">
+                                        <template #default="{ row }">
+                                            <el-button size="small" @click="editSensitiveWord(row)">编辑</el-button>
+                                            <el-button size="small" type="danger" @click="deleteSensitiveWord(row)">删除</el-button>
+                                        </template>
+                                    </el-table-column>
+                                </el-table>
+                                <div class="card-list-mobile">
+                                    <div v-for="item in sensitiveWords" :key="item.id" class="card-item">
+                                        <div><el-tag size="small">{{ item.category || '未分类' }}</el-tag> <strong>{{ item.word }}</strong></div>
+                                        <div style="margin-top:4px;">
+                                            <el-button size="small" @click="editSensitiveWord(item)">编辑</el-button>
+                                            <el-button size="small" type="danger" @click="deleteSensitiveWord(item)">删除</el-button>
+                                        </div>
+                                    </div>
+                                </div>
+                            </template>
+                        </template>
+
+                        <template v-if="configSubTab === 'cleanrules'">
+                            <div class="toolbar">
+                                <el-button type="primary" size="small" @click="showAddCleanRule">新增清理规则</el-button>
+                                <span style="font-size:13px;color:#909399;margin-left:8px;">共 {{ contentCleanRules.length }} 条</span>
+                            </div>
+                            <div v-if="ccrLoading" class="card-loading">加载中...</div>
+                            <template v-else-if="contentCleanRules.length === 0">
+                                <div class="empty-state">
+                                    <el-icon style="font-size:48px;color:#c0c4cc;"><IconWarning /></el-icon>
+                                    <div class="empty-text">暂无清理规则</div>
+                                </div>
+                            </template>
+                            <template v-else>
+                                <el-table :data="contentCleanRules" border stripe class="data-table" style="width:100%">
+                                    <el-table-column type="index" label="#" width="50"></el-table-column>
+                                    <el-table-column prop="rule_type" label="类型" width="110">
+                                        <template #default="{ row }">
+                                            <el-tag size="small">{{ row.rule_type }}</el-tag>
+                                        </template>
+                                    </el-table-column>
+                                    <el-table-column prop="description" label="描述" min-width="140"></el-table-column>
+                                    <el-table-column prop="pattern" label="模式" min-width="220">
+                                        <template #default="{ row }"><code style="font-size:12px;word-break:break-all;">{{ row.pattern }}</code></template>
+                                    </el-table-column>
+                                    <el-table-column label="状态" width="80">
+                                        <template #default="{ row }">
+                                            <el-tag :type="row.is_active ? 'success' : 'info'" size="small">{{ row.is_active ? '启用' : '停用' }}</el-tag>
+                                        </template>
+                                    </el-table-column>
+                                    <el-table-column label="操作" width="120">
+                                        <template #default="{ row }">
+                                            <el-button size="small" @click="editCleanRule(row)">编辑</el-button>
+                                            <el-button size="small" type="danger" @click="deleteCleanRule(row)">删除</el-button>
+                                        </template>
+                                    </el-table-column>
+                                </el-table>
+                                <div class="card-list-mobile">
+                                    <div v-for="item in contentCleanRules" :key="item.id" class="card-item">
+                                        <div><el-tag size="small">{{ item.rule_type }}</el-tag> <code style="font-size:12px;">{{ item.pattern }}</code></div>
+                                        <div style="margin-top:4px;">{{ item.description }}</div>
+                                        <div style="margin-top:4px;">
+                                            <el-tag :type="item.is_active ? 'success' : 'info'" size="small">{{ item.is_active ? '启用' : '停用' }}</el-tag>
+                                            <el-button size="small" style="margin-left:8px;" @click="editCleanRule(item)">编辑</el-button>
+                                            <el-button size="small" type="danger" @click="deleteCleanRule(item)">删除</el-button>
+                                        </div>
+                                    </div>
+                                </div>
+                            </template>
+                        </template>
+                    </div>
                 </div>
             </main>
         </div>
 
+        <el-dialog v-model="swDialogVisible" :title="editingSwId ? '编辑敏感词' : '新增敏感词'" width="450px" :close-on-click-modal="false">
+            <el-form :model="swForm" label-width="80px">
+                <el-form-item label="敏感词"><el-input v-model="swForm.word" placeholder="输入敏感词"/></el-form-item>
+                <el-form-item label="类别">
+                    <el-select v-model="swForm.category" style="width:100%">
+                        <el-option label="general" value="general" />
+                        <el-option label="政治敏感" value="政治敏感" />
+                        <el-option label="违禁内容" value="违禁内容" />
+                        <el-option label="不实信息" value="不实信息" />
+                        <el-option label="porn" value="porn" />
+                        <el-option label="ad" value="ad" />
+                        <el-option label="violence" value="violence" />
+                        <el-option label="other" value="other" />
+                    </el-select>
+                </el-form-item>
+            </el-form>
+            <template #footer>
+                <el-button @click="swDialogVisible = false">取消</el-button>
+                <el-button type="primary" @click="saveSensitiveWord" :loading="swSaving">保存</el-button>
+            </template>
+        </el-dialog>
+
+        <el-dialog v-model="ccrDialogVisible" :title="ccrDialogTitle" width="650px" :close-on-click-modal="false">
+            <el-form :model="ccrForm" label-width="100px">
+                <el-form-item label="类型">
+                    <el-select v-model="ccrForm.rule_type" style="width:100%">
+                        <el-option label="thinking (AI思考)" value="thinking" />
+                        <el-option label="preface (AI前缀)" value="preface" />
+                        <el-option label="verbosity (AI废话)" value="verbosity" />
+                        <el-option label="ai_telltale (AI套话)" value="ai_telltale" />
+                        <el-option label="html_thinking" value="html_thinking" />
+                    </el-select>
+                </el-form-item>
+                <el-form-item label="模式"><el-input v-model="ccrForm.pattern" placeholder="正则表达式"/></el-form-item>
+                <el-form-item label="描述"><el-input v-model="ccrForm.description" placeholder="简短说明"/></el-form-item>
+                <el-form-item label="排序"><el-input-number v-model="ccrForm.sort_order" :min="0" :max="999" style="width:100%"/></el-form-item>
+                <el-form-item label="启用"><el-switch v-model="ccrForm.is_active"/></el-form-item>
+            </el-form>
+            <template #footer>
+                <el-button @click="ccrDialogVisible = false">取消</el-button>
+                <el-button type="primary" @click="saveCleanRule" :loading="ccrSaving">保存</el-button>
+            </template>
+        </el-dialog>
+
         <el-dialog v-model="llmConfigDialogVisible" :title="llmConfigDialogTitle" width="750px" :close-on-click-modal="false">
             <el-form :model="llmConfigForm" label-width="110px">
                 <el-row :gutter="16">
                     <el-col :span="12"><el-form-item label="名称"><el-input v-model="llmConfigForm.name" placeholder="如：default_expand"/></el-form-item></el-col>
                     <el-col :span="12"><el-form-item label="供应商">
                         <el-select v-model="llmConfigForm.provider" style="width:100%">
-                            <el-option label="opencode-go (deepseek-v4-flash)" value="opencode-go"></el-option>
+                            <el-option label="sensenova" value="sensenova"></el-option>
-                            <el-option label="nvidia (step-3.7-flash)" value="nvidia"></el-option>
+                            <el-option label="nvidia" value="nvidia"></el-option>
-                            <el-option label="sensenova (deepseek-v4-flash)" value="sensenova"></el-option>
+                            <el-option label="opencode-go" value="opencode-go"></el-option>
                         </el-select>
                     </el-form-item></el-col>
                 </el-row>
@@ -630,6 +776,10 @@
                     <el-col :span="6"><el-form-item label="激活"><el-switch v-model="llmConfigForm.is_active"/></el-form-item></el-col>
                     <el-col :span="6"><el-form-item label="默认"><el-switch v-model="llmConfigForm.is_default"/></el-form-item></el-col>
                 </el-row>
+                <el-row :gutter="16">
+                    <el-col :span="12"><el-form-item label="调用上限"><el-input-number v-model="llmConfigForm.rate_limit" :min="0" :max="999999" style="width:100%"/></el-form-item></el-col>
+                    <el-col :span="12"><el-form-item label="时间窗口(分)"><el-input-number v-model="llmConfigForm.rate_limit_window_minutes" :min="1" :max="43200" style="width:100%"/></el-form-item></el-col>
+                </el-row>
                 <el-row :gutter="16">
                     <el-col :span="24"><el-form-item label="系统提示词"><el-input type="textarea" v-model="llmConfigForm.system_prompt" :rows="3" placeholder="你是一个专业的内容创作助手。"/></el-form-item></el-col>
                 </el-row>
@@ -692,6 +842,7 @@
                 };
 
                 const activeTab = ref('users');
+                const configSubTab = ref('sensitive');
                 const currentUser = ref({ username: '' });
                 const isAdmin = ref(false);
 
@@ -711,7 +862,7 @@ const llmConfigs = ref([]);
                 const llmConfigsLoading = ref(false);
                 const llmConfigDialogVisible = ref(false);
                 const llmConfigDialogTitle = ref('新增 LLM 配置');
-                const llmConfigForm = reactive({ id: null, name: '', system_prompt: '', user_prompt_template: '', temperature: 0.7, max_tokens: 131072, model: '', provider: 'opencode-go', base_url: '', api_key: '', is_active: true, is_default: false });
+                const llmConfigForm = reactive({ id: null, name: '', system_prompt: '', user_prompt_template: '', temperature: 0.7, max_tokens: 131072, model: '', provider: 'opencode-go', base_url: '', api_key: '', is_active: true, is_default: false, rate_limit: 0, rate_limit_window_minutes: 300 });
                 const editingLLMConfigId = ref(null);
                 const loadLLMConfigs = async () => {
                     llmConfigsLoading.value = true;
@@ -727,6 +878,7 @@ const llmConfigs = ref([]);
                         llmConfigForm.id = null; llmConfigForm.name = ''; llmConfigForm.provider = 'sensenova'; llmConfigForm.model = 'deepseek-v4-flash';
                         llmConfigForm.base_url = 'https://token.sensenova.cn/v1'; llmConfigForm.api_key = ''; llmConfigForm.temperature = 0.3;
                         llmConfigForm.max_tokens = 4000; llmConfigForm.system_prompt = ''; llmConfigForm.user_prompt_template = ''; llmConfigForm.is_active = true; llmConfigForm.is_default = false;
+                        llmConfigForm.rate_limit = 0; llmConfigForm.rate_limit_window_minutes = 300;
                     }
                     llmConfigDialogVisible.value = true;
                 };
@@ -1078,6 +1230,87 @@ const llmConfigs = ref([]);
                     catch (e) { ElMessage.error('重置失败: ' + e.message); }
                 };
 
+                const sensitiveWords = ref([]);
+                const swLoading = ref(false);
+                const swDialogVisible = ref(false);
+                const swSaving = ref(false);
+                const swForm = reactive({ word: '', category: 'general' });
+                const editingSwId = ref(null);
+                const loadSensitiveWords = async () => {
+                    swLoading.value = true;
+                    try { sensitiveWords.value = await api.get('/api/admin/config/sensitive-words'); }
+                    catch (e) { console.error(e); }
+                    finally { swLoading.value = false; }
+                };
+                const showAddSensitiveWord = () => { swForm.word = ''; swForm.category = 'general'; editingSwId.value = null; swDialogVisible.value = true; };
+                const editSensitiveWord = (row) => { swForm.word = row.word; swForm.category = row.category || 'general'; editingSwId.value = row.id; swDialogVisible.value = true; };
+                const saveSensitiveWord = async () => {
+                    swSaving.value = true;
+                    try {
+                        const body = { word: swForm.word, category: swForm.category };
+                        if (editingSwId.value) { await api.put('/api/admin/config/sensitive-words/' + editingSwId.value, body); ElMessage.success('更新成功'); }
+                        else { await api.post('/api/admin/config/sensitive-words', body); ElMessage.success('添加成功'); }
+                        swDialogVisible.value = false;
+                        await loadSensitiveWords();
+                    }
+                    catch (e) { ElMessage.error('保存失败: ' + e.message); }
+                    finally { swSaving.value = false; }
+                };
+                const deleteSensitiveWord = async (row) => {
+                    try { await ElMessageBox.confirm('确定删除「' + row.word + '」吗？', '提示', { type: 'warning' }); await api.delete('/api/admin/config/sensitive-words/' + row.id); ElMessage.success('删除成功'); await loadSensitiveWords(); }
+                    catch (e) { if (e !== 'cancel') ElMessage.error('删除失败: ' + e.message); }
+                };
+                const sensitiveWordsByCategory = computed(() => {
+                    const map = {};
+                    for (const w of sensitiveWords.value) {
+                        const cat = w.category || '未分类';
+                        if (!map[cat]) map[cat] = [];
+                        map[cat].push(w);
+                    }
+                    return Object.entries(map).sort((a, b) => a[0].localeCompare(b[0]));
+                });
+
+                const contentCleanRules = ref([]);
+                const ccrLoading = ref(false);
+                const ccrDialogVisible = ref(false);
+                const ccrDialogTitle = ref('新增清理规则');
+                const ccrSaving = ref(false);
+                const ccrForm = reactive({ id: null, rule_type: 'ai_telltale', pattern: '', description: '', sort_order: 0, is_active: true });
+                const editingCcrId = ref(null);
+                const loadContentCleanRules = async () => {
+                    ccrLoading.value = true;
+                    try { contentCleanRules.value = await api.get('/api/admin/config/content-clean-rules'); }
+                    catch (e) { console.error(e); }
+                    finally { ccrLoading.value = false; }
+                };
+                const showAddCleanRule = () => {
+                    ccrDialogTitle.value = '新增清理规则';
+                    editingCcrId.value = null;
+                    ccrForm.id = null; ccrForm.rule_type = 'ai_telltale'; ccrForm.pattern = ''; ccrForm.description = ''; ccrForm.sort_order = 0; ccrForm.is_active = true;
+                    ccrDialogVisible.value = true;
+                };
+                const editCleanRule = (row) => {
+                    ccrDialogTitle.value = '编辑清理规则';
+                    editingCcrId.value = row.id;
+                    Object.assign(ccrForm, { id: row.id, rule_type: row.rule_type, pattern: row.pattern, description: row.description || '', sort_order: row.sort_order || 0, is_active: row.is_active });
+                    ccrDialogVisible.value = true;
+                };
+                const saveCleanRule = async () => {
+                    ccrSaving.value = true;
+                    try {
+                        const body = { rule_type: ccrForm.rule_type, pattern: ccrForm.pattern, description: ccrForm.description, sort_order: ccrForm.sort_order, is_active: ccrForm.is_active };
+                        if (editingCcrId.value) { await api.put('/api/admin/config/content-clean-rules/' + editingCcrId.value, body); ElMessage.success('更新成功'); }
+                        else { await api.post('/api/admin/config/content-clean-rules', body); ElMessage.success('创建成功'); }
+                        ccrDialogVisible.value = false;
+                        await loadContentCleanRules();
+                    } catch (e) { ElMessage.error('保存失败: ' + e.message); }
+                    finally { ccrSaving.value = false; }
+                };
+                const deleteCleanRule = async (row) => {
+                    try { await ElMessageBox.confirm('确定删除该规则吗？', '提示', { type: 'warning' }); await api.delete('/api/admin/config/content-clean-rules/' + row.id); ElMessage.success('删除成功'); await loadContentCleanRules(); }
+                    catch (e) { if (e !== 'cancel') ElMessage.error('删除失败: ' + e.message); }
+                };
+
                 const formatDate = (dateStr) => { if (!dateStr) return '-'; return new Date(dateStr.replace(' ', 'T')).toLocaleString('zh-CN', { year: 'numeric', month: '2-digit', day: '2-digit', hour: '2-digit', minute: '2-digit' }); };
                 const logout = () => { localStorage.removeItem('authToken'); localStorage.removeItem('userRole'); localStorage.removeItem('currentUser'); window.location.href = '/login.html'; };
 
@@ -1086,6 +1319,7 @@ const llmConfigs = ref([]);
                     users: fetchUsers,
                     logs: loadLogTypes, orgs: loadOrgs, roles: loadRoles, menus: loadMenus, assistant: loadAssistantConfig,
                     searchproviders: loadSearchProviders,
+                    configitems: () => { loadSensitiveWords(); loadContentCleanRules(); },
                 };
                 const loadedTabs = new Set([]);
 
@@ -1098,6 +1332,7 @@ const llmConfigs = ref([]);
                         else console.warn('[switchTab] no loader for', name);
                     }
                 };
+                const switchConfigSubTab = (name) => { configSubTab.value = name; };
 
                 onMounted(() => {
                     const hash = window.location.hash.replace('#tab=', '');
@@ -1140,6 +1375,9 @@ const llmConfigs = ref([]);
                     searchProviderTestVisible, searchProviderTesting, searchProviderTestResult,
                     loadSearchProviders, addSearchProvider, editSearchProvider, saveSearchProvider, deleteSearchProvider, updateSearchProvider,
                     testSearchProvider, testAllSearchProviders, resetSearchUsage,
+                    sensitiveWords, swLoading, swDialogVisible, swSaving, swForm, editingSwId, loadSensitiveWords, showAddSensitiveWord, editSensitiveWord, saveSensitiveWord, deleteSensitiveWord, configSubTab, switchConfigSubTab,
+                    contentCleanRules, ccrLoading, ccrDialogVisible, ccrDialogTitle, ccrSaving, ccrForm, editingCcrId,
+                    loadContentCleanRules, showAddCleanRule, editCleanRule, saveCleanRule, deleteCleanRule,
                 };
             }
         });

platform/frontend/articles.html

@@ -23,7 +23,7 @@
     <script src="uni-nav.js"></script>
 </head>
 <body>
-    <div id="app">
+    <div id="app" v-cloak>
         <uni-nav title="文章管理" :username="currentUser.username" :is-admin="isAdmin" current-page="articles" @navigate="redirectToPage" @logout="handleLogout">
         </uni-nav>
         <div class="main-content">

platform/frontend/assets.html

@@ -24,7 +24,7 @@
     <script src="uni-nav.js"></script>
 </head>
 <body>
-    <div id="app">
+    <div id="app" v-cloak>
         <uni-nav title="素材库" :username="currentUser.username" :is-admin="isAdmin" current-page="assets" @navigate="redirectToPage" @logout="handleLogout">
         </uni-nav>
         <div class="main-content">

platform/frontend/calendar.html

@@ -87,7 +87,7 @@
     <script src="uni-nav.js"></script>
 </head>
 <body>
-    <div id="app">
+    <div id="app" v-cloak>
         <uni-nav title="内容日历" :username="currentUser.username" :is-admin="isAdmin" current-page="calendar" @navigate="redirectToPage" @logout="handleLogout">
         </uni-nav>
         <div class="main-content">

platform/frontend/index.html

@@ -36,7 +36,7 @@
     <script src="uni-nav.js"></script>
 </head>
 <body>
-    <div id="app">
+    <div id="app" v-cloak>
         <uni-nav title="仪表盘" :username="currentUser.username" :is-admin="isAdmin" current-page="dashboard" @navigate="redirectToPage" @logout="handleLogout">
         </uni-nav>
 <div class="main-content" v-if="isLoggedIn">

platform/frontend/metrics.html

@@ -41,7 +41,7 @@
     <script src="chart.umd.min.js"></script>
 </head>
 <body>
-    <div id="app">
+    <div id="app" v-cloak>
         <uni-nav title="数据分析" :username="currentUser.username" :is-admin="isAdmin" current-page="metrics" @navigate="redirectToPage" @logout="handleLogout">
         </uni-nav>
         <div class="main-content">

platform/frontend/platforms.html

@@ -38,7 +38,7 @@
     <script src="uni-nav.js"></script>
 </head>
 <body>
-    <div id="app">
+    <div id="app" v-cloak>
         <uni-nav title="平台配置" :username="currentUser.username" :is-admin="isAdmin" current-page="platforms" @navigate="redirectToPage" @logout="handleLogout">
         </uni-nav>
         <div class="main-content">

platform/frontend/tasks.html

@@ -65,7 +65,7 @@
     <script src="uni-nav.js"></script>
 </head>
 <body>
-    <div id="app">
+    <div id="app" v-cloak>
         <uni-nav title="任务管理" :username="currentUser.username" :is-admin="isAdmin" current-page="tasks" @navigate="redirectToPage" @logout="handleLogout">
         </uni-nav>
         <div class="main-content">
@@ -103,7 +103,7 @@
                                     <div><span>最后运行</span><span>{{ mod.last_run || '从未' }}<span v-if="mod.last_status" :style="{marginLeft:'6px',padding:'1px 6px',borderRadius:'8px',fontSize:'11px',fontWeight:500}"><span v-if="mod.last_status==='success'" style="color:#67c23a;">✅成功</span><span v-else-if="mod.last_status==='failed'" style="color:#f56c6c;">❌失败</span><span v-else-if="mod.last_status==='running'" style="color:#e6a23c;">⏳运行中</span></span></span></div>
                                     <div><span>下次运行</span><span>{{ mod.next_run || '—' }}</span></div>
                                     <div><span>累计运行</span><span>{{ mod.total_runs }} 次 <span style="color:#67c23a;">{{ mod.success_runs }} 成功</span> <span style="color:#f56c6c;">{{ mod.failed_runs }} 失败</span><span v-if="mod.running > 0" style="color:#e6a23c;"> {{ mod.running }} 运行中</span></span></div>
-                                    <div><span>LLM 模型</span><span>{{ (mod.params && mod.params.llm_provider) || (defaultLlmLabel || '默认') }}</span></div>
+                                    <div><span>LLM 模型</span><span>{{ llmLabel(mod) }}</span></div>
                                     <div style="margin-top:10px; border-bottom:none;">
                                         <el-button size="small" type="primary" @click.stop="triggerModule(mod.module_id)" :loading="runningModule === mod.module_id" :disabled="!mod.enabled">立即运行</el-button>
                                         <el-button size="small" @click.stop="openModuleDetail(mod)">查看详情</el-button>
@@ -349,15 +349,15 @@
                                     </div>
                                     <div style="display:flex;gap:16px;align-items:center;margin-bottom:12px;">
                                         <span style="font-size:13px;color:#606266;width:60px;">LLM</span>
-                                        <el-select v-model="drawerData.params.llm_provider" size="small" style="width:200px;" clearable placeholder="系统默认" @change="saveModuleConfig" :disabled="savingConfig">
+                                        <el-select v-model="drawerData.params.llm_provider" size="small" style="width:280px;" clearable placeholder="系统默认" @change="onLLMProviderChange" :disabled="savingConfig">
-                                            <el-option v-for="c in llmConfigs.filter(x=>x.is_active)" :key="c.id" :label="c.provider + ' (' + c.model + ')' + (c.is_default ? ' (默认)' : '')" :value="c.provider"/>
+                                            <el-option v-for="c in llmConfigs.filter(x=>x.is_active)" :key="c.id" :label="c.provider + '/' + c.model + (c.rate_limit ? ' (' + c.rate_limit + '次/' + (c.rate_limit_window_minutes||300) + '分)' : '') + (c.is_default ? ' (默认)' : '')" :value="c.provider"/>
                                         </el-select>
-                                        <span style="font-size:12px;color:#909399;">留空则使用系统默认模型</span>
+                                        <span style="font-size:12px;color:#909399;">留空则使用系统默认</span>
                                     </div>
                                 </div>
-                                <div v-if="Object.keys(drawerData.params || {}).filter(k => k !== 'llm_provider').length > 0">
+                                <div v-if="Object.keys(drawerData.params || {}).filter(k => k !== 'llm_provider' && k !== 'llm_model').length > 0">
                                     <div style="font-size:13px;font-weight:600;margin-bottom:12px;">参数配置</div>
-                                    <div v-for="(val, key) in drawerData.params" v-if="key !== 'llm_provider'" :key="key" style="margin-bottom:12px;display:flex;align-items:center;gap:12px;">
+                                    <div v-for="(val, key) in drawerData.params" v-if="key !== 'llm_provider' && key !== 'llm_model'" :key="key" style="margin-bottom:12px;display:flex;align-items:center;gap:12px;">
                                         <span style="font-size:13px;color:#606266;width:120px;">{{ key }}</span>
                                         <el-input v-if="typeof val === 'string'" v-model="drawerData.params[key]" size="small" style="flex:1;" @change="saveModuleConfig" :disabled="savingConfig"></el-input>
                                         <el-input-number v-else-if="typeof val === 'number'" v-model="drawerData.params[key]" size="small" :disabled="savingConfig" @change="saveModuleConfig"></el-input-number>
@@ -672,9 +672,29 @@ const TasksApp = {
                 const configs = await this.api('/api/admin/llmconfigs');
                 this.llmConfigs = configs || [];
                 const def = (configs || []).find(c => c.is_default);
-                this.defaultLlmLabel = def ? def.provider + ' (' + def.model + ')' : '';
+                this.defaultLlmLabel = def ? def.provider + '/' + def.model : '';
             } catch (e) { console.error('loadLLMConfigs error:', e); this.llmConfigs = []; }
         },
+        llmLabel(mod) {
+            const p = (mod.params && mod.params.llm_provider) || '';
+            const m = (mod.params && mod.params.llm_model) || '';
+            if (p && m) return p + '/' + m;
+            if (p) {
+                const match = (this.llmConfigs || []).find(c => c.provider === p && c.is_active);
+                if (match) return p + '/' + match.model;
+            }
+            return this.defaultLlmLabel || '默认';
+        },
+        onLLMProviderChange(val) {
+            if (!val) {
+                delete this.drawerData.params.llm_provider;
+                delete this.drawerData.params.llm_model;
+            } else {
+                const match = (this.llmConfigs || []).find(c => c.provider === val && c.is_active);
+                if (match) this.drawerData.params.llm_model = match.model;
+            }
+            this.saveModuleConfig();
+        },
         async openModuleDetail(mod) {
             this.showDrawer = true;
             this.drawerTitle = mod.title + ' 详情';

platform/frontend/theme-modern.css

@@ -3,6 +3,9 @@
   Inspired by opencode.ai — clean, modern, responsive
   ============================================ */
 
+/* ========== v-cloak (hide raw Vue templates) ========== */
+[v-cloak] { display: none !important; }
+
 /* ========== Design Tokens ========== */
 :root {
   /* Colors */
@@ -116,7 +119,7 @@ body {
   margin: 0 auto;
   min-height: calc(100vh - var(--nav-height));
 }
-.content-area { flex: 1; padding: var(--spacing-lg); overflow-y: auto; }
+.content-area { flex: 1; padding: var(--spacing-md); overflow-y: auto; }
 
 /* ========== Card ========== */
 .card {
@@ -373,7 +376,7 @@ body:has(.preview-dialog-custom.is-fullscreen) > [class*="el-overlay"] { z-index
 
 /* ========== Responsive ========== */
 @media (max-width: 768px) {
-  .content-area { padding: var(--spacing-md) !important; padding-bottom: 80px; }
+  .content-area { padding: var(--spacing-sm) !important; padding-bottom: 80px; }
   .card { padding: var(--spacing-md) !important; }
   .page-header { flex-direction: column; align-items: flex-start !important; gap: var(--spacing-sm) !important; }
 

platform/frontend/topics.html

@@ -32,7 +32,7 @@
     <script src="uni-nav.js"></script>
 </head>
 <body>
-    <div id="app">
+    <div id="app" v-cloak>
         <uni-nav title="选题管理" :username="currentUser.username" :is-admin="isAdmin" current-page="topics" @navigate="redirectToPage" @logout="handleLogout">
         </uni-nav>
         <div class="main-content">

scripts/compliance_checker.py

@@ -55,10 +55,29 @@ AI_TELTALES = [
     "众所周知",
     "毋庸置疑",
     "不知大家有没有发现",
+    "不可否认",
+    "毫无疑义",
+    "从某种意义上",
+    "从某种程度上",
+    "在一定程度上",
+    "换而言之",
+    "换言之",
+    "从本质",
+    "归根结底",
+    "说到底",
+    "这为我们提供了",
+    "为我们提供了宝贵的",
+    "引发了我们",
+    "不得不让人思考",
+    "引人深思",
+    "毫无悬念",
+    "毫无意外",
+    "毫无争议",
 ]
 
 _cached_sensitive_words = None
 _cached_platform_rules = None
+_cached_ai_telltales = None
 
 def _load_sensitive_words():
     global _cached_sensitive_words
@@ -121,6 +140,34 @@ def _load_platform_rules():
     _cached_platform_rules = PLATFORM_RULES
     return _cached_platform_rules
 
+
+def _load_ai_telltales():
+    """从 DB ContentCleanRule 加载 AI 套话模式（rule_type='ai_telltale'），无 DB 时回退硬编码列表"""
+    global _cached_ai_telltales
+    if _cached_ai_telltales is not None:
+        return _cached_ai_telltales
+
+    try:
+        from app.core.prompt_loader import _get_session
+        from app.models import ContentCleanRule
+        session = _get_session()
+        try:
+            rows = session.query(ContentCleanRule).filter(
+                ContentCleanRule.rule_type == 'ai_telltale',
+                ContentCleanRule.is_active == True
+            ).order_by(ContentCleanRule.sort_order).all()
+            if rows:
+                _cached_ai_telltales = [r.pattern for r in rows]
+                return _cached_ai_telltales
+        finally:
+            session.close()
+    except Exception:
+        pass
+
+    _cached_ai_telltales = list(AI_TELTALES)
+    return _cached_ai_telltales
+
+
 class ComplianceChecker:
     """合规审查器"""
 
@@ -349,7 +396,8 @@ class ComplianceChecker:
     def _check_ai_telltales(self, text: str):
         """检查AI套话——正文中出现这些模式说明AI写作痕迹明显"""
         plain = re.sub(r'<[^>]+>', '', text)
-        for pattern in AI_TELTALES:
+        patterns = _load_ai_telltales()
+        for pattern in patterns:
             if re.search(pattern, plain):
                 self.issues.append({
                     "type": "内容质量",
@@ -359,26 +407,16 @@ class ComplianceChecker:
                 })
 
     def _check_pronoun_consistency(self, text: str, platform: str):
-        """检查人称一致性（尤其是微信文章）"""
+        """检查人称一致性"""
-        if platform != "wechat":
-            return
         plain = re.sub(r'<[^>]+>', '', text)
         has_ni = '你' in plain
         has_nimen = '你们' in plain
-        has_women = '我们' in plain
         if has_nimen and has_ni:
             self.issues.append({
                 "type": "内容质量",
                 "category": "人称混用",
-                "detail": "微信文章中同时使用「你」和「你们」，建议统一为「你」",
+                "detail": "同时使用「你」和「你们」，建议统一为「你」",
-                "suggestion": "将所有「你们」替换为「你」，保持与读者的单数对话感"
+                "suggestion": "将「你们」替换为「你」，保持对话感"
-            })
-        if has_women and has_ni:
-            self.issues.append({
-                "type": "内容质量",
-                "category": "人称混用",
-                "detail": "微信文章中同时使用「我们」和「你」，建议统一视角",
-                "suggestion": "将「我们」替换为「你」或「我」，保持与读者对话而非说教"
             })
 
     def _check_reading_experience(self, text: str, platform: str = ""):

scripts/db_helper.py

@@ -89,8 +89,14 @@ def update_topic_status(topic_id: str, status: str, compliance_score: Optional[i
             topic.compliance_score = compliance_score
         if reviewed_at is not None:
             topic.reviewed_at = reviewed_at
-        if status in ['ready', 'published']:
+        if status == 'review' and topic.generated_at is None:
             topic.generated_at = datetime.now()
+            topic.reviewed_at = topic.reviewed_at or datetime.now()
+        if status == 'ready':
+            topic.ready_at = datetime.now().date()
+            topic.reviewed_at = topic.reviewed_at or datetime.now()
+        if status == 'published':
+            topic.published_at = datetime.now().date()
         db.commit()
         return True
     finally:
@@ -243,7 +249,6 @@ def save_article(topic_id: str, platform: str, html_content: str, db: Optional[S
         now = datetime.now()
         if existing:
             existing.html_content = html_content
-            existing.compliance_score = existing.compliance_score
         else:
             article = Article(
                 id=article_id,

scripts/prompt_loader.py

@@ -44,23 +44,33 @@ _PROMPT_DEFAULTS = {
         "temperature": 0.6, "max_tokens": 2000,
         "variables": ["gaps"],
     },
-    "section_expansion": {
+    "section_expansion_zhihu": {
-        "content": "你是一个资深作者，正在写一篇关于「{topic_title}」的文章。请写「{section_title}」这一节。\n\n今天日期：{date}。\n\n笔记要点：\n{content}\n\n输出要求（200-800字，3-6段）：\n- 每段3-5句，段间空行分隔\n- 用「你」或「我」视角，自然口语化\n- 从具体场景或痛点切入，禁止以「引言」「核心观点」「总结」「开头钩子」这类标签开头\n- 每个论点配具体案例或数据，不要空谈道理\n- 回答读者「所以呢」——这对他有什么用\n\n格式：\n- 纯段落文字，不要标题/列表/标记（#、-、*、1.等）\n- 不要出现「首先其次最后」「总的来说」「值得注意的是」「综上所述」等AI套话\n- 来源URL请放在文末括号内",
+        "content": "写「{topic_title}」文章中「{section_title}」这一节。\n\n今天：{date}。\n笔记要点：\n{content}\n\n写400-800字，3-6段。每段围绕一个论点展开：先摆事实/数据，再分析原因，最后说清这跟读者有什么关系。\n\n注意：\n- 每个论点必须配具体数据或案例（至少一个），不要空谈。引用研究、报告或行业事件\n- 用「你」称呼读者，像是在知乎回答里跟人讨论\n- 从具体问题切入，不要以「引言/总结/核心观点」这类标签开头\n- 回答「所以呢」——这条信息对读者有什么用\n- 避免「首先其次最后」「总的来说」「综上所述」「值得注意的是」这些表达\n- 来源放在文末括号内\n- 纯段落，不要列表/标题/序号标记",
-        "temperature": 0.75, "max_tokens": 3000,
+        "temperature": 0.78, "max_tokens": 4000,
+        "variables": ["topic_title", "section_title", "date", "content"],
+    },
+    "section_expansion_wechat": {
+        "content": "写「{topic_title}」文章中「{section_title}」这一节。\n\n今天：{date}。\n笔记要点：\n{content}\n\n写300-500字，用「我」的口吻叙述——像在跟朋友聊天，不是写文章。\n\n注意：\n- 从个人经历或感受切入，不要讲道理\n- 每个观点背后要有一个具体故事或观察\n- 段与段之间有情绪的起承转合，不是逻辑推理\n- 读起来像一个人在说话，而不是一篇文章在论述\n- 避免「首先其次最后」「总的来说」「综上所述」「值得注意的是」这些套路\n- 纯段落，不要标题/列表/序号标记",
+        "temperature": 0.8, "max_tokens": 3000,
+        "variables": ["topic_title", "section_title", "date", "content"],
+    },
+    "section_expansion_xiaohongshu": {
+        "content": "写「{topic_title}」笔记中「{section_title}」这一节。\n\n今天：{date}。\n笔记要点：\n{content}\n\n写100-200字，1-3小段。每段1-2句，直接给干货。\n\n注意：\n- 直接给结论或方法，不要铺垫\n- 可以用emoji（1个/段，不要堆砌）\n- 像博主在分享亲测有效的经验\n- 不要讲道理，只讲「是什么」和「怎么做」\n- 用「你」或直接省略主语\n- 避免「首先其次最后」「总的来说」「综上所述」「值得注意的是」这些套路\n- 纯段落，不要列表/标题/序号标记",
+        "temperature": 0.75, "max_tokens": 2000,
         "variables": ["topic_title", "section_title", "date", "content"],
     },
     "title_optimize_zhihu": {
-        "content": "你是一个知乎高赞标题专家。为以下文章起3个标题。\n\n文章主题：{title}\n核心观点：{core}\n受众痛点：{pain}\n领域：{field}\n\n要求：\n- 信息密度高，包含搜索关键词\n- 优先使用数字、对比、悬念、痛点\n- 20字以内\n- 不要「如何...」开头\n- 有独特判断和立场，能引发讨论\n- 制造「不点开就亏了」的紧迫感\n\n直接输出3个标题，每行一个，不要序号和说明。",
+        "content": "你是知乎老用户，经常上热榜。给这篇文章起3个标题候选：\n\n文章主题：{title}\n核心观点：{core}\n受众痛点：{pain}\n领域：{field}\n\n感觉像在跟人推荐一篇值得辩论的文章。信息要密、20字内、有明确立场、让人想点进来说两句。不要「如何」开头。\n\n每行一个，不要编号。",
         "temperature": 0.8, "max_tokens": 1500,
         "variables": ["title", "core", "pain", "field"],
     },
     "title_optimize_wechat": {
-        "content": "你是一个公众号标题专家。为以下文章起3个标题。\n\n文章主题：{title}\n核心观点：{core}\n\n要求：\n- 包含身份标签（如「打工人」「30岁后」「普通上班族」）\n- 包含情绪钩子（焦虑/反常识/后悔/稀缺）\n- 包含微信SEO关键词\n- 15-25字，口语化\n- 忌笼统，越具体越好\n- 不要感叹号堆砌\n\n参考公式：\n- 「身份+痛点+方案」：打工人学了一堆AI工具，为什么还在加班？\n- 「反常识+数据」：用了AI效率反而更低了？73%的人掉进了这个坑\n- 「结果+身份」：每天省出2小时后，我才发现自己以前有多傻\n\n直接输出3个标题，每行一个，不要序号和说明。",
+        "content": "你是公众号编辑。这篇文章面向{field}领域的读者，帮起3个标题：\n\n文章主题：{title}\n核心观点：{core}\n\n要有身份代入感（打工人/30岁+/普通上班族）、有情绪钩子（焦虑/反常识/后悔）、含微信搜索关键词。15-25字，口语化，越具体越好。参考这种风格：「用了AI效率反而更低了？73%的人掉进了这个坑」\n\n每行一个，不要编号。",
         "temperature": 0.8, "max_tokens": 1500,
-        "variables": ["title", "core"],
+        "variables": ["title", "core", "field"],
     },
     "title_optimize_xhs": {
-        "content": "你是一个小红书标题专家。为以下文章起3个标题。\n\n标题：{title}\n核心观点：{core}\n\n要求：\n- 18字以内\n- 爆款公式：身份/场景+数字+结果，或痛点+方案+反差\n- 包含小红书SEO关键词\n- 1个精确emoji（不要用🔥💥❌❓这几个滥用的）\n- 有场景感、结果感\n- 忌笼统：不要「必看/收藏/码住/绝了」\n\n直接输出3个标题，每行一个，不要序号和说明。",
+        "content": "你是小红书博主。给这篇笔记起3个标题：\n\n笔记主题：{title}\n核心：{core}\n\n每条18字以内，有具体场景和结果感。可以加1个精确emoji（不要🔥💥❌❓）。参考：「用了两年AI工具，推荐这5个就够了」这种风格。\n\n每行一个，不要编号。",
         "temperature": 0.8, "max_tokens": 1500,
         "variables": ["title", "core"],
     },

scripts/writer.py

@@ -43,9 +43,9 @@ logger = logging.getLogger(__name__)
 _md_parser = mistune.create_markdown()
 
 _FALLBACK_PLATFORM_CONFIG = {
-    "zhihu": {"max_chars": 3000, "style": "深度长文分析", "min_chars": 1500},
+    "zhihu": {"max_chars": 8000, "style": "深度长文分析", "min_chars": 3000},
-    "wechat": {"max_chars": 1500, "style": "亲切口语化", "min_chars": 800},
+    "wechat": {"max_chars": 4000, "style": "个人叙事对话感", "min_chars": 2000},
-    "xiaohongshu": {"max_chars": 800, "style": "图文笔记，emoji+标签", "min_chars": 300},
+    "xiaohongshu": {"max_chars": 1000, "style": "图文笔记，精炼实用", "min_chars": 400},
 }
 
 def _load_platform_config() -> dict:
@@ -144,28 +144,30 @@ class Writer:
         bullet_count = sum(1 for l in lines if l.startswith(('- ', '* ', '**', '+ ')))
         return bullet_count / len(lines) > 0.4
 
-    def _expand_section(self, section: Dict) -> str:
+    def _expand_section(self, section: Dict, platform: str = "zhihu") -> str:
         content = section.get('content', '').strip()
-        # 大纲要点格式（>40% 行以 -/*/** 开头）或内容过短（<100字）应由 LLM 展开为连贯段落
+        platform_prompt_key = f"section_expansion_{platform}"
-        if HAVE_LLM and (self._is_bullet_only(content) or (content and len(content) < 100)):
+        if platform not in ("zhihu", "wechat", "xiaohongshu"):
-            logger.info(f"使用 LLM 扩写章节(要点→段落): {section['title']}")
+            platform_prompt_key = "section_expansion_zhihu"
-            prompt = get_prompt("section_expansion",
+
+        if HAVE_LLM and content:
+            logger.info(f"LLM 扩写 [{platform}]: {section['title']} ({len(content)} chars)")
+            prompt = get_prompt(platform_prompt_key,
                 topic_title=self.topic['title'],
                 section_title=section['title'],
                 date=datetime.datetime.now().strftime('%Y年%m月%d日'),
                 content=content,
             )
             try:
-                params = get_prompt_params("section_expansion")
+                params = get_prompt_params(platform_prompt_key) or {"temperature": 0.75, "max_tokens": 3000}
                 expanded = call_llm(prompt, temperature=params.get("temperature", 0.75), max_tokens=params.get("max_tokens", 3000))
                 if expanded:
                     cleaned = self._clean_markdown(expanded.strip())
                     if cleaned:
                         return cleaned
             except Exception as e:
-                logger.warning(f"LLM 扩写失败: {e}")
+                logger.warning(f"LLM 扩写失败 [{platform}]: {e}")
 
-        # Fallback: 将 bullet points 展开为段落（过滤噪音行）
         lines = [l.strip() for l in content.split('\n') if not self._is_outline_noise(l)]
         if lines:
             sentences = []
@@ -185,17 +187,16 @@ class Writer:
                 return self._clean_markdown(result)
         return ''
 
-    def generate_full_markdown(self) -> str:
+    def generate_platform_markdown(self, platform: str = "zhihu") -> str:
         sections = self._parse_outline_sections()
         parts = []
         for sec in sections:
             if sec['level'] == 1:
                 if sec.get('content'):
-                    expanded = self._expand_section(sec)
+                    expanded = self._expand_section(sec, platform)
                     if expanded:
                         parts.append(expanded + "\n")
                 continue
-            # 跳过大纲结构噪音节点
             title_stripped = sec['title'].strip()
             if title_stripped in ('文章大纲', '大纲', '文章结构', '结构'):
                 continue
@@ -204,7 +205,7 @@ class Writer:
             heading = f"{'#' * sec['level']} {sec['title']}"
             parts.append(heading)
             if sec.get('content'):
-                expanded = self._expand_section(sec)
+                expanded = self._expand_section(sec, platform)
                 parts.append(expanded + "\n")
         full_md = "\n".join(parts).strip()
         # 收集所有引用来源，统一添加到文末
@@ -286,8 +287,6 @@ class Writer:
         if platform == "wechat":
             result = []
             for line in lines:
-                # 人称统一：我们→我，你们→你
-                line = line.replace('我们', '我').replace('你们', '你').replace('我', '你')
                 if line.startswith('### '):
                     result.append(f"\n**{line[4:]}**\n")
                 elif line.startswith('## '):
@@ -370,6 +369,7 @@ class Writer:
             prompt = get_prompt("title_optimize_wechat",
                 title=original,
                 core=self.topic.get('core_concept', ''),
+                field=self.topic.get('field', ''),
             )
         elif platform == "xiaohongshu":
             prompt = get_prompt("title_optimize_xhs",
@@ -459,10 +459,10 @@ class Writer:
         logger.info(f"选题 {self.topic_id} 状态已更新为待审查（数据库）")
 
     def run(self):
-        logger.info("开始撰写阶段")
+        logger.info("开始撰写阶段（三平台独立展开）")
-        markdown = self.generate_full_markdown()
         results = {}
         for platform in ["zhihu", "wechat", "xiaohongshu"]:
+            markdown = self.generate_platform_markdown(platform)
             html = self.generate_platform_html(markdown, platform)
             results[platform] = str(self.save_html(html, platform))
         self.mark_draft()