feihong-code 8.0.0 → 8.0.2

package/agent-team/.env.example

@@ -0,0 +1,13 @@
+# 复制为 .env 并填写
+# 模型供应商：mock（无 Key 跑通链路）| doubao（豆包大模型）
+LLM_PROVIDER=mock
+
+# 豆包大模型（火山方舟 OpenAI 兼容接口）
+DOUBAO_API_KEY=
+DOUBAO_MODEL=doubao-seed-1-6-250615
+DOUBAO_BASE_URL=https://ark.cn-beijing.volces.com/api/v3
+
+# 数据与调度
+DB_PATH=feihongzhi.db
+MAX_RETRY=3
+ALERT_LEVEL=log

package/agent-team/README.md

@@ -0,0 +1,42 @@
+# 飞虹智 · AI 员工军团系统（feihongzhi_agents）
+
+17 个 AI 员工模块 + 11 调度官的一人公司运营系统。按开发文档《AI 员工军团系统开发文档》实现。
+
+## 快速开始
+
+```bash
+# 1. 初始化数据库
+python3 main.py init-db
+
+# 2. 查看 17 个 AI 员工清单
+python3 main.py list-agents
+
+# 3. 跑通 M1 销售闭环演示（01线索雷达 → 02跟进销冠 → 04成交分析 → 09私域分层）
+python3 main.py demo-pipeline
+
+# 4. 提交单个任务并执行
+python3 main.py submit 12 --input '{"topic": "工厂老板如何用AI获客"}'
+python3 main.py run-once
+
+# 5. 任务统计
+python3 main.py stats
+```
+
+## 接入真实大模型
+
+复制 `.env.example` 为 `.env`，填入豆包 API Key 后设置 `LLM_PROVIDER=doubao`。
+
+## 目录结构
+
+```
+core/       数据模型(SQLite)、任务协议与状态机、11 调度官、BaseAgent 基类
+llm/        LLM 适配器（mock 本地模拟 / 豆包 API）
+agents/     17 个 AI 员工模块（按五层组织）
+tests/      单元测试
+```
+
+## 开发状态
+
+- [x] M0 基建：任务协议、数据模型、调度官骨架、模块框架（当前版本）
+- [ ] M1 销售闭环：接入真实 LLM 与真实线索源
+- [ ] M2 内容引擎 / M3 渠道扩展 / M4 战略决策（按路线图推进）

package/agent-team/agents/l1_acquisition.py

@@ -0,0 +1,207 @@
+"""L1 获客引流层：01 获客线索雷达 / 03 评论区商机识别官 / 06 每日精准获客选题官 / 07 AI外贸开发客户员工 / 08 小红书精准获客员工。
+
+每个模块的核心业务逻辑为模块级纯函数 run_logic_XX(data, llm)：
+- 确定性逻辑只依赖标准库，可被 skill_packager 内嵌导出为独立 Skill 的 run.py；
+- llm 非 mock 时追加 AI 增强说明字段 ai_note。
+"""
+import json
+import time
+from datetime import datetime
+
+from core.agent import BaseAgent
+
+
+def _ts() -> str:
+    return datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+
+
+# ---------------------------------------------------------------- 01
+def run_logic_01(data, llm=None):
+    """全域线索发现、清洗与入库（线索池）。"""
+    channels = data.get("channels") or ["评论区", "小红书", "公众号"]
+    keywords = data.get("keywords", "AI 获客")
+    now = int(time.time())
+    leads = []
+    for i, ch in enumerate(channels[:3], 1):
+        leads.append({
+            "lead_id": f"L{now % 100000000}{i}",
+            "source_channel": ch,
+            "company_name": f"示例{['制造','贸易','服务'][i-1]}公司{i}",
+            "contact_name": f"联系人{i}",
+            "contact_phone": f"1380000{i:04d}",
+            "grade": ["A", "B", "B"][i - 1],
+            "remark": f"来自{ch}，匹配关键词「{keywords}」",
+            "created_at": _ts(),
+        })
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"请基于输入 {json.dumps(data, ensure_ascii=False)} 给出线索质量评估与分级建议，不超过 80 字。",
+                system="你是资深销售线索运营专家。")
+        except Exception:
+            ai_note = ""
+    return {"total": len(leads), "leads": leads, "dedup_rule": "phone+company 联合去重", "ai_note": ai_note}
+
+
+class LeadRadarAgent(BaseAgent):
+    module_id = "01"
+    name = "AI获客线索雷达"
+    layer = "L1"
+    priority = "P0"
+    description = "全域线索发现、清洗与入库，构建线索池"
+    input_schema = {"channels": "线索渠道列表，如 [\"评论区\",\"小红书\",\"公众号\"]", "keywords": "获客关键词"}
+    output_keys = {"total": "线索总数", "leads": "线索列表", "dedup_rule": "去重规则"}
+    run_logic = staticmethod(run_logic_01)
+
+    def persist(self, output, task):
+        from core import models
+        for lead in output.get("leads", []):
+            models.insert_lead(lead)
+
+
+# ---------------------------------------------------------------- 03
+def run_logic_03(data, llm=None):
+    """评论区商机识别：监控目标账号评论区，提取意向商机。"""
+    accounts = data.get("target_accounts") or ["对标账号A", "对标账号B"]
+    found = []
+    for i, acc in enumerate(accounts[:2], 1):
+        found.append({
+            "opportunity_id": f"OP{i}{int(time.time()) % 100000}",
+            "account": acc,
+            "intent": "求购 / 咨询报价" if i == 1 else "寻求合作",
+            "quote": f"「请问你们这个方案怎么收费？想了解一下」",
+            "lead_ref": f"已转线索池待跟进（实时 {_ts()}）",
+        })
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"评论区内容 {json.dumps(found, ensure_ascii=False)}，请判断真实意向并排序，不超过 80 字。",
+                system="你是商机识别专家。")
+        except Exception:
+            ai_note = ""
+    return {"opportunities": found, "source_evidence": "原文链接与截图归档", "ai_note": ai_note}
+
+
+class CommentOpportunityAgent(BaseAgent):
+    module_id = "03"
+    name = "评论区商机识别官"
+    layer = "L1"
+    priority = "P1"
+    description = "监控目标账号与话题评论区，识别需求与意向，提取商机"
+    input_schema = {"target_accounts": "目标账号/话题列表"}
+    output_keys = {"opportunities": "商机列表", "source_evidence": "原文证据"}
+    run_logic = staticmethod(run_logic_03)
+
+    def persist(self, output, task):
+        from core import models
+        for op in output.get("opportunities", []):
+            models.insert_lead({
+                "lead_id": op.get("opportunity_id", ""),
+                "source_channel": "评论区",
+                "company_name": f"{op.get('account', '')}商机",
+                "contact_name": "待补",
+                "contact_phone": "",
+                "status": "new",
+                "grade": "B",
+                "remark": f"意向：{op.get('intent', '')}；原文：{op.get('quote', '')}",
+            })
+
+
+# ---------------------------------------------------------------- 06
+def run_logic_06(data, llm=None):
+    """每日精准获客选题：按目标客户画像与热点生成选题。"""
+    persona = data.get("persona", "工厂老板")
+    hot_topics = data.get("hot_topics") or ["AI 获客", "降本增效"]
+    topics = [
+        {"title": f"AI 如何帮{persona}每天自动获客", "direction": "痛点切入", "reason": f"结合热点「{hot_topics[0]}」"},
+        {"title": "一人公司怎么养 17 个 AI 员工", "direction": "案例教学", "reason": "差异化稀缺选题"},
+        {"title": "评论区才是最大的金矿", "direction": "方法拆解", "reason": f"呼应热点「{hot_topics[-1]}」"},
+    ]
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"画像：{persona}；热点：{hot_topics}。补充 1 个更高转化率的选题，不超过 60 字。",
+                system="你是短视频选题策划专家。")
+        except Exception:
+            ai_note = ""
+    return {"date": _ts()[:10], "topics": topics, "ai_note": ai_note}
+
+
+class DailyTopicAgent(BaseAgent):
+    module_id = "06"
+    name = "每日精准获客选题官"
+    layer = "L1"
+    priority = "P1"
+    description = "按目标客户画像与行业热点生成每日获客选题"
+    input_schema = {"persona": "目标客户画像", "hot_topics": "行业热点列表"}
+    output_keys = {"date": "选题日期", "topics": "选题清单"}
+    run_logic = staticmethod(run_logic_06)
+
+
+# ---------------------------------------------------------------- 07
+def run_logic_07(data, llm=None):
+    """外贸客户开发：目标市场客户开发与首轮触达。"""
+    market = data.get("market", "东南亚")
+    product = data.get("product", "智能家居")
+    letter = (
+        f"Subject: {product} 供应商直供方案（{market}）\n\n"
+        f"您好，我们专注 {product} 产能与交期，可支持 OEM/ODM，"
+        f"首单可样品验证。期待交流。"
+    )
+    candidates = [
+        {"company": f"{market}采购商A", "channel": "展会名录", "priority": "高"},
+        {"company": f"{market}渠道商B", "channel": "平台询盘", "priority": "中"},
+    ]
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"市场 {market}、产品 {product}，优化开发信开头两句话，不超过 60 字。",
+                system="你是外贸获客专家。")
+        except Exception:
+            ai_note = ""
+    return {"outreach_letter": letter, "candidates": candidates, "compliance": "触达前已过频控与合规检查", "ai_note": ai_note}
+
+
+class ForeignTradeAgent(BaseAgent):
+    module_id = "07"
+    name = "AI外贸开发客户员工"
+    layer = "L1"
+    priority = "P2"
+    description = "外贸场景客户开发与首轮触达"
+    input_schema = {"market": "目标市场", "product": "产品"}
+    output_keys = {"outreach_letter": "开发信", "candidates": "候选客户"}
+    run_logic = staticmethod(run_logic_07)
+
+
+# ---------------------------------------------------------------- 08
+def run_logic_08(data, llm=None):
+    """小红书获客：内容获客与线索收集。"""
+    category = data.get("category", "企业服务")
+    audience = data.get("audience", "工厂老板")
+    note_title = f"做了 3 个月小红书，{audience}客户主动找上门"
+    note_body = f"从 0 到 1 用小红书给{category}获客的真实过程：选题、封面、评论区运营三步走。"
+    guide_comment = "需要这套《小红书获客 SOP》的朋友，评论区扣「1」，我发你。"
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"品类 {category}、受众 {audience}，给出 1 个更高点击的标题，不超过 40 字。",
+                system="你是小红书内容运营专家。")
+        except Exception:
+            ai_note = ""
+    return {"note_title": note_title, "note_body": note_body, "guide_comment": guide_comment, "ai_note": ai_note}
+
+
+class XiaohongshuAgent(BaseAgent):
+    module_id = "08"
+    name = "小红书精准获客员工"
+    layer = "L1"
+    priority = "P2"
+    description = "小红书内容获客与线索收集"
+    input_schema = {"category": "品类", "audience": "目标人群"}
+    output_keys = {"note_title": "笔记标题", "note_body": "笔记正文", "guide_comment": "评论区引导话术"}
+    run_logic = staticmethod(run_logic_08)

package/agent-team/agents/l2_content.py

@@ -0,0 +1,237 @@
+"""L2 内容生产层：05 同行爆款雷达 / 12 老板短视频获客脚本官 / 13 博主账号深度拆解官 / 14 AI短视频自动剪辑师 / 15 公众号自动运营发布官。"""
+import json
+import time
+from datetime import datetime
+
+from core.agent import BaseAgent
+
+
+def _ts() -> str:
+    return datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+
+
+# ---------------------------------------------------------------- 05
+def run_logic_05(data, llm=None):
+    """同行爆款雷达：监控同行爆款，提炼可复制模式。"""
+    accounts = data.get("competitor_accounts") or ["同行账号A", "同行账号B"]
+    days = int(data.get("days", 7))
+    hot = [
+        {"title": "工厂老板必看的 3 个获客真相", "views": 128000, "account": accounts[0], "days": days},
+        {"title": "我靠评论区捡了 200 个客户", "views": 96000, "account": accounts[-1], "days": days},
+    ]
+    patterns = ["强反差标题 + 数字钩子", "真实案例收尾 + 行动号召", "评论区互动引导"]
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"近 {days} 天爆款 {json.dumps(hot, ensure_ascii=False)}，提炼 1 条可复用选题公式，不超过 60 字。",
+                system="你是内容增长分析师。")
+        except Exception:
+            ai_note = ""
+    return {"hot_contents": hot, "patterns": patterns, "ai_note": ai_note}
+
+
+class ViralRadarAgent(BaseAgent):
+    module_id = "05"
+    name = "同行爆款雷达"
+    layer = "L2"
+    priority = "P2"
+    description = "监控同行爆款内容，提炼可复制模式"
+    input_schema = {"competitor_accounts": "同行账号列表"}
+    output_keys = {"hot_contents": "爆款列表", "patterns": "可复用模式"}
+    run_logic = staticmethod(run_logic_05)
+
+    def persist(self, output, task):
+        from core import models
+        models.insert_asset({
+            "content_id": f"A{int(time.time()) % 100000000}",
+            "content_type": "爆款分析",
+            "topic": "同行爆款雷达",
+            "script": json.dumps(output.get("hot_contents", []), ensure_ascii=False),
+            "publish_status": "draft",
+        })
+
+
+# ---------------------------------------------------------------- 12
+def run_logic_12(data, llm=None):
+    """短视频获客脚本：生成口播 + 分镜脚本。"""
+    topic = data.get("topic", "AI 获客")
+    points = list(data.get("selling_points") or ["自动找客户", "降本增效"])
+    while len(points) < 2:
+        points.append("真实案例支撑")
+    audience = data.get("audience", "工厂老板")
+    script = {
+        "hook": f"{audience}注意：{topic}，90% 的人都做反了。",
+        "body": f"三步讲清：①{points[0]}；②{points[1]}；③真实案例。",
+        "cta": "评论区扣「方案」，免费领《获客清单》。",
+        "duration_sec": 45,
+        "shots": ["近景口播 3s", "痛点场景 10s", "操作演示 20s", "案例收尾 10s", "CTA 2s"],
+    }
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"选题 {topic}、卖点 {points}，重写黄金 3 秒钩子，不超过 40 字。",
+                system="你是短视频编导。")
+        except Exception:
+            ai_note = ""
+    return {"script": script, "ai_note": ai_note}
+
+
+class VideoScriptAgent(BaseAgent):
+    module_id = "12"
+    name = "老板短视频获客脚本官"
+    layer = "L2"
+    priority = "P1"
+    description = "生成获客向短视频脚本（口播 + 分镜）"
+    input_schema = {"topic": "选题", "selling_points": "卖点列表", "audience": "目标客户"}
+    output_keys = {"script": "脚本（hook/body/cta/shots）"}
+    run_logic = staticmethod(run_logic_12)
+
+    def persist(self, output, task):
+        from core import models
+        models.insert_asset({
+            "content_id": f"A{int(time.time()) % 100000000}",
+            "content_type": "脚本",
+            "topic": task.input_json.get("topic", ""),
+            "script": json.dumps(output.get("script", {}), ensure_ascii=False),
+            "publish_status": "draft",
+        })
+
+
+# ---------------------------------------------------------------- 13
+def run_logic_13(data, llm=None):
+    """博主账号拆解：拆解对标账号的定位、选题、结构、转化路径。"""
+    account = data.get("account", "对标账号A")
+    report = {
+        "account": account,
+        "positioning": "工厂老板 IP，讲透 AI 获客",
+        "topics": ["获客方法", "行业案例", "工具测评"],
+        "structure": "三段式：痛点 → 方法 → 案例",
+        "conversion": "评论区引流 → 私域 → 成交",
+        "takedown_at": _ts(),
+    }
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"拆解账号 {account}，给出 1 条可立即复用的打法，不超过 60 字。",
+                system="你是账号增长拆解专家。")
+        except Exception:
+            ai_note = ""
+    return {"report": report, "ai_note": ai_note}
+
+
+class BloggerTeardownAgent(BaseAgent):
+    module_id = "13"
+    name = "博主账号深度拆解官"
+    layer = "L2"
+    priority = "P2"
+    description = "拆解对标账号的定位、选题、结构与转化路径"
+    input_schema = {"account": "对标账号"}
+    output_keys = {"report": "拆解报告"}
+    run_logic = staticmethod(run_logic_13)
+
+    def persist(self, output, task):
+        from core import models
+        models.insert_asset({
+            "content_id": f"A{int(time.time()) % 100000000}",
+            "content_type": "账号拆解",
+            "topic": task.input_json.get("account", ""),
+            "script": json.dumps(output.get("report", {}), ensure_ascii=False),
+            "publish_status": "draft",
+        })
+
+
+# ---------------------------------------------------------------- 14
+def run_logic_14(data, llm=None):
+    """AI 短视频自动剪辑：素材自动剪辑成片。"""
+    clips = data.get("clips") or ["clip_a.mp4", "clip_b.mp4", "clip_c.mp4"]
+    script = data.get("script", "")
+    template = data.get("template", "口播标准模板")
+    video = {
+        "asset_id": f"V{int(time.time()) % 100000000}",
+        "clips_used": clips,
+        "template": template,
+        "duration_sec": 45,
+        "subtitles": "auto（准确率≥95%）",
+        "status": "rendered",
+        "rendered_at": _ts(),
+    }
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"脚本要点：{script[:100]}，给出剪辑节奏建议，不超过 50 字。",
+                system="你是视频剪辑师。")
+        except Exception:
+            ai_note = ""
+    return {"video": video, "ai_note": ai_note}
+
+
+class AutoEditorAgent(BaseAgent):
+    module_id = "14"
+    name = "AI短视频自动剪辑师"
+    layer = "L2"
+    priority = "P1"
+    description = "素材自动剪辑成片、批量出片"
+    input_schema = {"clips": "素材列表", "script": "脚本", "template": "模板"}
+    output_keys = {"video": "成片元数据"}
+    run_logic = staticmethod(run_logic_14)
+
+    def persist(self, output, task):
+        from core import models
+        video = output.get("video", {})
+        models.insert_asset({
+            "content_id": video.get("asset_id", f"A{int(time.time()) % 100000000}"),
+            "content_type": "成片",
+            "topic": task.input_json.get("script", "")[:60],
+            "script": json.dumps(video, ensure_ascii=False),
+            "asset_url": video.get("asset_id", ""),
+            "publish_status": video.get("status", "draft"),
+        })
+
+
+# ---------------------------------------------------------------- 15
+def run_logic_15(data, llm=None):
+    """公众号自动运营：内容生成与定时发布。"""
+    topic = data.get("topic", "AI 获客")
+    schedule = data.get("schedule", "明日 08:00")
+    article = {
+        "title": f"深度：{topic}的完整打法（建议收藏）",
+        "body": f"一、为什么{topic}是当下最大的红利；二、三步落地方案；三、真实数据复盘。",
+        "publish_time": schedule,
+        "status": "scheduled",
+        "created_at": _ts(),
+    }
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"公众号选题 {topic}，给 1 个标题备选，不超过 40 字。",
+                system="你是公众号主编。")
+        except Exception:
+            ai_note = ""
+    return {"article": article, "ai_note": ai_note}
+
+
+class GzhPublisherAgent(BaseAgent):
+    module_id = "15"
+    name = "公众号自动运营发布官"
+    layer = "L2"
+    priority = "P1"
+    description = "公众号内容生成、排期与定时发布"
+    input_schema = {"topic": "选题", "schedule": "发布时间"}
+    output_keys = {"article": "文章与排期"}
+    run_logic = staticmethod(run_logic_15)
+
+    def persist(self, output, task):
+        from core import models
+        article = output.get("article", {})
+        models.insert_asset({
+            "content_id": f"A{int(time.time()) % 100000000}",
+            "content_type": "文章",
+            "topic": article.get("title", ""),
+            "script": article.get("body", ""),
+            "publish_status": article.get("status", "draft"),
+        })

package/agent-team/agents/l3_conversion.py

@@ -0,0 +1,148 @@
+"""L3 销售转化层：02 AI销售跟进销冠 / 04 客户聊天成交分析官 / 09 私域客户分层跟进官。"""
+import json
+import time
+from datetime import datetime
+
+from core.agent import BaseAgent
+
+
+def _ts() -> str:
+    return datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+
+
+# ---------------------------------------------------------------- 02
+def run_logic_02(data, llm=None):
+    """销售跟进：按话术自动跟进、多轮触达客户。"""
+    customer = data.get("customer") or {"company_name": "示例公司", "contact_name": "王总"}
+    strategy = data.get("strategy", "标准 3 触达")
+    follow_up = {
+        "target": f"{customer.get('contact_name', '')}（{customer.get('company_name', '')}）",
+        "round": 1,
+        "message": f"您好，上次聊到的方案资料已整理好，方便发您参考吗？",
+        "channel": "企业微信",
+        "frequency_check": "24h 内仅 1 次，通过频控",
+        "next_action": "若 24h 未回复，执行第二轮价值提醒",
+        "need_human": True,
+        "at": _ts(),
+    }
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"客户 {json.dumps(customer, ensure_ascii=False)}，策略 {strategy}，生成第二轮跟进话术，不超过 60 字。",
+                system="你是顶级销售。")
+        except Exception:
+            ai_note = ""
+    return {"follow_up": follow_up, "ai_note": ai_note}
+
+
+class FollowUpAgent(BaseAgent):
+    module_id = "02"
+    name = "AI销售跟进销冠"
+    layer = "L3"
+    priority = "P0"
+    description = "按话术自动跟进、多轮触达客户（对外动作需人工确认）"
+    input_schema = {"customer": "客户对象（company_name/contact_name）", "strategy": "跟进策略"}
+    output_keys = {"follow_up": "跟进消息与下一步"}
+    need_human = True
+    run_logic = staticmethod(run_logic_02)
+
+    def persist(self, output, task):
+        from core import models
+        c = task.input_json.get("customer", {})
+        fu = output.get("follow_up", {})
+        models.upsert_customer(
+            company_name=c.get("company_name", ""),
+            contact_name=c.get("contact_name", ""),
+            contact_phone=c.get("contact_phone", ""),
+            remark=fu.get("message", ""),
+            tier="untiered",
+        )
+
+
+# ---------------------------------------------------------------- 04
+def run_logic_04(data, llm=None):
+    """聊天成交分析：分析聊天记录，诊断卡点，输出建议话术。"""
+    transcript = data.get("transcript", "客户：价格有点高；销售：可以谈；客户：我再想想。")
+    diagnosis = {
+        "pain_points": ["价格敏感"],
+        "blockers": ["价值量化不足", "未给决策紧迫感"],
+        "suggested_scripts": [
+            "先量化 ROI：按当前产能算 3 个月回本",
+            "再给限时政策：本月底前签约送首月服务",
+        ],
+        "confidence": "中高（抽样人工复核）",
+        "analyzed_at": _ts(),
+    }
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"聊天记录：{transcript[:200]}，指出最关键的一个成交卡点，不超过 50 字。",
+                system="你是成交分析专家。")
+        except Exception:
+            ai_note = ""
+    return {"diagnosis": diagnosis, "ai_note": ai_note}
+
+
+class ChatAnalysisAgent(BaseAgent):
+    module_id = "04"
+    name = "客户聊天成交分析官"
+    layer = "L3"
+    priority = "P0"
+    description = "分析聊天记录，诊断成交卡点，输出建议话术"
+    input_schema = {"transcript": "聊天记录 / 通话转写"}
+    output_keys = {"diagnosis": "卡点诊断与建议话术"}
+    run_logic = staticmethod(run_logic_04)
+
+
+# ---------------------------------------------------------------- 09
+def run_logic_09(data, llm=None):
+    """私域分层跟进：客户分层、差异化跟进计划。"""
+    customers = data.get("customers") or [
+        {"customer_id": "C1", "company_name": "客户A", "contact_name": "李总", "contact_phone": "13800000001"},
+        {"customer_id": "C2", "company_name": "客户B", "contact_name": "张总", "contact_phone": "13800000002"},
+        {"customer_id": "C3", "company_name": "客户C", "contact_name": "陈总", "contact_phone": "13800000003"},
+    ]
+    tiers = []
+    for i, c in enumerate(customers):
+        tier = ["A", "B", "C"][i % 3]
+        tiers.append({"customer_id": c["customer_id"], "company_name": c["company_name"], "tier": tier})
+    plan = [
+        {"tier": "A", "action": "每周 1 次价值触达 + 需求深挖", "interval": "7 天"},
+        {"tier": "B", "action": "每月 1 次方案更新 + 节点关怀", "interval": "30 天"},
+        {"tier": "C", "action": "季度盘点，重新评估意向", "interval": "90 天"},
+    ]
+    ai_note = ""
+    if llm is not None and getattr(llm, "provider", "mock") != "mock":
+        try:
+            ai_note = llm.complete(
+                f"客户列表 {json.dumps(customers, ensure_ascii=False)}，给出分层建议说明，不超过 60 字。",
+                system="你是私域运营专家。")
+        except Exception:
+            ai_note = ""
+    return {"tiers": tiers, "follow_up_plan": plan, "ai_note": ai_note}
+
+
+class TieringAgent(BaseAgent):
+    module_id = "09"
+    name = "私域客户分层跟进官"
+    layer = "L3"
+    priority = "P0"
+    description = "私域客户分层、差异化跟进与维护"
+    input_schema = {"customers": "客户列表"}
+    output_keys = {"tiers": "分层标签", "follow_up_plan": "跟进计划"}
+    run_logic = staticmethod(run_logic_09)
+
+    def persist(self, output, task):
+        from core import models
+        tier_map = {t["customer_id"]: t["tier"] for t in output.get("tiers", [])}
+        for c in task.input_json.get("customers", []):
+            models.upsert_customer(
+                customer_id=c.get("customer_id"),
+                company_name=c.get("company_name", ""),
+                contact_name=c.get("contact_name", ""),
+                contact_phone=c.get("contact_phone", ""),
+                tier=tier_map.get(c.get("customer_id"), "untiered"),
+                remark="分层跟进",
+            )