feat: 增强规则资产管理与审计页面运行时调试

后端新增规则资产版本管理和规则文件 CRUD 接口，优化风险规则生成模板执行和员工数据模型字段，知识库 RAG 增强本地回退和文档提取能力，清理旧风险规则文件统一由生成引擎管理，前端审计页面增加运行时调试面板和规则资产编辑交互，补充单元测试覆盖。
2026-05-24 21:44:17 +08:00
parent 575f093c74
commit 50b1c3f9a9
113 changed files with 13896 additions and 5044 deletions
--- a/.tmp/Yuxi
+++ b/.tmp/Yuxi
--- a/document/development/hermes_agent/01_architecture_overview.md
+++ b/document/development/hermes_agent/01_architecture_overview.md
@@ -0,0 +1,41 @@
+# Hermes 后台智能体架构总览
+
+## 1. 定位与愿景
+Hermes 是 X-Financial 系统中的**后台自动巡检与数据洞察中枢**。与处理实时对话的 UserAgent 不同，Hermes 专注于异步、长周期、大批量的任务，核心价值在于提供事前的**深度风险挖掘**和定期的**业财洞察报告**。
+
+## 2. 系统拓扑图
+```mermaid
+graph TD
+    subgraph 调度层
+        A[Cron Scheduler] -->|定时触发| B[Task Queue]
+    end
+    
+    subgraph Hermes Agent 层
+        B -->|消费任务| C(Hermes Worker)
+        C --> D[Task Skills Router]
+        D --> E{RiskScanner Skill}
+        D --> F{ExpenseReport Skill}
+        D --> G{KnowledgeCheck Skill}
+    end
+    
+    subgraph X-Financial 核心服务
+        E <--> H[(Expense DB)]
+        F <--> H
+        G <--> I[(LightRAG Graph/Vector)]
+        C <--> J[LLM Gateway / OpenAI]
+        C --> K[Notification Service / 企业微信]
+    end
+```
+
+## 3. 核心设计原则
+1. **防抖与限流**：后台全量扫表时，必须分片执行，防止对主数据库造成 I/O 拥堵。
+2. **幂等性保障**：每一个扫描任务和报告生成任务都应该具备唯一幂等键，避免因进程重启导致的重复发信或重复扣减信用分。
+3. **隔离性**：Hermes 的进程应与对外提供 API 服务的 Web Server 物理/逻辑隔离，大模型限流策略（Token Rate Limit）应配置相互独立的账单通道。
+
+## 4. 核心执行链路（示例：夜间风控巡检）
+1. 凌晨 02:00，Cron 触发 `trigger_risk_scan` 任务。
+2. Worker 拉取状态为 `draft` 和 `submitted` 且 `risk_scanned=False` 的单据。
+3. 将近三个月的相关人员单据聚类，构建 Context。
+4. 调用大模型，寻找“拆单”、“合谋”、“时间/地点异常”等隐蔽风险。
+5. 将发现的风险写入 `hermes_risk_report` 表，并标记对应单据。
+6. 任务结束，更新执行日志，等待早晨财务主管查阅。
--- a/document/development/hermes_agent/02_database_design.md
+++ b/document/development/hermes_agent/02_database_design.md
@@ -0,0 +1,46 @@
+# Hermes 数据库表结构设计
+
+为了支持后台异步任务的执行和长期记忆（风险标记、执行结果归档），我们需要在数据库中增加（或扩充）以下表结构。
+
+## 1. 任务调度与执行表
+
+### `hermes_task_config` (定时任务配置表)
+用于管理所有的后台巡检和推送任务，支持动态调整频率与开关。
+- `id`: string (UUID)
+- `task_type`: string (enum: `global_risk_scan`, `weekly_expense_report`, `kb_validation`...)
+- `cron_expression`: string (e.g., `0 2 * * *`)
+- `is_enabled`: boolean (默认 True)
+- `payload_template`: jsonb (预留参数，如扫描的时间窗口、特定部门过滤条件等)
+- `updated_at`: datetime
+
+### `hermes_task_execution_log` (任务执行日志表)
+记录每次任务的执行状态，便于排错与溯源。
+- `id`: string (UUID)
+- `config_id`: string (外键，关联 `hermes_task_config`)
+- `started_at`: datetime
+- `completed_at`: datetime
+- `status`: string (enum: `running`, `success`, `failed`)
+- `result_summary`: string (执行结果的简要说明，如“扫描了 1500 条单据，发现 12 条高危”)
+- `error_trace`: text (如果失败，存储错误堆栈)
+
+## 2. 深度分析结果表
+
+### `hermes_risk_report` (深度风险报告表)
+用于存储 LLM 找出的深层逻辑风险。
+- `id`: string (UUID)
+- `claim_id`: string (外键，关联存疑的主单据 `expense_claim`)
+- `execution_log_id`: string (外键，由哪次扫描任务产生的)
+- `risk_level`: string (enum: `low`, `medium`, `high`, `critical`)
+- `risk_type`: string (enum: `split_billing` 拆单, `collusion` 合谋, `policy_violation` 违规...)
+- `risk_description`: text (大模型生成的自然语言报告，如“该单据与前天提交的单据存在拆分可能...”)
+- `related_claim_ids`: jsonb (存储关联的同谋/相关单据 ID 列表，提供上下文线索)
+- `status`: string (enum: `pending_review` 待人工复核, `confirmed` 已确认为风险, `dismissed` 已忽略)
+
+## 3. 现有表的平滑改造
+
+### 修改 `employee` 表 (员工信用分预留)
+- **新增字段** `compliance_score`: int (默认 100，由 Hermes 动态扣减或恢复，用于风控引擎调节对该员工的抽查率和宽容度)
+
+### 修改 `expense_claim` 表 (风控标记)
+- **新增字段** `hermes_scanned_at`: datetime (记录该单据上次被 Hermes 扫描的时间，防止重复扫描)
+- **新增字段** `hermes_risk_flag`: boolean (快速判断该单子是否被挂载了 `hermes_risk_report`)
--- a/document/development/hermes_agent/03_risk_scan_module.md
+++ b/document/development/hermes_agent/03_risk_scan_module.md
@@ -0,0 +1,32 @@
+# 深度风险扫描模块设计 (Risk Scan Module)
+
+## 1. 业务目标
+将单点硬规则风控（如：发票大于 500 元是否合规）升级为**图谱式全局风控**。Hermes 将利用大语言模型（LLM）的逻辑推理能力，在海量历史数据中寻找隐藏的违规模式。
+
+## 2. 核心扫描链路
+本模块将作为一个独立的 Skill 被定时任务触发。
+
+### 第一步：数据快照聚合
+- **提取目标**：拉取状态为 `draft`、`submitted` 且最近 30 天内活跃的报销单，同时带出相关的发票明细。
+- **降维处理**：为避免超出大模型的 Token 上下文限制，必须对单据信息进行降维。仅提取：`申请人、时间、地点、商户名、金额、报销类型` 形成精简的 CSV 或 JSON Lines 格式。
+
+### 第二步：大模型批量推理 (LLM Batch Inference)
+- **风险定义植入**：通过 System Prompt 将目前财务最头疼的几类风险定义给模型（如拆单、套现、虚假连号发票）。
+- **执行方式**：将数据按“同部门”或“同地域”分块 (Chunking) 喂给大模型。
+- **Prompt 示例**：
+  ```markdown
+  你是一个内控审计 Agent。以下是某部门近半个月的报销流水清单。
+  请找出其中是否存在：
+  1. 拆单行为（同人、同地点、连日、小额累加）
+  2. 聚众套现行为（不同人、同偏僻餐馆、同日极高额）
+  如果发现风险，请输出对应的单号集合以及你的推理过程。
+  ```
+
+### 第三步：风险标记与处置
+- 解析大模型返回的结构化 JSON。
+- 对被判定的高危单据，在主库中插入 `hermes_risk_report` 记录。
+- **动作反馈**：如果该单据正处于 `submitted` 状态，并且得分极高（如虚假连号发票），可以通过 X-Financial 原有接口自动注入“退回”动作，并附加大模型的分析日志。
+
+## 3. 防抖与自我迭代
+- **扫描去重**：利用 `expense_claim.hermes_scanned_at` 防止已经出具过报告的单据被重复投入分析队列。
+- **人工纠偏 (Human-in-the-loop)**：当财务在前端驳回 Hermes 的风险提示（即认为没问题）时，事件将被记录。Hermes 可通过夜间的反思任务优化下一次 Prompt 中的判定容忍度。
--- a/document/development/hermes_agent/04_expense_report_module.md
+++ b/document/development/hermes_agent/04_expense_report_module.md
@@ -0,0 +1,34 @@
+# 动态费控与洞察报告模块设计 (Expense Report Module)
+
+## 1. 业务目标
+基于海量的流水账单，定期（周/月）由 Hermes Agent 为部门管理者或财务总监自动生成具有**业务洞察力**的归因分析报告，将冷冰冰的数字转化为具有指导意义的自然语言建议。
+
+## 2. 核心分析链路
+
+### 第一步：BI 数据聚合 (Data Aggregation)
+- 借助 ORM 或底层 Data Warehouse (如有)，Hermes 执行预置的聚合查询。
+- **采集核心指标**：
+  - 本期各部门总花费及环比/同比变动率。
+  - 各类目（如打车、机票、住宿、招待）的占比变化。
+  - Top 10 花费最多的商户（如特定几家酒店或订票平台）。
+  - 各类目超额/退回率最高的人员画像。
+
+### 第二步：大模型归因分析 (LLM Attribution Analysis)
+将硬性的聚合数据转化为结构化 Prompt，让 LLM 充当“财务分析师”。
+- **Prompt 示例**：
+  ```markdown
+  你是企业的财务总监助理，请阅读以下【本月报销聚合数据】。
+  请帮我撰写一份 300 字以内的执行摘要报告。
+  重点指出：
+  1. 哪个部门/哪类费用增长最快？原因可能是什么？
+  2. 我们的长尾开销集中在哪些商户？是否存在能够跟商户谈“协议价”的谈判空间？
+  ```
+
+### 第三步：多渠道报告分发 (Report Delivery)
+- **生成制品**：Hermes 利用代码解释器 (如有) 或 Markdown 引擎，将图表与文本融合成正式的 PDF 或长图。
+- **触达渠道**：
+  - **推送机制**：调用企微/钉钉机器人 API，直接向管理者的工作台推送“上周费控简报”。
+  - **交互追问**：管理者收到简报卡片后，可以在对话框里直接@Hermes 追问：“详细列一下研发部上周在北京住宿的那 5 万块钱是怎么花的”，Hermes 将调取缓存的报告上下文立即答复。
+
+## 3. 商业价值落地
+这项功能极大地解放了财务部的报表处理时间。通过提供前置的谈判线索（如发现某经济型酒店的高频住客其实都可以导流到协议酒店），可以给公司带来直接的差旅成本节约。
--- a/document/development/hermes_agent/05_deployment_and_cron.md
+++ b/document/development/hermes_agent/05_deployment_and_cron.md
@@ -0,0 +1,63 @@
+# 部署与任务调度架构方案 (Deployment & Cron)
+
+## 1. 业务诉求分析
+Hermes 作为纯后台的智能体，它的执行过程长达几分钟甚至几小时。它绝不能与提供给前台 HTTP 请求的主 Web 服务混合在同一个事件循环（Event Loop）或同步进程中，否则会导致 API 严重堵塞和超时崩溃。
+
+因此，Hermes 的部署需要进行**进程级解耦**。
+
+## 2. 选型对比与推荐方案
+
+### 方案 A：Celery + Redis (重型/工业级标准)
+- **优势**：业界最成熟的 Python 异步任务队列，支持极其复杂的 Cron 配置，原生支持任务重试、失败回调以及分布式扩展。
+- **劣势**：增加系统组件（必须额外部署 Redis / RabbitMQ 容器），运维成本相对较高。
+- **结论**：如果 X-Financial 后续要承载上千人的企业报销，这是**首选必经之路**。
+
+### 方案 B：APScheduler + Background Worker (中型/轻量级)
+- **优势**：直接在 Python 进程内运行，无需额外的消息队列组件。可以用一个单独的 Docker 容器运行 `python run_hermes_scheduler.py`。
+- **劣势**：多节点部署时难以控制并发（可能会多个节点同时执行同样的任务），需要引入基于数据库表或 Redis 的分布式锁。
+- **结论**：适合初期快速跑通 MVP 的方案。
+
+## 3. 推荐架构：基于 Redis 分布式锁的独立容器方案
+结合当前现状，建议采用 **方案B 叠加 分布式锁** 的轻量微服务架构。
+
+### 部署拓扑：
+```yaml
+# docker-compose.yml 示例切片
+services:
+  x-financial-api:
+    build: .
+    command: uvicorn main:app
+    ports:
+      - "8000:8000"
+
+  x-financial-hermes:
+    build: .
+    command: python scripts/start_hermes_daemon.py
+    # 这个容器不暴露外部端口，纯粹在后台运行定时任务和消费队列
+```
+
+### 执行伪代码 (`start_hermes_daemon.py`)
+```python
+from apscheduler.schedulers.blocking import BlockingScheduler
+from app.services.system_hermes import SystemHermesService
+
+scheduler = BlockingScheduler()
+hermes = SystemHermesService()
+
+# 每天凌晨 3 点执行深度风控扫表
+@scheduler.scheduled_job('cron', hour=3, minute=0)
+def job_risk_scan():
+    # 获取分布式锁防止集群脑裂重复执行
+    if acquire_redis_lock("hermes:lock:risk_scan"):
+        try:
+            hermes.run_query("执行全局风控扫描技能...", skills=["global_risk_scan"])
+        finally:
+            release_redis_lock("hermes:lock:risk_scan")
+
+if __name__ == "__main__":
+    scheduler.start()
+```
+
+## 4. 容灾与可观测性保障
+- **日志采集**：确保 `hermes.run_query` 及其后台生成的 stdout/stderr 日志能够写入 ELK 或文件系统中，方便第二天的运维排查。
+- **告警链路**：如果调度系统挂掉或者大模型连续多次返回失败状态码，必须通过 Webhook 飞书/钉钉及时告警系统管理员。
--- a/fix_init.py
+++ b/fix_init.py
@@ -0,0 +1,24 @@
+import re
+
+file_path = 'server/src/app/models/__init__.py'
+with open(file_path, 'r', encoding='utf-8') as f:
+    content = f.read()
+
+# Add imports
+imports_to_add = "from app.models.hermes_config import HermesTaskConfig, HermesTaskExecutionLog\nfrom app.models.hermes_report import HermesRiskReport\n"
+content = re.sub(
+    r'(from app\.models\.organization import OrganizationUnit)',
+    imports_to_add + r'\1',
+    content
+)
+
+# Add to __all__
+content = re.sub(
+    r'(\s*"OrganizationUnit",)',
+    r'\n    "HermesTaskConfig",\n    "HermesTaskExecutionLog",\n    "HermesRiskReport",\1',
+    content
+)
+
+with open(file_path, 'w', encoding='utf-8') as f:
+    f.write(content)
+print('Done.')
--- a/fix_knowledge.py
+++ b/fix_knowledge.py
@@ -0,0 +1,45 @@
+import re
+
+file_path = 'server/src/app/services/user_agent_knowledge.py'
+with open(file_path, 'r', encoding='utf-8') as f:
+    content = f.read()
+
+content = re.sub(
+    r'(def _build_fast_knowledge_answer\([\s\S]*?-> str \| None:\n)',
+    r'\1        return None\n',
+    content
+)
+
+content = re.sub(
+    r'heading_text = f" > \{heading\}" if heading else ""',
+    r'if "表格行级检索线索" in heading:\n                heading = heading.replace("表格行级检索线索", "").strip(" >")\n            heading_text = f"（{heading}）" if heading else ""\n            item_title = item.get("title") or title',
+    content
+)
+
+content = re.sub(
+    r'evidence_lines\.append\(f"- 《\{item\.get\(\'title\'\) or title\}》\{heading_text\}：\{summary\}\\n\{preview\}"\)',
+    r'evidence_lines.append(f"- **《{item_title}》** {heading_text}\\n  {summary}\\n{preview}")',
+    content
+)
+
+content = re.sub(
+    r'evidence_lines\.append\(f"- 《\{item\.get\(\'title\'\) or title\}》\{heading_text\}：\\n\{rendered\}"\)',
+    r'evidence_lines.append(f"- **《{item_title}》** {heading_text}\\n{rendered}")',
+    content
+)
+
+content = re.sub(
+    r'evidence_lines\.append\(f"- 《\{item\.get\(\'title\'\) or title\}》\{heading_text\}：\{rendered\}"\)',
+    r'evidence_lines.append(f"- **《{item_title}》** {heading_text}\\n  {rendered}")',
+    content
+)
+
+content = re.sub(
+    r'evidence_lines\.append\(f"- 《\{item_title\}》：\{excerpt\}"\)',
+    r'evidence_lines.append(f"- **《{item_title}》**：{excerpt}")',
+    content
+)
+
+with open(file_path, 'w', encoding='utf-8') as f:
+    f.write(content)
+print('Done.')
--- a/fix_response.py
+++ b/fix_response.py
@@ -0,0 +1,15 @@
+import re
+
+file_path = 'server/src/app/services/user_agent_response.py'
+with open(file_path, 'r', encoding='utf-8') as f:
+    content = f.read()
+
+content = re.sub(
+    r'\s*"fallback_answer": fallback_answer,',
+    '',
+    content
+)
+
+with open(file_path, 'w', encoding='utf-8') as f:
+    f.write(content)
+print('Done.')
--- a/server/rules/risk-rules/risk.expense.consecutive_transport_receipts.json
+++ b/server/rules/risk-rules/risk.expense.consecutive_transport_receipts.json
@@ -1,32 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.expense.consecutive_transport_receipts",
-  "name": "连号交通票据",
-  "enabled": true,
-  "risk_dimension": "consecutive_receipts",
-  "ontology_signal": "consecutive_transport_receipts",
-  "evaluator": "consecutive_transport_receipts",
-  "applies_to": {
-    "expense_types": ["transport", "travel"],
-    "min_attachments": 2
-  },
-  "inputs": {
-    "invoice_no": "attachment.invoice_no"
-  },
-  "params": {
-    "min_consecutive_count": 3
-  },
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "medium",
-      "action": "manual_review"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 三、车辆交通 / 连号票集中报销",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.expense.entertainment_missing_detail.json
+++ b/server/rules/risk-rules/risk.expense.entertainment_missing_detail.json
@@ -1,29 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.expense.entertainment_missing_detail",
-  "name": "招待费事由不完整",
-  "enabled": true,
-  "risk_dimension": "entertainment_detail",
-  "ontology_signal": "entertainment_missing_detail",
-  "evaluator": "entertainment_reason_missing",
-  "applies_to": {
-    "domains": ["meal"]
-  },
-  "inputs": {
-    "reason": "claim.reason_corpus"
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "medium",
-      "action": "warn"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 三、餐费招待 / 业务招待无事由对象",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.expense.generated_20260523010818.json
+++ b/server/rules/risk-rules/risk.expense.generated_20260523010818.json
@@ -1,90 +0,0 @@
-{
-  "schema_version": "2.0",
-  "rule_code": "risk.expense.generated_20260523010818",
-  "name": "住宿城市必须出现在本次差旅行程城市中风险规则",
-  "description": "当报销业务满足“住宿城市必须出现在本次差旅行程城市中，如果酒店发票城市与申报目的地或交通票行程城市都不一致，则判定为高风险，并要求补充差旅说明。”时，系统会按高风险进行提示，并要求经办人或审核人补充核对依据。",
-  "enabled": true,
-  "risk_dimension": "natural_language_rule",
-  "risk_category": "报销",
-  "ontology_signal": "natural_language_risk",
-  "evaluator": "template_rule",
-  "template_key": "field_compare_v1",
-  "applies_to": {
-    "domains": [
-      "expense"
-    ]
-  },
-  "inputs": {
-    "fields": [
-      {
-        "key": "claim.reason",
-        "label": "报销事由",
-        "type": "text",
-        "source": "claim"
-      },
-      {
-        "key": "claim.location",
-        "label": "申报地点",
-        "type": "text",
-        "source": "claim"
-      },
-      {
-        "key": "attachment.hotel_city",
-        "label": "住宿城市",
-        "type": "text",
-        "source": "attachment"
-      },
-      {
-        "key": "attachment.route_cities",
-        "label": "行程城市",
-        "type": "list",
-        "source": "attachment"
-      }
-    ]
-  },
-  "params": {
-    "template_key": "field_compare_v1",
-    "field_keys": [
-      "claim.reason",
-      "claim.location",
-      "attachment.hotel_city",
-      "attachment.route_cities"
-    ],
-    "condition_summary": "对比报销事由、申报地点、住宿城市之间是否一致或存在交集",
-    "natural_language": "住宿城市必须出现在本次差旅行程城市中，如果酒店发票城市与申报目的地或交通票行程城市都不一致，则判定为高风险，并要求补充差旅说明。",
-    "conditions": [
-      {
-        "left": "claim.reason",
-        "operator": "overlap",
-        "right": "claim.location"
-      }
-    ]
-  },
-  "outcomes": {
-    "pass": {
-      "severity": "none",
-      "action": "continue"
-    },
-    "fail": {
-      "severity": "high",
-      "action": "manual_review"
-    }
-  },
-  "metadata": {
-    "owner": "WangMin",
-    "stability": "generated_draft",
-    "source_ref": "自然语言风险规则",
-    "created_at": "2026-05-23T01:08:18.310751+00:00",
-    "created_by": "WangMin",
-    "natural_language": "住宿城市必须出现在本次差旅行程城市中，如果酒店发票城市与申报目的地或交通票行程城市都不一致，则判定为高风险，并要求补充差旅说明。",
-    "business_explanation": "当报销业务满足“住宿城市必须出现在本次差旅行程城市中，如果酒店发票城市与申报目的地或交通票行程城市都不一致，则判定为高风险，并要求补充差旅说明。”时，系统会按高风险进行提示，并要求经办人或审核人补充核对依据。",
-    "condition_summary": "对比报销事由、申报地点、住宿城市之间是否一致或存在交集",
-    "flow": {
-      "start": "报销单据提交",
-      "evidence": "读取报销事由、申报地点、住宿城市",
-      "decision": "对比报销事由、申报地点、住宿城市之间是否一致或存在交集",
-      "pass": "未命中风险，继续业务流转",
-      "fail": "命中高风险，提示复核"
-    }
-  }
-}
--- a/server/rules/risk-rules/risk.expense.generated_20260523010846.json
+++ b/server/rules/risk-rules/risk.expense.generated_20260523010846.json
@@ -1,90 +0,0 @@
-{
-  "schema_version": "2.0",
-  "rule_code": "risk.expense.generated_20260523010846",
-  "name": "酒店发票城市必须与申报目的地或交通票风险规则",
-  "description": "当报销业务满足“酒店发票城市必须与申报目的地或交通票行程城市一致，如果都不一致，则判定为高风险，并要求报销人补充异常行程说明。”时，系统会按高风险进行提示，并要求经办人或审核人补充核对依据。",
-  "enabled": true,
-  "risk_dimension": "natural_language_rule",
-  "risk_category": "报销",
-  "ontology_signal": "natural_language_risk",
-  "evaluator": "template_rule",
-  "template_key": "field_compare_v1",
-  "applies_to": {
-    "domains": [
-      "expense"
-    ]
-  },
-  "inputs": {
-    "fields": [
-      {
-        "key": "claim.reason",
-        "label": "报销事由",
-        "type": "text",
-        "source": "claim"
-      },
-      {
-        "key": "claim.location",
-        "label": "申报地点",
-        "type": "text",
-        "source": "claim"
-      },
-      {
-        "key": "claim.employee_name",
-        "label": "报销人",
-        "type": "text",
-        "source": "claim"
-      },
-      {
-        "key": "attachment.route_cities",
-        "label": "行程城市",
-        "type": "list",
-        "source": "attachment"
-      }
-    ]
-  },
-  "params": {
-    "template_key": "field_compare_v1",
-    "field_keys": [
-      "claim.reason",
-      "claim.location",
-      "claim.employee_name",
-      "attachment.route_cities"
-    ],
-    "condition_summary": "对比报销事由、申报地点、报销人之间是否一致或存在交集",
-    "natural_language": "酒店发票城市必须与申报目的地或交通票行程城市一致，如果都不一致，则判定为高风险，并要求报销人补充异常行程说明。",
-    "conditions": [
-      {
-        "left": "claim.reason",
-        "operator": "overlap",
-        "right": "claim.location"
-      }
-    ]
-  },
-  "outcomes": {
-    "pass": {
-      "severity": "none",
-      "action": "continue"
-    },
-    "fail": {
-      "severity": "high",
-      "action": "manual_review"
-    }
-  },
-  "metadata": {
-    "owner": "min.wang@xfinance.com",
-    "stability": "generated_draft",
-    "source_ref": "自然语言风险规则",
-    "created_at": "2026-05-23T01:08:46.286513+00:00",
-    "created_by": "min.wang@xfinance.com",
-    "natural_language": "酒店发票城市必须与申报目的地或交通票行程城市一致，如果都不一致，则判定为高风险，并要求报销人补充异常行程说明。",
-    "business_explanation": "当报销业务满足“酒店发票城市必须与申报目的地或交通票行程城市一致，如果都不一致，则判定为高风险，并要求报销人补充异常行程说明。”时，系统会按高风险进行提示，并要求经办人或审核人补充核对依据。",
-    "condition_summary": "对比报销事由、申报地点、报销人之间是否一致或存在交集",
-    "flow": {
-      "start": "报销单据提交",
-      "evidence": "读取报销事由、申报地点、报销人",
-      "decision": "对比报销事由、申报地点、报销人之间是否一致或存在交集",
-      "pass": "未命中风险，继续业务流转",
-      "fail": "命中高风险，提示复核"
-    }
-  }
-}
--- a/server/rules/risk-rules/risk.expense.generated_20260523011139.json
+++ b/server/rules/risk-rules/risk.expense.generated_20260523011139.json
@@ -1,90 +0,0 @@
-{
-  "schema_version": "2.0",
-  "rule_code": "risk.expense.generated_20260523011139",
-  "name": "酒店发票城市一致性校验",
-  "description": "校验酒店发票城市是否与申报目的地或行程城市一致，不一致时标记为高风险并要求补充说明",
-  "enabled": true,
-  "risk_dimension": "natural_language_rule",
-  "risk_category": "报销",
-  "ontology_signal": "natural_language_risk",
-  "evaluator": "template_rule",
-  "template_key": "field_compare_v1",
-  "applies_to": {
-    "domains": [
-      "expense"
-    ]
-  },
-  "inputs": {
-    "fields": [
-      {
-        "key": "attachment.route_cities",
-        "label": "行程城市",
-        "type": "list",
-        "source": "attachment"
-      },
-      {
-        "key": "claim.location",
-        "label": "申报地点",
-        "type": "text",
-        "source": "claim"
-      },
-      {
-        "key": "attachment.hotel_city",
-        "label": "住宿城市",
-        "type": "text",
-        "source": "attachment"
-      },
-      {
-        "key": "claim.reason",
-        "label": "报销事由",
-        "type": "text",
-        "source": "claim"
-      }
-    ]
-  },
-  "params": {
-    "template_key": "field_compare_v1",
-    "field_keys": [
-      "attachment.hotel_city",
-      "claim.location",
-      "attachment.route_cities",
-      "claim.reason"
-    ],
-    "condition_summary": "对比住宿城市、申报地点、行程城市之间是否一致或存在交集",
-    "natural_language": "酒店发票城市必须与申报目的地或交通票行程城市一致，如果都不一致，则判定为高风险，并要求补充异常行程说明。",
-    "conditions": [
-      {
-        "left": "attachment.hotel_city",
-        "operator": "overlap",
-        "right": "claim.location"
-      }
-    ]
-  },
-  "outcomes": {
-    "pass": {
-      "severity": "none",
-      "action": "continue"
-    },
-    "fail": {
-      "severity": "high",
-      "action": "manual_review"
-    }
-  },
-  "metadata": {
-    "owner": "WangMin",
-    "stability": "generated_draft",
-    "source_ref": "自然语言风险规则",
-    "created_at": "2026-05-23T01:11:39.165281+00:00",
-    "created_by": "WangMin",
-    "natural_language": "酒店发票城市必须与申报目的地或交通票行程城市一致，如果都不一致，则判定为高风险，并要求补充异常行程说明。",
-    "business_explanation": "校验酒店发票城市是否与申报目的地或行程城市一致，不一致时标记为高风险并要求补充说明",
-    "condition_summary": "对比住宿城市、申报地点、行程城市之间是否一致或存在交集",
-    "flow": {
-      "start": "提交酒店发票",
-      "evidence": "读取住宿城市、申报地点、行程城市",
-      "decision": "对比住宿城市、申报地点、行程城市之间是否一致或存在交集",
-      "pass": "继续流转",
-      "fail": "提示高风险：酒店发票城市与申报目的地及行程城市均不一致，需补充异常行程说明"
-    }
-  }
-}
--- a/server/rules/risk-rules/risk.expense.meal_localized_as_travel.json
+++ b/server/rules/risk-rules/risk.expense.meal_localized_as_travel.json
@@ -1,30 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.expense.meal_localized_as_travel",
-  "name": "同城餐饮混入差旅",
-  "enabled": true,
-  "risk_dimension": "meal_travel_mix",
-  "ontology_signal": "meal_as_travel",
-  "evaluator": "meal_as_travel_same_city",
-  "applies_to": {
-    "domains": ["travel"]
-  },
-  "inputs": {
-    "declared": "claim.location",
-    "meal_city": "attachment.cities"
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "medium",
-      "action": "warn"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 三、餐费招待 / 同城餐饮归集异地差旅",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.expense.reason_too_brief.json
+++ b/server/rules/risk-rules/risk.expense.reason_too_brief.json
@@ -1,29 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.expense.reason_too_brief",
-  "name": "报销事由过短",
-  "enabled": true,
-  "risk_dimension": "reason_quality",
-  "ontology_signal": "reason_too_brief",
-  "evaluator": "reason_too_brief",
-  "applies_to": {},
-  "inputs": {
-    "reason": "claim.reason_corpus"
-  },
-  "params": {
-    "min_reason_length": 6
-  },
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "medium",
-      "action": "warn"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 通用 / 事由不足以支撑真实性判断",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.invoice.claimant_buyer_name_match.json
+++ b/server/rules/risk-rules/risk.invoice.claimant_buyer_name_match.json
@@ -1,32 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.invoice.claimant_buyer_name_match",
-  "name": "报销人与发票抬头一致",
-  "enabled": true,
-  "risk_dimension": "identity_consistency",
-  "ontology_signal": "buyer_name_mismatch",
-  "evaluator": "identity_consistency",
-  "applies_to": {
-    "min_attachments": 1
-  },
-  "inputs": {
-    "claimant": "claim.employee_name",
-    "buyer": "attachment.buyer_name"
-  },
-  "params": {
-    "allow_keywords": ["代报", "集团", "公司", "有限公司"]
-  },
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "high",
-      "action": "manual_review"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 二、发票类 / 抬头错误",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.invoice.cross_year_invoice.json
+++ b/server/rules/risk-rules/risk.invoice.cross_year_invoice.json
@@ -1,30 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.invoice.cross_year_invoice",
-  "name": "跨年发票入账",
-  "enabled": true,
-  "risk_dimension": "cross_year_invoice",
-  "ontology_signal": "cross_year_invoice",
-  "evaluator": "cross_year_invoice",
-  "applies_to": {
-    "min_attachments": 1
-  },
-  "inputs": {
-    "invoice_date": "attachment.invoice_date",
-    "claim_date": ["claim.occurred_at", "item.item_date"]
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "medium",
-      "action": "warn"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 二、发票类 / 跨年发票",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.invoice.document_expense_mismatch.json
+++ b/server/rules/risk-rules/risk.invoice.document_expense_mismatch.json
@@ -1,30 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.invoice.document_expense_mismatch",
-  "name": "开票内容与报销场景不符",
-  "enabled": true,
-  "risk_dimension": "document_expense_mismatch",
-  "ontology_signal": "document_expense_mismatch",
-  "evaluator": "document_expense_mismatch",
-  "applies_to": {
-    "min_attachments": 1
-  },
-  "inputs": {
-    "document_type": "attachment.document_type",
-    "expense_type": ["claim.expense_type", "item.item_type"]
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "medium",
-      "action": "warn"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 二、发票类 / 开票内容与业务不符",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.invoice.duplicate_invoice.json
+++ b/server/rules/risk-rules/risk.invoice.duplicate_invoice.json
@@ -1,29 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.invoice.duplicate_invoice",
-  "name": "发票重复报销",
-  "enabled": true,
-  "risk_dimension": "duplicate_invoice",
-  "ontology_signal": "duplicate_invoice",
-  "evaluator": "duplicate_invoice",
-  "applies_to": {
-    "min_attachments": 1
-  },
-  "inputs": {
-    "invoice_no": "attachment.invoice_no"
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "high",
-      "action": "block"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 二、发票类 / 重复报销",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.invoice.vague_goods_description.json
+++ b/server/rules/risk-rules/risk.invoice.vague_goods_description.json
@@ -1,30 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.invoice.vague_goods_description",
-  "name": "发票品名过于笼统",
-  "enabled": true,
-  "risk_dimension": "vague_goods_description",
-  "ontology_signal": "vague_goods_description",
-  "evaluator": "vague_goods_description",
-  "applies_to": {
-    "expense_types": ["office", "other"],
-    "min_attachments": 1
-  },
-  "inputs": {
-    "ocr": "attachment.ocr_text"
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "medium",
-      "action": "warn"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 二、发票类 / 品名笼统",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.invoice.void_or_red_invoice.json
+++ b/server/rules/risk-rules/risk.invoice.void_or_red_invoice.json
@@ -1,30 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.invoice.void_or_red_invoice",
-  "name": "作废或红冲发票",
-  "enabled": true,
-  "risk_dimension": "void_or_red_invoice",
-  "ontology_signal": "void_or_red_invoice",
-  "evaluator": "invoice_void_or_red",
-  "applies_to": {
-    "min_attachments": 1
-  },
-  "inputs": {
-    "status": "attachment.invoice_status",
-    "ocr": "attachment.ocr_text"
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "high",
-      "action": "block"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 二、发票类 / 作废红冲发票",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.travel.base_location_overlap.json
+++ b/server/rules/risk-rules/risk.travel.base_location_overlap.json
@@ -1,30 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.travel.base_location_overlap",
-  "name": "常驻地重合出差风险",
-  "enabled": true,
-  "risk_dimension": "base_location_overlap",
-  "ontology_signal": "base_location_overlap",
-  "evaluator": "base_location_overlap",
-  "applies_to": {
-    "domains": ["travel"]
-  },
-  "inputs": {
-    "employee_base": "employee.location",
-    "declared": "claim.location"
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "high",
-      "action": "manual_review"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 一、出差类 / 两头在外",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.travel.destination_receipt_location.json
+++ b/server/rules/risk-rules/risk.travel.destination_receipt_location.json
@@ -1,29 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.travel.destination_receipt_location",
-  "name": "申报地点与票据地点一致",
-  "risk_dimension": "location_consistency",
-  "ontology_signal": "location_mismatch",
-  "evaluator": "location_consistency",
-  "inputs": {
-    "declared": "claim.location",
-    "evidence": ["attachment.cities", "item.item_location"]
-  },
-  "params": {
-    "match_mode": "city_fuzzy",
-    "missing_evidence": "warn"
-  },
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "high",
-      "action": "manual_review",
-      "message_template": "申报地点 {declared} 与票据识别地点 {evidence} 不一致"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "updated_at": "2026-05-18"
-  }
-}
--- a/server/rules/risk-rules/risk.travel.hotel_without_itinerary.json
+++ b/server/rules/risk-rules/risk.travel.hotel_without_itinerary.json
@@ -1,32 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.travel.hotel_without_itinerary",
-  "name": "住宿城市与行程不一致",
-  "enabled": true,
-  "risk_dimension": "hotel_itinerary",
-  "ontology_signal": "hotel_itinerary_mismatch",
-  "evaluator": "hotel_without_itinerary",
-  "applies_to": {
-    "domains": ["travel"],
-    "expense_types": ["hotel", "travel"]
-  },
-  "inputs": {
-    "declared": "claim.location",
-    "hotel": "attachment.hotel_city",
-    "itinerary": "attachment.route_cities"
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "high",
-      "action": "manual_review"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 三、住宿费 / 夜间异地住宿、酒店连续多天",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.travel.intracity_travel_claim.json
+++ b/server/rules/risk-rules/risk.travel.intracity_travel_claim.json
@@ -1,30 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.travel.intracity_travel_claim",
-  "name": "同城虚报差旅补贴",
-  "enabled": true,
-  "risk_dimension": "intracity_travel",
-  "ontology_signal": "intracity_travel",
-  "evaluator": "intracity_travel_claim",
-  "applies_to": {
-    "domains": ["travel"]
-  },
-  "inputs": {
-    "declared": "claim.location",
-    "evidence": ["attachment.route", "attachment.cities"]
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "high",
-      "action": "manual_review"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 一、出差类 / 同城虚报差旅",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/rules/risk-rules/risk.travel.multi_city_reason_required.json
+++ b/server/rules/risk-rules/risk.travel.multi_city_reason_required.json
@@ -1,30 +0,0 @@
-{
-  "schema_version": "1.0",
-  "rule_code": "risk.travel.multi_city_reason_required",
-  "name": "多城市行程需说明",
-  "enabled": true,
-  "risk_dimension": "multi_city_itinerary",
-  "ontology_signal": "multi_city_itinerary",
-  "evaluator": "multi_city_reason_required",
-  "applies_to": {
-    "domains": ["travel"]
-  },
-  "inputs": {
-    "reason": "claim.reason_corpus",
-    "cities": ["attachment.cities", "item.item_location"]
-  },
-  "params": {},
-  "outcomes": {
-    "pass": { "severity": "none", "action": "continue" },
-    "fail": {
-      "severity": "medium",
-      "action": "warn"
-    }
-  },
-  "metadata": {
-    "owner": "风控与审计部",
-    "stability": "platform_builtin",
-    "source_ref": "常用risk.txt / 一、出差类 / 绕道出行、行程不符",
-    "updated_at": "2026-05-19"
-  }
-}
--- a/server/scripts/start_hermes_daemon.py
+++ b/server/scripts/start_hermes_daemon.py
@@ -0,0 +1,68 @@
+import os
+import sys
+import time
+import logging
+
+# Ensure src is in the python path
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "../src")))
+
+from app.core.logging import setup_logging
+from app.db.session import get_session_factory
+from app.models.hermes_config import HermesTaskConfig
+from app.services.hermes_scheduler import hermes_scheduler
+
+logger = logging.getLogger("hermes_daemon")
+
+def init_default_config():
+    """Ensure there is at least one active global_risk_scan task in the database."""
+    session_factory = get_session_factory()
+    db = session_factory()
+    try:
+        # 初始化 global_risk_scan
+        existing_risk = db.query(HermesTaskConfig).filter_by(task_type="global_risk_scan").first()
+        if not existing_risk:
+            logger.info("No global_risk_scan config found. Initializing default config.")
+            db.add(HermesTaskConfig(
+                task_type="global_risk_scan",
+                cron_expression="0 2 * * *", 
+                is_enabled=True
+            ))
+            
+        # 初始化 weekly_expense_report
+        existing_report = db.query(HermesTaskConfig).filter_by(task_type="weekly_expense_report").first()
+        if not existing_report:
+            logger.info("No weekly_expense_report config found. Initializing default config.")
+            db.add(HermesTaskConfig(
+                task_type="weekly_expense_report",
+                cron_expression="0 9 * * 1", # 每周一早9点（在简化版中暂时代表周报频率）
+                is_enabled=True
+            ))
+            
+        db.commit()
+    except Exception as e:
+        logger.error(f"Failed to initialize default config: {e}")
+    finally:
+        db.close()
+
+
+def main():
+    setup_logging()
+    logger.info("Initializing Hermes Background Daemon...")
+    
+    # 注入默认配置
+    init_default_config()
+    
+    # 启动调度器
+    hermes_scheduler.start()
+    
+    logger.info("Hermes Daemon is running. Press Ctrl+C to stop.")
+    try:
+        while True:
+            time.sleep(1) # 主线程保持存活
+    except KeyboardInterrupt:
+        logger.info("Keyboard interrupt received. Shutting down...")
+        hermes_scheduler.shutdown()
+        logger.info("Shutdown complete.")
+
+if __name__ == "__main__":
+    main()
--- a/server/src/app/api/v1/endpoints/agent_assets.py
+++ b/server/src/app/api/v1/endpoints/agent_assets.py
@@ -23,7 +23,16 @@ from app.schemas.agent_asset import (
    AgentAssetRead,
    AgentAssetReviewCreate,
    AgentAssetReviewRead,
+    AgentAssetRiskRuleEnabledUpdate,
    AgentAssetRiskRuleGenerateRequest,
+    AgentAssetRiskRuleLatestTestSummary,
+    AgentAssetRiskRuleReportRequest,
+    AgentAssetRiskRuleReturnRequest,
+    AgentAssetRiskRuleSampleTestRequest,
+    AgentAssetRiskRuleScenarioTestRequest,
+    AgentAssetRiskRuleSimulationRead,
+    AgentAssetRiskRuleSimulationRequest,
+    AgentAssetRiskRuleTestRunRead,
    AgentAssetRuleJsonRead,
    AgentAssetRuleJsonWrite,
    AgentAssetSpreadsheetChangeRecordRead,
@@ -131,6 +140,116 @@ def get_agent_asset_rule_json(
        _handle_asset_error(exc)


+@router.get(
+    "/{asset_id}/risk-rule-tests/latest",
+    response_model=AgentAssetRiskRuleLatestTestSummary,
+    summary="读取风险规则最近测试摘要",
+    description="返回当前风险规则工作版本最近一次样例测试、场景试运行和测试报告。",
+)
+def get_agent_asset_risk_rule_latest_test(
+    asset_id: str,
+    _: CurrentUser,
+    db: DbSession,
+) -> AgentAssetRiskRuleLatestTestSummary:
+    try:
+        return AgentAssetService(db).get_latest_risk_rule_test_summary(asset_id)
+    except Exception as exc:
+        _handle_asset_error(exc)
+
+
+@router.post(
+    "/{asset_id}/risk-rule-tests/simulate",
+    response_model=AgentAssetRiskRuleSimulationRead,
+    summary="执行风险规则对话仿真",
+    description="基于临时对话输入和附件元信息执行风险识别，不创建业务单据，不写入测试记录。",
+)
+def simulate_agent_asset_risk_rule_test(
+    asset_id: str,
+    payload: AgentAssetRiskRuleSimulationRequest,
+    _: RuleEditorUser,
+    db: DbSession,
+) -> AgentAssetRiskRuleSimulationRead:
+    try:
+        return AgentAssetService(db).simulate_risk_rule_message(asset_id, payload)
+    except Exception as exc:
+        _handle_asset_error(exc)
+
+
+@router.post(
+    "/{asset_id}/risk-rule-tests/sample",
+    response_model=AgentAssetRiskRuleTestRunRead,
+    summary="执行风险规则快速样例测试",
+    description="使用人工样例或系统默认样例执行当前 JSON 风险规则，不依赖大模型判断结果。",
+)
+def run_agent_asset_risk_rule_sample_test(
+    asset_id: str,
+    payload: AgentAssetRiskRuleSampleTestRequest,
+    current_user: RuleEditorUser,
+    db: DbSession,
+    x_actor: ActorHeader = None,
+    x_request_id: RequestIdHeader = None,
+) -> AgentAssetRiskRuleTestRunRead:
+    try:
+        return AgentAssetService(db).run_risk_rule_sample_test(
+            asset_id,
+            payload,
+            actor=(x_actor or current_user.name or "system").strip() or "system",
+            request_id=x_request_id,
+        )
+    except Exception as exc:
+        _handle_asset_error(exc)
+
+
+@router.post(
+    "/{asset_id}/risk-rule-tests/scenario",
+    response_model=AgentAssetRiskRuleTestRunRead,
+    summary="执行风险规则真实场景试运行",
+    description="按测试意图读取真实业务样本并沙盒执行风险规则，不写回业务单据。",
+)
+def run_agent_asset_risk_rule_scenario_test(
+    asset_id: str,
+    payload: AgentAssetRiskRuleScenarioTestRequest,
+    current_user: RuleEditorUser,
+    db: DbSession,
+    x_actor: ActorHeader = None,
+    x_request_id: RequestIdHeader = None,
+) -> AgentAssetRiskRuleTestRunRead:
+    try:
+        return AgentAssetService(db).run_risk_rule_scenario_test(
+            asset_id,
+            payload,
+            actor=(x_actor or current_user.name or "system").strip() or "system",
+            request_id=x_request_id,
+        )
+    except Exception as exc:
+        _handle_asset_error(exc)
+
+
+@router.post(
+    "/{asset_id}/risk-rule-tests/report",
+    response_model=AgentAssetRiskRuleTestRunRead,
+    summary="确认风险规则测试报告",
+    description="在样例测试和真实场景试运行通过后，保存当前版本测试通过记录。",
+)
+def confirm_agent_asset_risk_rule_test_report(
+    asset_id: str,
+    payload: AgentAssetRiskRuleReportRequest,
+    current_user: RuleEditorUser,
+    db: DbSession,
+    x_actor: ActorHeader = None,
+    x_request_id: RequestIdHeader = None,
+) -> AgentAssetRiskRuleTestRunRead:
+    try:
+        return AgentAssetService(db).confirm_risk_rule_test_report(
+            asset_id,
+            payload,
+            actor=(x_actor or current_user.name or "system").strip() or "system",
+            request_id=x_request_id,
+        )
+    except Exception as exc:
+        _handle_asset_error(exc)
+
+
@router.put(
    "/{asset_id}/rule-json",
    response_model=AgentAssetRuleJsonRead,
@@ -586,6 +705,112 @@ def activate_agent_asset(
        _handle_asset_error(exc)


+@router.post(
+    "/{asset_id}/risk-rule-enabled",
+    response_model=AgentAssetRead,
+    summary="设置风险规则启用状态",
+    description=(
+        "高级管理人员可独立启用或停用 JSON 风险规则；停用后即使已上线也不会进入真实业务扫描。"
+    ),
+)
+def set_agent_asset_risk_rule_enabled(
+    asset_id: str,
+    payload: AgentAssetRiskRuleEnabledUpdate,
+    current_user: RuleReviewerUser,
+    db: DbSession,
+    x_actor: ActorHeader = None,
+    x_request_id: RequestIdHeader = None,
+) -> AgentAssetRead:
+    try:
+        asset = AgentAssetService(db).set_risk_rule_enabled(
+            asset_id,
+            enabled=payload.enabled,
+            actor=(x_actor or current_user.name or "system").strip() or "system",
+            request_id=x_request_id,
+        )
+        detail = AgentAssetService(db).get_asset(asset.id)
+        if detail is None:
+            raise LookupError("Asset not found")
+        return detail
+    except Exception as exc:
+        _handle_asset_error(exc)
+
+
+@router.post(
+    "/{asset_id}/return",
+    response_model=AgentAssetRiskRuleLatestTestSummary,
+    summary="回退待审核风险规则",
+    description="高级管理人员将待审核风险规则回退到草稿，并记录回退原因。",
+)
+def return_agent_asset_risk_rule(
+    asset_id: str,
+    payload: AgentAssetRiskRuleReturnRequest,
+    current_user: RuleReviewerUser,
+    db: DbSession,
+    x_actor: ActorHeader = None,
+    x_request_id: RequestIdHeader = None,
+) -> AgentAssetRiskRuleLatestTestSummary:
+    try:
+        return AgentAssetService(db).return_risk_rule(
+            asset_id,
+            note=payload.note,
+            actor=(x_actor or current_user.name or "system").strip() or "system",
+            request_id=x_request_id,
+        )
+    except Exception as exc:
+        _handle_asset_error(exc)
+
+
+@router.post(
+    "/{asset_id}/publish",
+    response_model=AgentAssetRead,
+    summary="审核并发布风险规则",
+    description="高级管理人员确认测试通过后，将待审核风险规则一次性审核通过并发布上线。",
+)
+def publish_agent_asset_risk_rule(
+    asset_id: str,
+    current_user: RuleReviewerUser,
+    db: DbSession,
+    x_actor: ActorHeader = None,
+    x_request_id: RequestIdHeader = None,
+) -> AgentAssetRead:
+    try:
+        asset = AgentAssetService(db).publish_risk_rule(
+            asset_id,
+            actor=(x_actor or current_user.name or "system").strip() or "system",
+            request_id=x_request_id,
+        )
+        detail = AgentAssetService(db).get_asset(asset.id)
+        if detail is None:
+            raise LookupError("Asset not found")
+        return detail
+    except Exception as exc:
+        _handle_asset_error(exc)
+
+
+@router.delete(
+    "/{asset_id}",
+    status_code=status.HTTP_204_NO_CONTENT,
+    summary="删除未发布风险规则",
+    description="仅允许删除从未发布过的 JSON 风险规则，并同步删除规则 JSON 文件。",
+)
+def delete_agent_asset(
+    asset_id: str,
+    current_user: RuleEditorUser,
+    db: DbSession,
+    x_actor: ActorHeader = None,
+    x_request_id: RequestIdHeader = None,
+) -> None:
+    try:
+        AgentAssetService(db).delete_unpublished_asset(
+            asset_id,
+            actor=(x_actor or current_user.name or "system").strip() or "system",
+            request_id=x_request_id,
+        )
+    except Exception as exc:
+        _handle_asset_error(exc)
+
+
@router.post(
    "/{asset_id}/versions/{version}/restore",
    response_model=AgentAssetRead,
--- a/server/src/app/db/base.py
+++ b/server/src/app/db/base.py
@@ -1,6 +1,6 @@
 from app.db.base_class import Base
 from app.models.agent_conversation import AgentConversation, AgentConversationMessage
-from app.models.agent_asset import AgentAsset, AgentAssetReview, AgentAssetVersion
+from app.models.agent_asset import AgentAsset, AgentAssetReview, AgentAssetTestRun, AgentAssetVersion
 from app.models.agent_run import AgentRun, AgentToolCall, SemanticParseLog
 from app.models.approval import ApprovalRecord
 from app.models.audit_log import AuditLog
@@ -27,6 +27,7 @@ __all__ = [
    "AgentConversationMessage",
    "AgentAsset",
    "AgentAssetReview",
+    "AgentAssetTestRun",
    "AgentAssetVersion",
    "AgentRun",
    "AgentToolCall",
--- a/server/src/app/models/init.py
+++ b/server/src/app/models/init.py
@@ -11,6 +11,8 @@ from app.models.financial_record import (
    ExpenseClaim,
    ExpenseClaimItem,
 )
+from app.models.hermes_config import HermesTaskConfig, HermesTaskExecutionLog
+from app.models.hermes_report import HermesRiskReport
 from app.models.organization import OrganizationUnit
 from app.models.reimbursement import ReimbursementRequest
 from app.models.role import Role
@@ -34,6 +36,9 @@ __all__ = [
    "EmployeeChangeLog",
    "ExpenseClaim",
    "ExpenseClaimItem",
+    "HermesTaskConfig",
+    "HermesTaskExecutionLog",
+    "HermesRiskReport",
    "OrganizationUnit",
    "ReimbursementRequest",
    "Role",
--- a/server/src/app/models/agent_asset.py
+++ b/server/src/app/models/agent_asset.py
@@ -4,7 +4,7 @@ import uuid
 from datetime import datetime
 from typing import Any

-from sqlalchemy import DateTime, ForeignKey, String, Text, UniqueConstraint, func
+from sqlalchemy import Boolean, DateTime, ForeignKey, String, Text, UniqueConstraint, func
 from sqlalchemy.orm import Mapped, mapped_column, relationship
 from sqlalchemy.types import JSON

@@ -46,6 +46,12 @@ class AgentAsset(Base):
        order_by="desc(AgentAssetReview.created_at)",
    )
    scheduled_runs = relationship("AgentRun", back_populates="task_asset")
+    test_runs = relationship(
+        "AgentAssetTestRun",
+        back_populates="asset",
+        cascade="all, delete-orphan",
+        order_by="desc(AgentAssetTestRun.created_at)",
+    )


 class AgentAssetVersion(Base):
@@ -79,3 +85,21 @@ class AgentAssetReview(Base):
    created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())

    asset = relationship("AgentAsset", back_populates="reviews")
+
+
+class AgentAssetTestRun(Base):
+    __tablename__ = "agent_asset_test_runs"
+
+    id: Mapped[str] = mapped_column(String(36), primary_key=True, default=lambda: str(uuid.uuid4()))
+    asset_id: Mapped[str] = mapped_column(ForeignKey("agent_assets.id"), index=True)
+    version: Mapped[str] = mapped_column(String(30), index=True)
+    test_type: Mapped[str] = mapped_column(String(30), index=True)
+    status: Mapped[str] = mapped_column(String(20), index=True)
+    passed: Mapped[bool] = mapped_column(Boolean, default=False, index=True)
+    summary: Mapped[str] = mapped_column(Text(), default="")
+    input_json: Mapped[dict[str, Any]] = mapped_column(JSON, default=dict)
+    result_json: Mapped[dict[str, Any]] = mapped_column(JSON, default=dict)
+    created_by: Mapped[str] = mapped_column(String(100))
+    created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
+
+    asset = relationship("AgentAsset", back_populates="test_runs")
--- a/server/src/app/models/employee.py
+++ b/server/src/app/models/employee.py
@@ -3,7 +3,7 @@ from __future__ import annotations
 import uuid
 from datetime import date, datetime

-from sqlalchemy import Boolean, Column, Date, DateTime, ForeignKey, String, Table, func
+from sqlalchemy import Boolean, Column, Date, DateTime, ForeignKey, Integer, String, Table, func
 from sqlalchemy.orm import Mapped, mapped_column, relationship

 from app.db.base_class import Base
@@ -35,6 +35,7 @@ class Employee(Base):
    password_hash: Mapped[str | None] = mapped_column(String(255), nullable=True)
    employment_status: Mapped[str] = mapped_column(String(30), default="在职", index=True)
    sync_state: Mapped[str] = mapped_column(String(30), default="已同步")
+    compliance_score: Mapped[int] = mapped_column(Integer, default=100)
    spotlight: Mapped[bool] = mapped_column(Boolean, default=False)
    last_sync_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
    organization_unit_id: Mapped[str | None] = mapped_column(
--- a/server/src/app/models/financial_record.py
+++ b/server/src/app/models/financial_record.py
@@ -5,7 +5,7 @@ from datetime import date, datetime
 from decimal import Decimal
 from typing import Any

-from sqlalchemy import Date, DateTime, ForeignKey, Integer, Numeric, String, Text, func
+from sqlalchemy import Boolean, Date, DateTime, ForeignKey, Integer, Numeric, String, Text, func
 from sqlalchemy.orm import Mapped, mapped_column, relationship
 from sqlalchemy.types import JSON

@@ -39,6 +39,8 @@ class ExpenseClaim(Base):
    status: Mapped[str] = mapped_column(String(30), index=True)
    approval_stage: Mapped[str | None] = mapped_column(String(50), nullable=True)
    risk_flags_json: Mapped[list[Any]] = mapped_column(JSON, default=list)
+    hermes_scanned_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
+    hermes_risk_flag: Mapped[bool] = mapped_column(Boolean, default=False, index=True)
    created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
    updated_at: Mapped[datetime] = mapped_column(
        DateTime(timezone=True), server_default=func.now(), onupdate=func.now()
--- a/server/src/app/models/hermes_config.py
+++ b/server/src/app/models/hermes_config.py
@@ -0,0 +1,48 @@
+from __future__ import annotations
+
+import uuid
+from datetime import datetime
+from typing import Any
+
+from sqlalchemy import Boolean, DateTime, ForeignKey, String, Text, func
+from sqlalchemy.orm import Mapped, mapped_column, relationship
+from sqlalchemy.types import JSON
+
+from app.db.base_class import Base
+
+
+class HermesTaskConfig(Base):
+    __tablename__ = "hermes_task_configs"
+
+    id: Mapped[str] = mapped_column(String(36), primary_key=True, default=lambda: str(uuid.uuid4()))
+    task_type: Mapped[str] = mapped_column(String(50), index=True)
+    cron_expression: Mapped[str] = mapped_column(String(100))
+    is_enabled: Mapped[bool] = mapped_column(Boolean, default=True)
+    payload_template: Mapped[dict[str, Any]] = mapped_column(JSON, default=dict)
+    
+    created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
+    updated_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), server_default=func.now(), onupdate=func.now()
+    )
+
+    execution_logs = relationship(
+        "HermesTaskExecutionLog",
+        back_populates="config",
+        cascade="all, delete-orphan",
+        order_by="desc(HermesTaskExecutionLog.started_at)",
+    )
+
+
+class HermesTaskExecutionLog(Base):
+    __tablename__ = "hermes_task_execution_logs"
+
+    id: Mapped[str] = mapped_column(String(36), primary_key=True, default=lambda: str(uuid.uuid4()))
+    config_id: Mapped[str] = mapped_column(String(36), ForeignKey("hermes_task_configs.id"), index=True)
+    status: Mapped[str] = mapped_column(String(30), index=True)
+    result_summary: Mapped[str | None] = mapped_column(String(255), nullable=True)
+    error_trace: Mapped[str | None] = mapped_column(Text(), nullable=True)
+    
+    started_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
+    completed_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
+
+    config = relationship("HermesTaskConfig", back_populates="execution_logs")
--- a/server/src/app/models/hermes_report.py
+++ b/server/src/app/models/hermes_report.py
@@ -0,0 +1,34 @@
+from __future__ import annotations
+
+import uuid
+from datetime import datetime
+from typing import Any
+
+from sqlalchemy import DateTime, ForeignKey, String, Text, func
+from sqlalchemy.orm import Mapped, mapped_column, relationship
+from sqlalchemy.types import JSON
+
+from app.db.base_class import Base
+
+
+class HermesRiskReport(Base):
+    __tablename__ = "hermes_risk_reports"
+
+    id: Mapped[str] = mapped_column(String(36), primary_key=True, default=lambda: str(uuid.uuid4()))
+    claim_id: Mapped[str] = mapped_column(ForeignKey("expense_claims.id"), index=True)
+    execution_log_id: Mapped[str] = mapped_column(ForeignKey("hermes_task_execution_logs.id"), index=True)
+    
+    risk_level: Mapped[str] = mapped_column(String(20), index=True)
+    risk_type: Mapped[str] = mapped_column(String(50), index=True)
+    risk_description: Mapped[str] = mapped_column(Text())
+    
+    related_claim_ids: Mapped[list[str]] = mapped_column(JSON, default=list)
+    status: Mapped[str] = mapped_column(String(30), default="pending_review", index=True)
+    
+    created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
+    updated_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), server_default=func.now(), onupdate=func.now()
+    )
+
+    claim = relationship("ExpenseClaim", foreign_keys=[claim_id])
+    execution_log = relationship("HermesTaskExecutionLog", foreign_keys=[execution_log_id])
--- a/server/src/app/repositories/agent_asset.py
+++ b/server/src/app/repositories/agent_asset.py
@@ -3,7 +3,12 @@ from __future__ import annotations
 from sqlalchemy import or_, select
 from sqlalchemy.orm import Session

-from app.models.agent_asset import AgentAsset, AgentAssetReview, AgentAssetVersion
+from app.models.agent_asset import (
+    AgentAsset,
+    AgentAssetReview,
+    AgentAssetTestRun,
+    AgentAssetVersion,
+)


 class AgentAssetRepository:
@@ -84,6 +89,17 @@ class AgentAssetRepository:
            stmt = stmt.limit(limit)
        return list(self.db.scalars(stmt).all())

+    def list_reviews_for_assets(self, asset_ids: list[str]) -> list[AgentAssetReview]:
+        if not asset_ids:
+            return []
+
+        stmt = (
+            select(AgentAssetReview)
+            .where(AgentAssetReview.asset_id.in_(asset_ids))
+            .order_by(AgentAssetReview.asset_id, AgentAssetReview.created_at.desc())
+        )
+        return list(self.db.scalars(stmt).all())
+
    def get_review(
        self, asset_id: str, version: str, review_status: str | None = None
    ) -> AgentAssetReview | None:
@@ -119,3 +135,54 @@ class AgentAssetRepository:
        self.db.commit()
        self.db.refresh(review)
        return review
+
+    def list_test_runs(
+        self,
+        asset_id: str,
+        *,
+        version: str | None = None,
+        test_type: str | None = None,
+        status: str | None = None,
+        limit: int | None = None,
+    ) -> list[AgentAssetTestRun]:
+        stmt = (
+            select(AgentAssetTestRun)
+            .where(AgentAssetTestRun.asset_id == asset_id)
+            .order_by(AgentAssetTestRun.created_at.desc())
+        )
+        if version:
+            stmt = stmt.where(AgentAssetTestRun.version == version)
+        if test_type:
+            stmt = stmt.where(AgentAssetTestRun.test_type == test_type)
+        if status:
+            stmt = stmt.where(AgentAssetTestRun.status == status)
+        if limit is not None:
+            stmt = stmt.limit(limit)
+        return list(self.db.scalars(stmt).all())
+
+    def get_latest_test_run(
+        self,
+        asset_id: str,
+        *,
+        version: str | None = None,
+        test_type: str | None = None,
+        status: str | None = None,
+    ) -> AgentAssetTestRun | None:
+        items = self.list_test_runs(
+            asset_id,
+            version=version,
+            test_type=test_type,
+            status=status,
+            limit=1,
+        )
+        return items[0] if items else None
+
+    def create_test_run(self, test_run: AgentAssetTestRun) -> AgentAssetTestRun:
+        self.db.add(test_run)
+        self.db.commit()
+        self.db.refresh(test_run)
+        return test_run
+
+    def delete_asset(self, asset: AgentAsset) -> None:
+        self.db.delete(asset)
+        self.db.commit()
--- a/server/src/app/schemas/agent_asset.py
+++ b/server/src/app/schemas/agent_asset.py
@@ -112,8 +112,111 @@ class AgentAssetRuleJsonRead(BaseModel):

 class AgentAssetRiskRuleGenerateRequest(BaseModel):
    business_domain: AgentAssetDomain = AgentAssetDomain.EXPENSE
+    expense_category: str | None = Field(default=None, max_length=40)
    risk_level: str = Field(default="medium", pattern="^(low|medium|high)$")
    natural_language: str = Field(min_length=8, max_length=2000)
+    requires_attachment: bool = False
+
+
+class AgentAssetRiskRuleSampleCase(BaseModel):
+    case_id: str | None = Field(default=None, max_length=60)
+    name: str = Field(default="测试样例", min_length=1, max_length=80)
+    values: dict[str, Any] = Field(default_factory=dict)
+    expected_hit: bool = True
+    expected_severity: str | None = Field(default=None, max_length=20)
+    note: str | None = None
+
+
+class AgentAssetRiskRuleSampleTestRequest(BaseModel):
+    version: str | None = Field(default=None, max_length=30)
+    cases: list[AgentAssetRiskRuleSampleCase] = Field(default_factory=list)
+
+
+class AgentAssetRiskRuleScenarioTestRequest(BaseModel):
+    version: str | None = Field(default=None, max_length=30)
+    intent: str = Field(default="", max_length=1000)
+    filters: dict[str, Any] = Field(default_factory=dict)
+
+
+class AgentAssetRiskRuleReportRequest(BaseModel):
+    version: str | None = Field(default=None, max_length=30)
+    confirm_passed: bool = True
+    note: str | None = Field(default=None, max_length=1000)
+
+
+class AgentAssetRiskRuleSimulationAttachment(BaseModel):
+    name: str = Field(default="", max_length=240)
+    content_type: str | None = Field(default=None, max_length=120)
+    size: int | None = Field(default=None, ge=0)
+    note: str | None = Field(default=None, max_length=500)
+    ocr_text: str | None = Field(default=None, max_length=20000)
+    summary: str | None = Field(default=None, max_length=2000)
+    document_type: str | None = Field(default=None, max_length=80)
+    document_type_label: str | None = Field(default=None, max_length=120)
+    scene_code: str | None = Field(default=None, max_length=80)
+    scene_label: str | None = Field(default=None, max_length=120)
+    avg_score: float | None = Field(default=None, ge=0.0, le=1.0)
+    recognition_status: str | None = Field(default=None, max_length=40)
+    document_fields: list[dict[str, Any]] = Field(default_factory=list)
+
+
+class AgentAssetRiskRuleSimulationRequest(BaseModel):
+    version: str | None = Field(default=None, max_length=30)
+    message: str = Field(default="", max_length=4000)
+    field_values: dict[str, Any] = Field(default_factory=dict)
+    attachments: list[AgentAssetRiskRuleSimulationAttachment] = Field(default_factory=list)
+
+
+class AgentAssetRiskRuleSimulationRead(BaseModel):
+    version: str
+    ready: bool = True
+    stage: str = "executed"
+    hit: bool
+    severity: str = "none"
+    severity_label: str = "未命中"
+    summary: str
+    blocking_reason: str = ""
+    message: str = ""
+    field_values: dict[str, Any] = Field(default_factory=dict)
+    evidence: dict[str, Any] = Field(default_factory=dict)
+    attachments: list[dict[str, Any]] = Field(default_factory=list)
+    recognized_fields: list[dict[str, Any]] = Field(default_factory=list)
+    missing_fields: list[dict[str, Any]] = Field(default_factory=list)
+    recognition_summary: list[dict[str, Any]] = Field(default_factory=list)
+    execution_mode: str = "risk_rule_simulation"
+    created_at: datetime
+
+
+class AgentAssetRiskRuleReturnRequest(BaseModel):
+    note: str = Field(min_length=1, max_length=1000)
+
+
+class AgentAssetRiskRuleEnabledUpdate(BaseModel):
+    enabled: bool
+
+
+class AgentAssetRiskRuleTestRunRead(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    id: str
+    asset_id: str
+    version: str
+    test_type: str
+    status: str
+    passed: bool
+    summary: str
+    input_json: dict[str, Any] = Field(default_factory=dict)
+    result_json: dict[str, Any] = Field(default_factory=dict)
+    created_by: str
+    created_at: datetime
+
+
+class AgentAssetRiskRuleLatestTestSummary(BaseModel):
+    version: str = ""
+    sample: AgentAssetRiskRuleTestRunRead | None = None
+    scenario: AgentAssetRiskRuleTestRunRead | None = None
+    report: AgentAssetRiskRuleTestRunRead | None = None
+    test_passed: bool = False


 class AgentAssetVersionTimelineItemRead(BaseModel):
@@ -187,6 +290,8 @@ class AgentAssetListItem(BaseModel):
    config_json: dict[str, Any]
    change_count: int = 0
    modified_by: str | None = None
+    published_by: str | None = None
+    published_at: datetime | None = None
    created_at: datetime
    updated_at: datetime

@@ -197,3 +302,4 @@ class AgentAssetRead(AgentAssetListItem):
    current_version_change_note: str | None = None
    recent_versions: list[AgentAssetVersionRead] = Field(default_factory=list)
    latest_review: AgentAssetReviewRead | None = None
+    latest_test_summary: AgentAssetRiskRuleLatestTestSummary | None = None
--- a/server/src/app/schemas/settings.py
+++ b/server/src/app/schemas/settings.py
@@ -164,6 +164,7 @@ class SettingsRead(BaseModel):
    companyForm: SettingsCompanyForm
    adminForm: SettingsAdminForm
    sessionForm: SettingsSessionForm
+    hermesForm: dict
    llmForm: SettingsLlmForm
    renderForm: SettingsRenderForm
    logForm: SettingsLogForm
@@ -174,6 +175,7 @@ class SettingsWrite(BaseModel):
    companyForm: SettingsCompanyForm
    adminForm: SettingsAdminForm
    sessionForm: SettingsSessionForm
+    hermesForm: dict
    llmForm: SettingsLlmForm
    renderForm: SettingsRenderForm
    logForm: SettingsLogForm
--- a/server/src/app/services/agent_asset_risk_rule_simulation.py
+++ b/server/src/app/services/agent_asset_risk_rule_simulation.py
@@ -0,0 +1,552 @@
+from __future__ import annotations
+
+import re
+from datetime import UTC, date, datetime
+from typing import Any
+
+from app.schemas.agent_asset import (
+    AgentAssetRiskRuleSimulationAttachment,
+    AgentAssetRiskRuleSimulationRead,
+    AgentAssetRiskRuleSimulationRequest,
+)
+from app.services.risk_rule_template_executor import RiskRuleTemplateExecutor
+
+
+class AgentAssetRiskRuleSimulationMixin:
+    def simulate_risk_rule_message(
+        self,
+        asset_id: str,
+        body: AgentAssetRiskRuleSimulationRequest,
+    ) -> AgentAssetRiskRuleSimulationRead:
+        _, version, manifest = self._load_risk_rule_for_test(asset_id, body.version)
+        attachments = self._normalize_simulation_attachments(body.attachments)
+        field_values, source_map, recognized_fields = self._build_simulation_field_values(
+            manifest,
+            message=body.message,
+            explicit_values=body.field_values,
+            attachments=attachments,
+        )
+        recognition_summary = self._build_recognition_summary(attachments)
+        required_keys = self._extract_execution_field_keys(manifest)
+        missing_fields = self._build_missing_fields(
+            manifest,
+            field_values=field_values,
+            source_map=source_map,
+            required_keys=required_keys,
+        )
+        block = self._resolve_simulation_block(
+            manifest,
+            message=body.message,
+            attachments=attachments,
+            missing_fields=missing_fields,
+        )
+        if block:
+            return AgentAssetRiskRuleSimulationRead(
+                version=version,
+                ready=False,
+                stage=block["stage"],
+                hit=False,
+                severity="none",
+                severity_label="待补充",
+                summary=block["summary"],
+                blocking_reason=block["reason"],
+                field_values=field_values,
+                attachments=attachments,
+                recognized_fields=recognized_fields,
+                missing_fields=missing_fields,
+                recognition_summary=recognition_summary,
+                created_at=datetime.now(UTC),
+            )
+
+        claim, contexts = self._build_synthetic_claim(field_values, manifest)
+        result = RiskRuleTemplateExecutor().evaluate(manifest, claim=claim, contexts=contexts)
+        hit = result is not None
+        severity = (
+            str((manifest.get("outcomes") or {}).get("fail", {}).get("severity") or "medium")
+            if hit
+            else "none"
+        )
+        severity_label = self._risk_severity_label(severity)
+        message = str(result.get("message") or "") if isinstance(result, dict) else ""
+        summary = (
+            f"本次仿真命中{severity_label}，仅生成风险识别结果，不创建业务单据。"
+            if hit
+            else "本次仿真未命中风险，仅完成规则识别，不创建业务单据。"
+        )
+        evidence = result.get("evidence") if isinstance(result, dict) else {}
+        return AgentAssetRiskRuleSimulationRead(
+            version=version,
+            ready=True,
+            stage="executed",
+            hit=hit,
+            severity=severity,
+            severity_label=severity_label,
+            summary=summary,
+            message=message,
+            field_values=field_values,
+            evidence=evidence if isinstance(evidence, dict) else {},
+            attachments=attachments,
+            recognized_fields=recognized_fields,
+            missing_fields=[],
+            recognition_summary=recognition_summary,
+            created_at=datetime.now(UTC),
+        )
+
+    def _build_simulation_field_values(
+        self,
+        manifest: dict[str, Any],
+        *,
+        message: str,
+        explicit_values: dict[str, Any],
+        attachments: list[dict[str, Any]],
+    ) -> tuple[dict[str, Any], dict[str, str], list[dict[str, Any]]]:
+        fields = self._extract_manifest_fields(manifest)
+        values: dict[str, Any] = {}
+        source_map: dict[str, str] = {}
+        safe_explicit_values = explicit_values if isinstance(explicit_values, dict) else {}
+        corpus = self._build_simulation_corpus(message, attachments)
+        city_mentions = self._extract_city_mentions(corpus)
+
+        for field in fields:
+            key = field["key"]
+            explicit_value = safe_explicit_values.get(key)
+            if self._has_meaningful_value(explicit_value):
+                values[key] = explicit_value
+                source_map[key] = "manual"
+                continue
+            attachment_value = self._find_attachment_field_value(
+                key,
+                field.get("label") or key,
+                attachments,
+            )
+            if self._has_meaningful_value(attachment_value):
+                values[key] = attachment_value
+                source_map[key] = "ocr"
+                continue
+            inferred = self._infer_simulation_value(
+                key,
+                field.get("label") or key,
+                corpus=corpus,
+                city_mentions=city_mentions,
+            )
+            if self._has_meaningful_value(inferred):
+                values[key] = inferred
+                source_map[key] = "inferred"
+
+        self._apply_compare_city_hints(manifest, values, source_map, city_mentions)
+        recognized_fields = self._build_recognized_fields(fields, values, source_map)
+        return values, source_map, recognized_fields
+
+    def _infer_simulation_value(
+        self,
+        field_key: str,
+        label: str,
+        *,
+        corpus: str,
+        city_mentions: list[str],
+    ) -> Any:
+        key_text = f"{field_key} {label}".lower()
+        if field_key.endswith("route_cities"):
+            return city_mentions or []
+        if "city" in field_key or "location" in field_key:
+            if any(
+                token in key_text
+                for token in ("hotel", "invoice", "attachment", "发票", "酒店", "住宿")
+            ):
+                return city_mentions[0] if city_mentions else ""
+            if any(token in key_text for token in ("route", "trip", "目的", "行程", "申报")):
+                return (
+                    city_mentions[1]
+                    if len(city_mentions) > 1
+                    else (city_mentions[0] if city_mentions else "")
+                )
+            return city_mentions[0] if city_mentions else ""
+        if field_key.endswith("amount"):
+            return self._extract_amount(corpus)
+        if field_key.endswith("issue_date") or field_key.endswith("item_date"):
+            return self._extract_iso_date(corpus)
+        if field_key.endswith("invoice_no"):
+            return self._extract_invoice_no(corpus)
+        if field_key.endswith("ocr_text"):
+            return corpus
+        if field_key.endswith("goods_name"):
+            return self._infer_goods_name(corpus)
+        if field_key.endswith("item_type"):
+            return self._infer_item_type(corpus)
+        if field_key.endswith("reason") or field_key.endswith("item_reason"):
+            return corpus or "仿真测试报销事由"
+        return None
+
+    def _apply_compare_city_hints(
+        self,
+        manifest: dict[str, Any],
+        values: dict[str, Any],
+        source_map: dict[str, str],
+        city_mentions: list[str],
+    ) -> None:
+        if len(city_mentions) < 2:
+            return
+        params = manifest.get("params") if isinstance(manifest.get("params"), dict) else {}
+        conditions = params.get("conditions") if isinstance(params.get("conditions"), list) else []
+        for condition in conditions:
+            if not isinstance(condition, dict):
+                continue
+            left = str(condition.get("left") or "").strip()
+            right = str(condition.get("right") or "").strip()
+            if not left or not right:
+                continue
+            if self._looks_like_city_field(left):
+                values[left] = city_mentions[0]
+                source_map[left] = source_map.get(left) or "inferred"
+            if self._looks_like_city_field(right):
+                values[right] = city_mentions[1]
+                source_map[right] = source_map.get(right) or "inferred"
+
+    @staticmethod
+    def _normalize_simulation_attachments(
+        attachments: list[AgentAssetRiskRuleSimulationAttachment],
+    ) -> list[dict[str, Any]]:
+        normalized: list[dict[str, Any]] = []
+        for item in list(attachments or [])[:12]:
+            normalized.append(
+                {
+                    "name": str(item.name or "").strip(),
+                    "content_type": str(item.content_type or "").strip(),
+                    "size": item.size or 0,
+                    "note": str(item.note or "").strip(),
+                    "ocr_text": str(item.ocr_text or "").strip(),
+                    "summary": str(item.summary or "").strip(),
+                    "document_type": str(item.document_type or "").strip(),
+                    "document_type_label": str(item.document_type_label or "").strip(),
+                    "scene_code": str(item.scene_code or "").strip(),
+                    "scene_label": str(item.scene_label or "").strip(),
+                    "avg_score": float(item.avg_score or 0.0),
+                    "recognition_status": str(item.recognition_status or "").strip(),
+                    "document_fields": AgentAssetRiskRuleSimulationMixin._normalize_document_fields(
+                        item.document_fields
+                    ),
+                }
+            )
+        return normalized
+
+    @staticmethod
+    def _build_simulation_corpus(message: str, attachments: list[dict[str, Any]]) -> str:
+        parts = [str(message or "").strip()]
+        for item in attachments:
+            parts.append(str(item.get("name") or "").strip())
+            parts.append(str(item.get("note") or "").strip())
+            parts.append(str(item.get("summary") or "").strip())
+            parts.append(str(item.get("ocr_text") or "").strip())
+            for field in list(item.get("document_fields") or []):
+                if isinstance(field, dict):
+                    parts.append(str(field.get("value") or "").strip())
+        return "\n".join(part for part in parts if part)
+
+    @staticmethod
+    def _normalize_document_fields(fields: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        normalized: list[dict[str, Any]] = []
+        for field in list(fields or [])[:30]:
+            if not isinstance(field, dict):
+                continue
+            key = str(field.get("key") or "").strip()
+            label = str(field.get("label") or "").strip()
+            value = field.get("value")
+            if key and label and AgentAssetRiskRuleSimulationMixin._has_meaningful_value(value):
+                normalized.append({"key": key, "label": label, "value": value})
+        return normalized
+
+    def _find_attachment_field_value(
+        self,
+        field_key: str,
+        label: str,
+        attachments: list[dict[str, Any]],
+    ) -> Any:
+        short_key = field_key.removeprefix("attachment.")
+        for attachment in attachments:
+            if short_key == "ocr_text":
+                value = attachment.get("ocr_text") or attachment.get("summary")
+                if self._has_meaningful_value(value):
+                    return value
+            for field in list(attachment.get("document_fields") or []):
+                if not isinstance(field, dict):
+                    continue
+                candidate_key = str(field.get("key") or "").strip().lower()
+                candidate_label = str(field.get("label") or "").strip()
+                if self._field_matches_simulation_key(
+                    candidate_key, candidate_label, short_key, label
+                ):
+                    return field.get("value")
+        return None
+
+    @staticmethod
+    def _field_matches_simulation_key(
+        candidate_key: str,
+        candidate_label: str,
+        short_key: str,
+        target_label: str,
+    ) -> bool:
+        compact_candidate = candidate_key.replace("_", "")
+        compact_target = short_key.replace("_", "").lower()
+        if compact_target and compact_target in compact_candidate:
+            return True
+        label_text = f"{candidate_label} {target_label}"
+        label_map = {
+            "invoice_no": ("发票号", "发票号码", "票号"),
+            "hotel_city": ("住宿城市", "酒店城市", "酒店地点", "住宿", "酒店"),
+            "route_cities": ("行程", "路线", "目的地", "出差城市"),
+            "goods_name": ("品名", "商品", "服务名称"),
+            "amount": ("金额", "价税合计", "合计"),
+            "issue_date": ("日期", "开票日期", "发票日期"),
+        }
+        return any(token in label_text for token in label_map.get(short_key, ()))
+
+    def _extract_execution_field_keys(self, manifest: dict[str, Any]) -> list[str]:
+        params = manifest.get("params") if isinstance(manifest.get("params"), dict) else {}
+        template_key = str(manifest.get("template_key") or params.get("template_key") or "").strip()
+        keys: list[str] = []
+        if template_key == "field_compare_v1":
+            conditions = (
+                params.get("conditions") if isinstance(params.get("conditions"), list) else []
+            )
+            for condition in conditions:
+                if not isinstance(condition, dict):
+                    continue
+                for side in ("left", "right"):
+                    key = str(condition.get(side) or "").strip()
+                    if key and key not in keys:
+                        keys.append(key)
+        elif template_key == "keyword_match_v1":
+            for key in self._read_string_list(
+                params.get("search_fields") or params.get("field_keys")
+            ):
+                if key not in keys:
+                    keys.append(key)
+        elif template_key == "field_required_v1":
+            return []
+        return keys
+
+    def _build_missing_fields(
+        self,
+        manifest: dict[str, Any],
+        *,
+        field_values: dict[str, Any],
+        source_map: dict[str, str],
+        required_keys: list[str],
+    ) -> list[dict[str, Any]]:
+        labels = {field["key"]: field["label"] for field in self._extract_manifest_fields(manifest)}
+        missing: list[dict[str, Any]] = []
+        for key in required_keys:
+            value = field_values.get(key)
+            if key not in source_map or not self._has_meaningful_value(value):
+                missing.append({"key": key, "label": labels.get(key, key)})
+        return missing
+
+    def _resolve_simulation_block(
+        self,
+        manifest: dict[str, Any],
+        *,
+        message: str,
+        attachments: list[dict[str, Any]],
+        missing_fields: list[dict[str, Any]],
+    ) -> dict[str, str] | None:
+        has_attachment = bool(attachments)
+        requires_attachment = self._rule_requires_attachment(manifest)
+        has_recognition = any(
+            self._has_meaningful_value(item.get("ocr_text"))
+            or self._has_meaningful_value(item.get("summary"))
+            or self._has_meaningful_value(item.get("document_fields"))
+            for item in attachments
+        )
+        has_user_evidence = self._has_meaningful_user_message(message)
+        if requires_attachment and not has_attachment:
+            return {
+                "stage": "needs_attachment",
+                "summary": "当前规则要求上传附件，暂不能仅凭文字执行风险判断。",
+                "reason": "请上传测试单据，并填写本次测试意图后再执行仿真。",
+            }
+        if requires_attachment and not has_user_evidence:
+            return {
+                "stage": "needs_test_intent",
+                "summary": "当前规则要求附件和测试说明一起进入仿真判断。",
+                "reason": "请补充本次测试意图或关键业务事实，再执行风险识别。",
+            }
+        if has_attachment and not has_recognition and not has_user_evidence:
+            return {
+                "stage": "needs_recognition",
+                "summary": "单据尚未完成识别，暂不能执行风险规则。",
+                "reason": "请先完成 OCR 识别，或在对话中补充票据城市、金额、发票号等关键信息。",
+            }
+        template_key = str(
+            manifest.get("template_key") or (manifest.get("params") or {}).get("template_key") or ""
+        ).strip()
+        if template_key != "field_required_v1" and missing_fields:
+            labels = "、".join(
+                str(item.get("label") or item.get("key")) for item in missing_fields[:4]
+            )
+            return {
+                "stage": "needs_field_confirmation",
+                "summary": f"还缺少规则执行所需字段：{labels}，暂不能判断是否命中。",
+                "reason": "请补充缺失字段，或上传可识别出这些字段的票据后再执行。",
+            }
+        if not has_attachment and not has_user_evidence:
+            return {
+                "stage": "needs_input",
+                "summary": "请先描述测试单据或上传票据，再执行风险识别。",
+                "reason": "当前没有可用于规则判断的业务事实。",
+            }
+        return None
+
+    @staticmethod
+    def _rule_requires_attachment(manifest: dict[str, Any]) -> bool:
+        if bool(manifest.get("requires_attachment")):
+            return True
+        metadata = manifest.get("metadata") if isinstance(manifest.get("metadata"), dict) else {}
+        return bool(metadata.get("requires_attachment"))
+
+    @staticmethod
+    def _has_meaningful_user_message(message: str) -> bool:
+        text = str(message or "").strip()
+        if not text:
+            return False
+        generic_prompts = (
+            "请识别我上传的临时单据是否命中这条风险规则",
+            "请识别上传单据是否命中风险规则",
+        )
+        return not any(prompt in text for prompt in generic_prompts)
+
+    @staticmethod
+    def _build_recognized_fields(
+        fields: list[dict[str, str]],
+        values: dict[str, Any],
+        source_map: dict[str, str],
+    ) -> list[dict[str, Any]]:
+        labels = {field["key"]: field["label"] for field in fields}
+        return [
+            {
+                "key": key,
+                "label": labels.get(key, key),
+                "value": value,
+                "source": source_map.get(key, ""),
+            }
+            for key, value in values.items()
+            if source_map.get(key)
+        ]
+
+    @staticmethod
+    def _build_recognition_summary(attachments: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        return [
+            {
+                "name": item.get("name") or "",
+                "status": item.get("recognition_status")
+                or (
+                    "recognized"
+                    if item.get("ocr_text") or item.get("document_fields")
+                    else "pending"
+                ),
+                "document_type_label": item.get("document_type_label") or "",
+                "scene_label": item.get("scene_label") or "",
+                "summary": item.get("summary") or "",
+                "field_count": len(list(item.get("document_fields") or [])),
+                "avg_score": item.get("avg_score") or 0.0,
+            }
+            for item in attachments
+        ]
+
+    @staticmethod
+    def _extract_city_mentions(text: str) -> list[str]:
+        city_names = [
+            "北京",
+            "上海",
+            "广州",
+            "深圳",
+            "杭州",
+            "南京",
+            "成都",
+            "武汉",
+            "重庆",
+            "天津",
+            "苏州",
+            "西安",
+        ]
+        pattern = "|".join(re.escape(city) for city in city_names)
+        found: list[str] = []
+        for match in re.finditer(pattern, text):
+            city = match.group(0)
+            if city not in found:
+                found.append(city)
+        return found
+
+    @staticmethod
+    def _extract_amount(text: str) -> str:
+        match = re.search(r"(\d{2,8}(?:\.\d{1,2})?)\s*(?:元|块|人民币|CNY)?", text, re.IGNORECASE)
+        return match.group(1) if match else ""
+
+    @staticmethod
+    def _extract_iso_date(text: str) -> str:
+        match = re.search(r"(20\d{2})[-/.年](\d{1,2})[-/.月](\d{1,2})", text)
+        if not match:
+            return ""
+        year, month, day = (int(part) for part in match.groups())
+        try:
+            return date(year, month, day).isoformat()
+        except ValueError:
+            return ""
+
+    @staticmethod
+    def _extract_invoice_no(text: str) -> str:
+        match = re.search(r"(?:发票号|发票号码|票号)[:：\s]*([A-Z0-9-]{6,32})", text, re.IGNORECASE)
+        return match.group(1) if match else ""
+
+    @staticmethod
+    def _infer_item_type(text: str) -> str:
+        if not text:
+            return ""
+        if any(keyword in text for keyword in ("酒店", "住宿", "宾馆")):
+            return "住宿费"
+        if any(keyword in text for keyword in ("机票", "航班", "火车", "高铁", "打车")):
+            return "交通费"
+        if any(keyword in text for keyword in ("餐饮", "餐费", "招待")):
+            return "餐饮费"
+        return "差旅费"
+
+    @staticmethod
+    def _infer_goods_name(text: str) -> str:
+        if not text:
+            return ""
+        if any(keyword in text for keyword in ("酒店", "住宿", "宾馆")):
+            return "住宿服务"
+        if any(keyword in text for keyword in ("机票", "航班", "火车", "高铁", "打车")):
+            return "交通服务"
+        if any(keyword in text for keyword in ("餐饮", "餐费", "招待")):
+            return "餐饮服务"
+        return "报销服务"
+
+    @staticmethod
+    def _looks_like_city_field(field_key: str) -> bool:
+        lowered = field_key.lower()
+        return "city" in lowered or "location" in lowered or lowered.endswith("route_cities")
+
+    @staticmethod
+    def _has_meaningful_value(value: Any) -> bool:
+        if value is None:
+            return False
+        if isinstance(value, str):
+            return bool(value.strip())
+        if isinstance(value, (list, tuple, set, dict)):
+            return bool(value)
+        return True
+
+    @staticmethod
+    def _risk_severity_label(severity: str) -> str:
+        return {
+            "low": "低风险",
+            "medium": "中风险",
+            "high": "高风险",
+            "none": "未命中",
+        }.get(str(severity or "").strip().lower(), "风险")
+
+    @staticmethod
+    def _read_string_list(value: Any) -> list[str]:
+        if not isinstance(value, list):
+            return []
+        return [str(item or "").strip() for item in value if str(item or "").strip()]
--- a/server/src/app/services/agent_asset_risk_rule_testing.py
+++ b/server/src/app/services/agent_asset_risk_rule_testing.py
@@ -0,0 +1,723 @@
+from __future__ import annotations
+
+import re
+from datetime import UTC, date, datetime, timedelta
+from decimal import Decimal, InvalidOperation
+from typing import Any
+
+from sqlalchemy import or_, select
+
+from app.core.agent_enums import (
+    AgentAssetDomain,
+    AgentAssetStatus,
+    AgentAssetType,
+    AgentReviewStatus,
+)
+from app.models.agent_asset import AgentAsset, AgentAssetReview, AgentAssetTestRun
+from app.models.financial_record import ExpenseClaim, ExpenseClaimItem
+from app.schemas.agent_asset import (
+    AgentAssetRiskRuleLatestTestSummary,
+    AgentAssetRiskRuleReportRequest,
+    AgentAssetRiskRuleSampleCase,
+    AgentAssetRiskRuleSampleTestRequest,
+    AgentAssetRiskRuleScenarioTestRequest,
+    AgentAssetRiskRuleTestRunRead,
+)
+from app.services.expense_claims import ExpenseClaimService
+from app.services.risk_rule_template_executor import RiskRuleTemplateExecutor
+
+
+class AgentAssetRiskRuleTestingMixin:
+    def get_latest_risk_rule_test_summary(
+        self,
+        asset_or_id: AgentAsset | str,
+        *,
+        version: str | None = None,
+    ) -> AgentAssetRiskRuleLatestTestSummary:
+        asset = self._resolve_asset(asset_or_id)
+        target_version = self._resolve_target_version(asset, version)
+        sample = self.repository.get_latest_test_run(
+            asset.id, version=target_version, test_type="sample"
+        )
+        scenario = self.repository.get_latest_test_run(
+            asset.id, version=target_version, test_type="scenario"
+        )
+        report = self.repository.get_latest_test_run(
+            asset.id, version=target_version, test_type="report", status="passed"
+        )
+        return AgentAssetRiskRuleLatestTestSummary(
+            version=target_version,
+            sample=self._serialize_test_run(sample),
+            scenario=self._serialize_test_run(scenario),
+            report=self._serialize_test_run(report),
+            test_passed=bool(report and report.passed),
+        )
+
+    def run_risk_rule_sample_test(
+        self,
+        asset_id: str,
+        body: AgentAssetRiskRuleSampleTestRequest,
+        *,
+        actor: str,
+        request_id: str | None = None,
+    ) -> AgentAssetRiskRuleTestRunRead:
+        asset, version, manifest = self._load_risk_rule_for_test(asset_id, body.version)
+        cases = body.cases or self._build_default_sample_cases(manifest)
+        results = [self._run_sample_case(manifest, case) for case in cases]
+        passed = bool(results) and all(item["passed"] for item in results)
+        summary = f"快速样例测试 {'通过' if passed else '未通过'}，共 {len(results)} 条。"
+        return self._create_test_run(
+            asset,
+            version=version,
+            test_type="sample",
+            passed=passed,
+            summary=summary,
+            input_json={"cases": [case.model_dump() for case in cases]},
+            result_json={"cases": results, "case_count": len(results)},
+            actor=actor,
+            request_id=request_id,
+        )
+
+    def run_risk_rule_scenario_test(
+        self,
+        asset_id: str,
+        body: AgentAssetRiskRuleScenarioTestRequest,
+        *,
+        actor: str,
+        request_id: str | None = None,
+    ) -> AgentAssetRiskRuleTestRunRead:
+        asset, version, manifest = self._load_risk_rule_for_test(asset_id, body.version)
+        if asset.domain != AgentAssetDomain.EXPENSE.value:
+            raise ValueError("一期真实场景试运行仅支持报销业务域。")
+
+        parsed_scope = self._parse_scenario_scope(body.intent, body.filters)
+        claims = self._query_expense_claim_samples(parsed_scope)
+        claim_results = [self._run_claim_scenario(manifest, claim) for claim in claims]
+        hit_items = [item for item in claim_results if item["hit"]]
+        severity_counts: dict[str, int] = {}
+        for item in hit_items:
+            severity = str(item.get("severity") or "unknown")
+            severity_counts[severity] = severity_counts.get(severity, 0) + 1
+
+        passed = bool(claim_results)
+        summary = (
+            f"真实场景试运行完成，样本 {len(claim_results)} 条，命中 {len(hit_items)} 条。"
+            if passed
+            else "真实场景试运行未找到可测样本。"
+        )
+        return self._create_test_run(
+            asset,
+            version=version,
+            test_type="scenario",
+            passed=passed,
+            summary=summary,
+            input_json={
+                "intent": body.intent,
+                "filters": body.filters,
+                "parsed_scope": parsed_scope,
+            },
+            result_json={
+                "total_count": len(claim_results),
+                "hit_count": len(hit_items),
+                "severity_counts": severity_counts,
+                "items": claim_results[:50],
+            },
+            actor=actor,
+            request_id=request_id,
+        )
+
+    def confirm_risk_rule_test_report(
+        self,
+        asset_id: str,
+        body: AgentAssetRiskRuleReportRequest,
+        *,
+        actor: str,
+        request_id: str | None = None,
+    ) -> AgentAssetRiskRuleTestRunRead:
+        asset, version, _ = self._load_risk_rule_for_test(asset_id, body.version)
+        sample = self.repository.get_latest_test_run(
+            asset.id, version=version, test_type="sample", status="passed"
+        )
+        scenario = self.repository.get_latest_test_run(
+            asset.id, version=version, test_type="scenario"
+        )
+        if sample is None:
+            raise ValueError("提交审核前必须先完成快速样例测试。")
+        if not body.confirm_passed:
+            raise ValueError("请确认测试通过后再保存测试报告。")
+
+        summary = "测试报告已确认，当前版本可提交审核。"
+        if scenario is None:
+            summary = "快速样例测试已确认通过，真实场景试运行未执行。"
+        elif not scenario.passed:
+            summary = "快速样例测试已确认通过，真实场景试运行未找到可测样本。"
+        return self._create_test_run(
+            asset,
+            version=version,
+            test_type="report",
+            passed=True,
+            summary=summary,
+            input_json={"confirm_passed": True, "note": body.note or ""},
+            result_json={
+                "sample_test_run_id": sample.id,
+                "scenario_test_run_id": scenario.id,
+                "sample_summary": sample.summary,
+                "scenario_summary": scenario.summary,
+            },
+            actor=actor,
+            request_id=request_id,
+        )
+
+    def delete_unpublished_asset(
+        self,
+        asset_id: str,
+        *,
+        actor: str,
+        request_id: str | None = None,
+    ) -> None:
+        asset = self._resolve_asset(asset_id)
+        self._require_json_risk_asset(asset)
+        if str(asset.published_version or "").strip():
+            raise PermissionError("已发布过的风险规则不能删除。")
+
+        before = self._asset_snapshot(asset)
+        self._delete_risk_rule_json_file(asset)
+        self.repository.delete_asset(asset)
+        self.audit_service.log_action(
+            actor=actor,
+            action="delete_agent_asset",
+            resource_type=AgentAssetType.RULE.value,
+            resource_id=asset_id,
+            before_json=before,
+            after_json={"deleted": True},
+            request_id=request_id,
+        )
+
+    def return_risk_rule(
+        self,
+        asset_id: str,
+        *,
+        note: str,
+        actor: str,
+        request_id: str | None = None,
+    ) -> AgentAssetRiskRuleLatestTestSummary:
+        asset = self._resolve_asset(asset_id)
+        self._require_json_risk_asset(asset)
+        version = self._resolve_target_version(asset, None)
+        if asset.status != AgentAssetStatus.REVIEW.value:
+            raise ValueError("只有待审核风险规则可以回退。")
+
+        before = self._asset_snapshot(asset)
+        review = AgentAssetReview(
+            asset_id=asset.id,
+            version=version,
+            reviewer=actor,
+            review_status=AgentReviewStatus.REJECTED.value,
+            review_note=str(note or "审核回退").strip() or "审核回退",
+            reviewed_at=datetime.now(UTC),
+        )
+        self.db.add(review)
+        asset.reviewer = actor
+        asset.status = AgentAssetStatus.DRAFT.value
+        self.db.add(asset)
+        self.db.commit()
+        self.audit_service.log_action(
+            actor=actor,
+            action="return_agent_asset",
+            resource_type=AgentAssetType.RULE.value,
+            resource_id=asset.id,
+            before_json=before,
+            after_json={"version": version, "status": asset.status, "note": note},
+            request_id=request_id,
+        )
+        return self.get_latest_risk_rule_test_summary(asset)
+
+    def publish_risk_rule(
+        self,
+        asset_id: str,
+        *,
+        actor: str,
+        request_id: str | None = None,
+    ) -> AgentAsset:
+        asset = self._resolve_asset(asset_id)
+        self._require_json_risk_asset(asset)
+        version = self._resolve_target_version(asset, None)
+        if asset.status != AgentAssetStatus.REVIEW.value:
+            raise ValueError("只有待审核风险规则可以发布上线。")
+        if not self.get_latest_risk_rule_test_summary(asset, version=version).test_passed:
+            raise PermissionError("当前规则版本尚未完成测试通过确认，不能发布。")
+
+        before = self._asset_snapshot(asset)
+        approved_review = self.repository.get_review(
+            asset.id, version, AgentReviewStatus.APPROVED.value
+        )
+        if approved_review is None:
+            self.db.add(
+                AgentAssetReview(
+                    asset_id=asset.id,
+                    version=version,
+                    reviewer=actor,
+                    review_status=AgentReviewStatus.APPROVED.value,
+                    review_note="发布上线前审核通过。",
+                    reviewed_at=datetime.now(UTC),
+                )
+            )
+        asset.reviewer = actor
+        asset.published_version = version
+        asset.status = AgentAssetStatus.ACTIVE.value
+        self.db.add(asset)
+        self.db.commit()
+        self.audit_service.log_action(
+            actor=actor,
+            action="publish_agent_asset",
+            resource_type=AgentAssetType.RULE.value,
+            resource_id=asset.id,
+            before_json=before,
+            after_json=self._asset_snapshot(asset),
+            request_id=request_id,
+        )
+        refreshed = self.repository.get(asset.id)
+        if refreshed is None:
+            raise LookupError("Asset not found")
+        return refreshed
+
+    def set_risk_rule_enabled(
+        self,
+        asset_id: str,
+        *,
+        enabled: bool,
+        actor: str,
+        request_id: str | None = None,
+    ) -> AgentAsset:
+        asset = self._resolve_asset(asset_id)
+        self._require_json_risk_asset(asset)
+        before = self._asset_snapshot(asset)
+        rule_library, file_name = self._resolve_json_risk_rule_document(asset)
+        manifest = self.rule_library_manager.read_rule_library_json(
+            library=rule_library,
+            file_name=file_name,
+        )
+        manifest["enabled"] = bool(enabled)
+        self.rule_library_manager.write_rule_library_json(
+            library=rule_library,
+            file_name=file_name,
+            payload=manifest,
+        )
+
+        config_json = dict(asset.config_json or {})
+        config_json["enabled"] = bool(enabled)
+        asset.config_json = config_json
+        updated = self.repository.save_asset(asset)
+        self.audit_service.log_action(
+            actor=actor,
+            action="set_risk_rule_enabled",
+            resource_type=AgentAssetType.RULE.value,
+            resource_id=asset.id,
+            before_json=before,
+            after_json=self._asset_snapshot(updated),
+            request_id=request_id,
+        )
+        return updated
+
+    def _load_risk_rule_for_test(
+        self, asset_id: str, version: str | None
+    ) -> tuple[AgentAsset, str, dict[str, Any]]:
+        asset = self._resolve_asset(asset_id)
+        self._require_json_risk_asset(asset)
+        target_version = self._resolve_target_version(asset, version)
+        if self.repository.get_version(asset.id, target_version) is None:
+            raise LookupError(f"版本 {target_version} 不存在")
+
+        rule_library, file_name = self._resolve_json_risk_rule_document(asset)
+        manifest = self.rule_library_manager.read_rule_library_json(
+            library=rule_library,
+            file_name=file_name,
+        )
+        return asset, target_version, manifest
+
+    def _create_test_run(
+        self,
+        asset: AgentAsset,
+        *,
+        version: str,
+        test_type: str,
+        passed: bool,
+        summary: str,
+        input_json: dict[str, Any],
+        result_json: dict[str, Any],
+        actor: str,
+        request_id: str | None,
+    ) -> AgentAssetRiskRuleTestRunRead:
+        status = "passed" if passed else "failed"
+        created = self.repository.create_test_run(
+            AgentAssetTestRun(
+                asset_id=asset.id,
+                version=version,
+                test_type=test_type,
+                status=status,
+                passed=passed,
+                summary=summary,
+                input_json=input_json,
+                result_json=result_json,
+                created_by=actor,
+            )
+        )
+        self.audit_service.log_action(
+            actor=actor,
+            action=f"risk_rule_test_{test_type}",
+            resource_type=AgentAssetType.RULE.value,
+            resource_id=asset.id,
+            before_json=None,
+            after_json={"version": version, "status": status, "summary": summary},
+            request_id=request_id,
+        )
+        return AgentAssetRiskRuleTestRunRead.model_validate(created)
+
+    def _run_sample_case(
+        self,
+        manifest: dict[str, Any],
+        case: AgentAssetRiskRuleSampleCase,
+    ) -> dict[str, Any]:
+        claim, contexts = self._build_synthetic_claim(case.values, manifest)
+        result = RiskRuleTemplateExecutor().evaluate(manifest, claim=claim, contexts=contexts)
+        actual_hit = result is not None
+        actual_severity = (
+            str((manifest.get("outcomes") or {}).get("fail", {}).get("severity") or "").strip()
+            if actual_hit
+            else "none"
+        )
+        expected_severity = str(case.expected_severity or "").strip()
+        severity_passed = (
+            not actual_hit or not expected_severity or expected_severity == actual_severity
+        )
+        passed = actual_hit == case.expected_hit and severity_passed
+        return {
+            "case_id": case.case_id or "",
+            "name": case.name,
+            "values": case.values,
+            "expected_hit": case.expected_hit,
+            "expected_severity": expected_severity,
+            "actual_hit": actual_hit,
+            "actual_severity": actual_severity,
+            "passed": passed,
+            "message": str(result.get("message") or "") if isinstance(result, dict) else "",
+            "evidence": result.get("evidence") if isinstance(result, dict) else {},
+        }
+
+    def _run_claim_scenario(self, manifest: dict[str, Any], claim: ExpenseClaim) -> dict[str, Any]:
+        contexts = ExpenseClaimService(self.db)._build_claim_attachment_contexts(claim)
+        result = RiskRuleTemplateExecutor().evaluate(manifest, claim=claim, contexts=contexts)
+        hit = result is not None
+        return {
+            "claim_id": claim.id,
+            "claim_no": claim.claim_no,
+            "employee_name": claim.employee_name,
+            "department_name": claim.department_name,
+            "expense_type": claim.expense_type,
+            "amount": float(claim.amount or 0),
+            "status": claim.status,
+            "occurred_at": claim.occurred_at.isoformat() if claim.occurred_at else "",
+            "hit": hit,
+            "severity": str((manifest.get("outcomes") or {}).get("fail", {}).get("severity") or "")
+            if hit
+            else "none",
+            "message": str(result.get("message") or "") if isinstance(result, dict) else "",
+            "evidence": result.get("evidence") if isinstance(result, dict) else {},
+        }
+
+    def _build_synthetic_claim(
+        self,
+        values: dict[str, Any],
+        manifest: dict[str, Any],
+    ) -> tuple[ExpenseClaim, list[dict[str, Any]]]:
+        claim = ExpenseClaim(
+            claim_no="TEST-RISK-RULE",
+            employee_name=str(values.get("claim.employee_name") or "测试员工"),
+            department_name=str(values.get("claim.department_name") or "测试部门"),
+            expense_type=str(values.get("item.item_type") or "差旅费"),
+            reason=str(values.get("claim.reason") or "测试报销事由"),
+            location=str(values.get("claim.location") or "北京"),
+            amount=self._to_decimal(values.get("claim.amount")),
+            currency="CNY",
+            invoice_count=1,
+            occurred_at=datetime.now(UTC),
+            status="draft",
+        )
+        item = ExpenseClaimItem(
+            item_date=date.today(),
+            item_type=str(values.get("item.item_type") or "住宿费"),
+            item_reason=str(values.get("item.item_reason") or claim.reason),
+            item_location=str(values.get("item.item_location") or claim.location),
+            item_amount=self._to_decimal(values.get("item.item_amount") or claim.amount),
+        )
+        claim.items = [item]
+
+        attachment_fields = []
+        document_info: dict[str, Any] = {"fields": attachment_fields}
+        for field in self._extract_manifest_fields(manifest):
+            key = field["key"]
+            if key not in values:
+                continue
+            value = self._coerce_sample_value(key, values.get(key))
+            if key.startswith("claim."):
+                setattr(claim, key.removeprefix("claim."), value)
+            elif key.startswith("item."):
+                setattr(item, key.removeprefix("item."), value)
+            elif key.startswith("attachment."):
+                short_key = key.removeprefix("attachment.")
+                document_info[short_key] = value
+                attachment_fields.append(
+                    {"key": short_key, "label": field["label"], "value": value}
+                )
+        return claim, [
+            {
+                "document_info": document_info,
+                "ocr_text": document_info.get("ocr_text", ""),
+            }
+        ]
+
+    def _build_default_sample_cases(
+        self,
+        manifest: dict[str, Any],
+    ) -> list[AgentAssetRiskRuleSampleCase]:
+        fields = self._extract_manifest_fields(manifest)
+        severity = str((manifest.get("outcomes") or {}).get("fail", {}).get("severity") or "")
+        template_key = str(manifest.get("template_key") or "").strip()
+        hit_values = self._find_case_values_for_expected(manifest, fields, expected_hit=True)
+        pass_values = self._find_case_values_for_expected(manifest, fields, expected_hit=False)
+        cases = [
+            AgentAssetRiskRuleSampleCase(
+                case_id="hit",
+                name="应该命中风险",
+                values=hit_values,
+                expected_hit=True,
+                expected_severity=severity,
+                note="验证规则能识别异常样本。",
+            ),
+            AgentAssetRiskRuleSampleCase(
+                case_id="pass",
+                name="应该不命中",
+                values=pass_values,
+                expected_hit=False,
+                expected_severity="none",
+                note="验证正常样本不会误触发。",
+            ),
+        ]
+        if template_key == "field_required_v1":
+            cases.append(
+                AgentAssetRiskRuleSampleCase(
+                    case_id="missing",
+                    name="关键字段缺失",
+                    values={key: "" for key in hit_values},
+                    expected_hit=True,
+                    expected_severity=severity,
+                    note="验证缺字段时会进入复核。",
+                )
+            )
+        return cases
+
+    def _find_case_values_for_expected(
+        self,
+        manifest: dict[str, Any],
+        fields: list[dict[str, str]],
+        *,
+        expected_hit: bool,
+    ) -> dict[str, Any]:
+        candidates = [
+            self._build_case_values(manifest, fields, hit=expected_hit),
+            {field["key"]: self._default_value_for_field(field["key"]) for field in fields},
+            {
+                field["key"]: ("上海" if index == 0 else "北京")
+                for index, field in enumerate(fields)
+            },
+            {field["key"]: "北京" for field in fields},
+            {field["key"]: "" for field in fields},
+        ]
+        severity = str((manifest.get("outcomes") or {}).get("fail", {}).get("severity") or "")
+        for values in candidates:
+            probe = AgentAssetRiskRuleSampleCase(
+                name="默认样例探测",
+                values=values,
+                expected_hit=expected_hit,
+                expected_severity=severity if expected_hit else "none",
+            )
+            result = self._run_sample_case(manifest, probe)
+            if bool(result["actual_hit"]) == expected_hit:
+                return values
+        return candidates[0]
+
+    def _build_case_values(
+        self,
+        manifest: dict[str, Any],
+        fields: list[dict[str, str]],
+        *,
+        hit: bool,
+    ) -> dict[str, Any]:
+        values = {field["key"]: self._default_value_for_field(field["key"]) for field in fields}
+        template_key = str(manifest.get("template_key") or "").strip()
+        params = manifest.get("params") if isinstance(manifest.get("params"), dict) else {}
+        if template_key == "field_compare_v1":
+            condition = next(
+                (item for item in params.get("conditions", []) if isinstance(item, dict)),
+                {},
+            )
+            left = str(condition.get("left") or "").strip()
+            right = str(condition.get("right") or "").strip()
+            operator = str(condition.get("operator") or "overlap").strip()
+            if left and operator == "is_empty":
+                values[left] = "测试值" if hit else ""
+            elif left and right and operator in {"not_equals", "not_in", "not_overlap"}:
+                values[left] = "北京" if hit else "上海"
+                values[right] = "北京"
+            elif left and right:
+                values[left] = "上海" if hit else "北京"
+                values[right] = "北京"
+        elif template_key == "field_required_v1" and hit and fields:
+            values[fields[0]["key"]] = ""
+        elif template_key == "keyword_match_v1":
+            keywords = params.get("keywords") if isinstance(params.get("keywords"), list) else []
+            keyword = str(next(iter(keywords), "咨询费") or "咨询费")
+            target_key = fields[0]["key"] if fields else "claim.reason"
+            values[target_key] = f"本次报销包含{keyword}" if hit else "正常差旅报销"
+        return values
+
+    @staticmethod
+    def _default_value_for_field(field_key: str) -> Any:
+        if field_key.endswith("amount"):
+            return "100.00"
+        if field_key.endswith("issue_date"):
+            return date.today().isoformat()
+        if field_key.endswith("route_cities"):
+            return ["北京"]
+        if field_key.endswith("ocr_text"):
+            return "正常发票内容"
+        if "city" in field_key or "location" in field_key:
+            return "北京"
+        if field_key.endswith("item_type"):
+            return "住宿费"
+        return "测试值"
+
+    def _query_expense_claim_samples(self, parsed_scope: dict[str, Any]) -> list[ExpenseClaim]:
+        days = int(parsed_scope.get("days") or 30)
+        limit = min(max(int(parsed_scope.get("limit") or 50), 1), 200)
+        since = datetime.now(UTC) - timedelta(days=days)
+        stmt = select(ExpenseClaim).where(ExpenseClaim.created_at >= since)
+
+        expense_keyword = str(parsed_scope.get("expense_keyword") or "").strip()
+        if expense_keyword:
+            like_keyword = f"%{expense_keyword}%"
+            stmt = stmt.where(
+                or_(
+                    ExpenseClaim.expense_type.ilike(like_keyword),
+                    ExpenseClaim.reason.ilike(like_keyword),
+                )
+            )
+
+        cities = [str(item or "").strip() for item in parsed_scope.get("cities", []) if item]
+        if cities:
+            city_filters = []
+            for city in cities[:8]:
+                like_city = f"%{city}%"
+                city_filters.extend(
+                    [
+                        ExpenseClaim.location.ilike(like_city),
+                        ExpenseClaim.reason.ilike(like_city),
+                    ]
+                )
+            stmt = stmt.where(or_(*city_filters))
+
+        stmt = stmt.order_by(ExpenseClaim.created_at.desc()).limit(limit)
+        return list(self.db.scalars(stmt).all())
+
+    @staticmethod
+    def _parse_scenario_scope(intent: str, filters: dict[str, Any]) -> dict[str, Any]:
+        text = str(intent or "")
+        raw_days = filters.get("days") or filters.get("recent_days")
+        days = int(raw_days) if str(raw_days or "").isdigit() else 30
+        match = re.search(r"最近\s*(\d{1,3})\s*天", text)
+        if match:
+            days = int(match.group(1))
+        limit = filters.get("limit") if str(filters.get("limit") or "").isdigit() else 50
+        expense_keyword = str(filters.get("expense_keyword") or "").strip()
+        if not expense_keyword and any(keyword in text for keyword in ("酒店", "住宿")):
+            expense_keyword = "住宿"
+
+        city_candidates = ("北京", "上海", "广州", "深圳", "武汉", "杭州", "成都", "南京")
+        cities = [
+            city
+            for city in city_candidates
+            if city in text or city in [str(item) for item in filters.get("cities", []) or []]
+        ]
+        return {
+            "business_domain": "expense",
+            "days": max(1, min(days, 365)),
+            "limit": max(1, min(int(limit), 200)),
+            "expense_keyword": expense_keyword,
+            "cities": cities,
+            "execution_mode": "dry_run",
+        }
+
+    @staticmethod
+    def _extract_manifest_fields(manifest: dict[str, Any]) -> list[dict[str, str]]:
+        inputs = manifest.get("inputs") if isinstance(manifest.get("inputs"), dict) else {}
+        fields = inputs.get("fields") if isinstance(inputs.get("fields"), list) else []
+        normalized = []
+        for item in fields:
+            if not isinstance(item, dict):
+                continue
+            key = str(item.get("key") or "").strip()
+            if key:
+                normalized.append({"key": key, "label": str(item.get("label") or key).strip()})
+        return normalized
+
+    @staticmethod
+    def _coerce_sample_value(field_key: str, value: Any) -> Any:
+        if field_key.endswith("route_cities") and isinstance(value, str):
+            return [item.strip() for item in re.split(r"[,，、/ ]+", value) if item.strip()]
+        return value
+
+    @staticmethod
+    def _to_decimal(value: Any) -> Decimal:
+        try:
+            return Decimal(str(value or "0"))
+        except (InvalidOperation, ValueError):
+            return Decimal("0")
+
+    def _resolve_asset(self, asset_or_id: AgentAsset | str) -> AgentAsset:
+        if isinstance(asset_or_id, AgentAsset):
+            return asset_or_id
+        asset = self.repository.get(str(asset_or_id))
+        if asset is None:
+            raise LookupError("Asset not found")
+        return asset
+
+    @staticmethod
+    def _require_json_risk_asset(asset: AgentAsset) -> None:
+        config_json = asset.config_json if isinstance(asset.config_json, dict) else {}
+        if asset.asset_type != AgentAssetType.RULE.value:
+            raise ValueError("仅规则资产支持风险规则操作。")
+        if str(config_json.get("detail_mode") or "").strip().lower() != "json_risk":
+            raise ValueError("仅 JSON 风险规则支持该操作。")
+
+    def _resolve_target_version(self, asset: AgentAsset, version: str | None) -> str:
+        target = str(version or self._resolve_working_version(asset) or "").strip()
+        if not target:
+            raise ValueError("当前规则尚未配置工作版本。")
+        return target
+
+    def _delete_risk_rule_json_file(self, asset: AgentAsset) -> None:
+        try:
+            rule_library, file_name = self._resolve_json_risk_rule_document(asset)
+            target = self.rule_library_manager.resolve_rule_library_path(
+                library=rule_library,
+                file_name=file_name,
+            )
+            target.unlink(missing_ok=True)
+        except (FileNotFoundError, ValueError):
+            return
+
+    @staticmethod
+    def _serialize_test_run(
+        run: AgentAssetTestRun | None,
+    ) -> AgentAssetRiskRuleTestRunRead | None:
+        return AgentAssetRiskRuleTestRunRead.model_validate(run) if run is not None else None
--- a/server/src/app/services/agent_assets.py
+++ b/server/src/app/services/agent_assets.py
@@ -4,6 +4,7 @@ import json
 from collections import defaultdict
 from datetime import UTC, datetime
 from typing import Any
+
 from sqlalchemy.orm import Session

 from app.core.agent_enums import (
@@ -27,13 +28,14 @@ from app.schemas.agent_asset import (
 )
 from app.services.agent_asset_json_rules import AgentAssetJsonRuleMixin
 from app.services.agent_asset_onlyoffice import AgentAssetOnlyOfficeMixin
+from app.services.agent_asset_risk_rule_simulation import AgentAssetRiskRuleSimulationMixin
+from app.services.agent_asset_risk_rule_testing import AgentAssetRiskRuleTestingMixin
 from app.services.agent_asset_rule_library import AgentAssetRuleLibraryManager
+from app.services.agent_asset_spreadsheet import AgentAssetSpreadsheetManager
 from app.services.agent_asset_spreadsheet_helpers import AgentAssetSpreadsheetHelperMixin
 from app.services.agent_asset_timeline import AgentAssetTimelineMixin
-from app.services.agent_asset_spreadsheet import AgentAssetSpreadsheetManager
 from app.services.agent_foundation import AgentFoundationService
 from app.services.audit import AuditLogService
-from app.services.settings import resolve_onlyoffice_settings

 logger = get_logger("app.services.agent_assets")

@@ -41,6 +43,8 @@ logger = get_logger("app.services.agent_assets")
 class AgentAssetService(
    AgentAssetOnlyOfficeMixin,
    AgentAssetSpreadsheetHelperMixin,
+    AgentAssetRiskRuleTestingMixin,
+    AgentAssetRiskRuleSimulationMixin,
    AgentAssetTimelineMixin,
    AgentAssetJsonRuleMixin,
 ):
@@ -66,10 +70,7 @@ class AgentAssetService(
            asset_type=asset_type, status=status, domain=domain, keyword=keyword
        )
        version_stats = self._collect_version_stats(assets)
-        return [
-            self._serialize_list_item(asset, version_stats.get(asset.id))
-            for asset in assets
-        ]
+        return [self._serialize_list_item(asset, version_stats.get(asset.id)) for asset in assets]

    def get_asset(self, asset_id: str) -> AgentAssetRead | None:
        self._ensure_ready()
@@ -88,9 +89,7 @@ class AgentAssetService(
            else next(iter(self.repository.list_reviews(asset_id, limit=1)), None)
        )
        current_version = (
-            self.repository.get_version(asset_id, working_version)
-            if working_version
-            else None
+            self.repository.get_version(asset_id, working_version) if working_version else None
        )
        version_stats = self._collect_version_stats([asset]).get(asset.id)
        return AgentAssetRead(
@@ -100,12 +99,14 @@ class AgentAssetService(
            else None,
            current_version_content_type=current_version.content_type if current_version else None,
            current_version_change_note=current_version.change_note if current_version else None,
-            recent_versions=[
-                self._serialize_version(item, asset) for item in recent_versions
-            ],
+            recent_versions=[self._serialize_version(item, asset) for item in recent_versions],
            latest_review=AgentAssetReviewRead.model_validate(latest_review)
            if latest_review
            else None,
+            latest_test_summary=self.get_latest_risk_rule_test_summary(asset)
+            if str((asset.config_json or {}).get("detail_mode") or "").strip().lower()
+            == "json_risk"
+            else None,
        )

    def create_asset(
@@ -301,6 +302,13 @@ class AgentAssetService(
        if self.repository.get_version(asset_id, payload.version) is None:
            raise LookupError(f"版本 {payload.version} 不存在")
        if asset.asset_type == AgentAssetType.RULE.value:
+            if (
+                str((asset.config_json or {}).get("detail_mode") or "").strip().lower()
+                == "json_risk"
+                and payload.review_status == AgentReviewStatus.PENDING
+                and not self.get_latest_risk_rule_test_summary(asset).test_passed
+            ):
+                raise PermissionError("当前规则版本尚未完成测试通过确认，不能提交审核。")
            working_version = self._resolve_working_version(asset)
            if payload.version != working_version:
                raise ValueError("只能对当前工作版本发起审核。")
@@ -594,11 +602,10 @@ class AgentAssetService(
            ),
        )

-    def _collect_version_stats(
-        self, assets: list[AgentAsset]
-    ) -> dict[str, dict[str, int | str | None]]:
+    def _collect_version_stats(self, assets: list[AgentAsset]) -> dict[str, dict[str, Any]]:
        asset_ids = [item.id for item in assets]
        versions = self.repository.list_versions_for_assets(asset_ids)
+        reviews = self.repository.list_reviews_for_assets(asset_ids)
        spreadsheet_logs = self.audit_service.repository.list_for_resources(
            resource_type=AgentAssetType.RULE.value,
            resource_ids=[
@@ -610,23 +617,33 @@ class AgentAssetService(
            ],
            action="edit_rule_spreadsheet",
        )
-        working_versions = {
-            item.id: self._resolve_working_version(item) for item in assets
-        }
+        working_versions = {item.id: self._resolve_working_version(item) for item in assets}
        version_counts: dict[str, int] = defaultdict(int)
        modified_by: dict[str, str | None] = {item.id: None for item in assets}
+        published_versions = {item.id: self._resolve_published_version(item) for item in assets}
+        published_by: dict[str, str | None] = {}
+        published_at: dict[str, datetime | None] = {}
        spreadsheet_edit_counts: dict[str, int] = defaultdict(int)
        spreadsheet_last_actor: dict[str, str | None] = {}
        spreadsheet_last_changed_at: dict[str, datetime] = {}

        for version in versions:
            version_counts[version.asset_id] += 1
-            if (
-                modified_by.get(version.asset_id) is None
-                and version.version == working_versions.get(version.asset_id)
-            ):
+            if modified_by.get(
+                version.asset_id
+            ) is None and version.version == working_versions.get(version.asset_id):
                modified_by[version.asset_id] = version.created_by

+        for review in reviews:
+            if review.asset_id in published_at:
+                continue
+            if review.version != published_versions.get(review.asset_id):
+                continue
+            if review.review_status != AgentReviewStatus.APPROVED.value:
+                continue
+            published_by[review.asset_id] = review.reviewer
+            published_at[review.asset_id] = review.reviewed_at or review.created_at
+
        for log in spreadsheet_logs:
            spreadsheet_edit_counts[log.resource_id] += 1
            last_changed_at = spreadsheet_last_changed_at.get(log.resource_id)
@@ -652,6 +669,8 @@ class AgentAssetService(
                    and spreadsheet_last_actor.get(item.id)
                    else modified_by.get(item.id)
                ),
+                "published_by": published_by.get(item.id),
+                "published_at": published_at.get(item.id),
            }
            for item in assets
        }
@@ -663,9 +682,11 @@ class AgentAssetService(
    ) -> AgentAssetListItem:
        payload = AgentAssetListItem.model_validate(asset).model_dump()
        payload["change_count"] = int((version_stats or {}).get("change_count") or 0)
-        payload["modified_by"] = (
-            str((version_stats or {}).get("modified_by") or "").strip() or None
+        payload["modified_by"] = str((version_stats or {}).get("modified_by") or "").strip() or None
+        payload["published_by"] = (
+            str((version_stats or {}).get("published_by") or "").strip() or None
        )
+        payload["published_at"] = (version_stats or {}).get("published_at")
        return AgentAssetListItem.model_validate(payload)

    @staticmethod
--- a/server/src/app/services/agent_foundation.py
+++ b/server/src/app/services/agent_foundation.py
@@ -2,7 +2,7 @@ from __future__ import annotations

 import threading

-from sqlalchemy import select
+from sqlalchemy import inspect, select, text
 from sqlalchemy.orm import Session

 from app.core.config import get_settings
@@ -75,6 +75,7 @@ class AgentFoundationService(
        try:
            Base.metadata.create_all(bind=self.db.get_bind())
            self._ensure_agent_asset_schema()
+            self._ensure_financial_record_schema()
            self._seed_agent_assets()
            self._sync_demo_financial_records()
            self._seed_runs_and_logs()
@@ -88,6 +89,36 @@ class AgentFoundationService(
        bind = self.db.get_bind()
        return str(getattr(bind, "url", "") or id(bind))

+    def _ensure_financial_record_schema(self) -> None:
+        bind = self.db.get_bind()
+        inspector = inspect(bind)
+        if "expense_claims" not in inspector.get_table_names():
+            return
+
+        column_names = {column["name"] for column in inspector.get_columns("expense_claims")}
+        dialect_name = bind.dialect.name
+        timestamp_type = "TIMESTAMP WITH TIME ZONE" if dialect_name == "postgresql" else "DATETIME"
+        boolean_default = "FALSE" if dialect_name == "postgresql" else "0"
+
+        if "hermes_scanned_at" not in column_names:
+            self.db.execute(
+                text(f"ALTER TABLE expense_claims ADD COLUMN hermes_scanned_at {timestamp_type}")
+            )
+        if "hermes_risk_flag" not in column_names:
+            self.db.execute(
+                text(
+                    "ALTER TABLE expense_claims "
+                    f"ADD COLUMN hermes_risk_flag BOOLEAN DEFAULT {boolean_default} NOT NULL"
+                )
+            )
+            self.db.execute(
+                text(
+                    "CREATE INDEX IF NOT EXISTS ix_expense_claims_hermes_risk_flag "
+                    "ON expense_claims (hermes_risk_flag)"
+                )
+            )
+        self.db.flush()
+
    def _sync_demo_financial_records(self) -> None:
        if get_settings().seed_demo_financial_records:
            self._seed_financial_records()
--- a/server/src/app/services/employee.py
+++ b/server/src/app/services/employee.py
@@ -651,7 +651,11 @@ class EmployeeService:
        column_names = {column["name"] for column in inspector.get_columns("employees")}
        if "password_hash" not in column_names:
            self.db.execute(text("ALTER TABLE employees ADD COLUMN password_hash VARCHAR(255)"))
-            self.db.flush()
+        if "compliance_score" not in column_names:
+            self.db.execute(
+                text("ALTER TABLE employees ADD COLUMN compliance_score INTEGER DEFAULT 100 NOT NULL")
+            )
+        self.db.flush()

    def _seed_employee_history(self, employee: Employee, definition: dict[str, Any]) -> None:
        existing_keys = {
--- a/server/src/app/services/expense_type_keywords.py
+++ b/server/src/app/services/expense_type_keywords.py
@@ -141,6 +141,10 @@ EXPENSE_TYPE_KEYWORD_GROUPS: tuple[tuple[str, str, tuple[str, ...]], ...] = (
            "办公用品",
            "办公耗材",
            "办公设备",
+            "采购",
+            "集中采购",
+            "物资采购",
+            "办公采购",
            "办公",
            "文具",
            "耗材",
--- a/server/src/app/services/hermes_expense_report.py
+++ b/server/src/app/services/hermes_expense_report.py
@@ -0,0 +1,104 @@
+from __future__ import annotations
+
+import json
+from datetime import datetime, timedelta, timezone
+from typing import Any
+
+from sqlalchemy import func, select
+from sqlalchemy.orm import Session
+
+from app.core.logging import get_logger
+from app.models.financial_record import ExpenseClaim
+from app.services.runtime_chat import RuntimeChatService
+
+logger = get_logger("app.services.hermes_expense_report")
+
+
+class HermesExpenseReportService:
+    def __init__(self, db: Session) -> None:
+        self.db = db
+        self.chat_service = RuntimeChatService(db)
+
+    def generate_weekly_report(self, log_id: str | None = None) -> None:
+        logger.info("Starting Hermes weekly expense report generation...")
+        
+        # 1. 聚合数据
+        aggregated_data = self._aggregate_recent_expenses(days=7)
+        if not aggregated_data.get("total_amount"):
+            logger.info("No expense data in the last 7 days. Skipping report.")
+            return
+
+        # 2. 传入大模型分析
+        report_markdown = self._generate_insights_with_llm(aggregated_data)
+        
+        if not report_markdown:
+            logger.warning("Failed to generate expense report from LLM.")
+            return
+
+        # 3. 模拟发送报告
+        self._deliver_report(report_markdown, log_id)
+        logger.info("Hermes weekly expense report generation completed.")
+
+    def _aggregate_recent_expenses(self, days: int = 7) -> dict[str, Any]:
+        target_date = datetime.now(timezone.utc) - timedelta(days=days)
+        
+        # 基础过滤：最近N天且不是驳回状态的单据
+        base_filter = [
+            ExpenseClaim.occurred_at >= target_date,
+            ExpenseClaim.status != "rejected"
+        ]
+
+        # 1. 按部门汇总
+        dept_stmt = select(
+            ExpenseClaim.department_name, 
+            func.sum(ExpenseClaim.amount).label("total")
+        ).where(*base_filter).group_by(ExpenseClaim.department_name)
+        
+        dept_results = self.db.execute(dept_stmt).all()
+        by_department = {row.department_name or "Unknown": float(row.total or 0) for row in dept_results}
+
+        # 2. 按类目汇总
+        type_stmt = select(
+            ExpenseClaim.expense_type, 
+            func.sum(ExpenseClaim.amount).label("total")
+        ).where(*base_filter).group_by(ExpenseClaim.expense_type)
+        
+        type_results = self.db.execute(type_stmt).all()
+        by_expense_type = {row.expense_type or "Unknown": float(row.total or 0) for row in type_results}
+
+        # 3. 总花费
+        total_amount = sum(by_department.values())
+
+        return {
+            "period": f"Last {days} days",
+            "total_amount": total_amount,
+            "by_department": by_department,
+            "by_expense_type": by_expense_type
+        }
+
+    def _generate_insights_with_llm(self, data: dict[str, Any]) -> str | None:
+        system_prompt = (
+            "你是公司的财务分析专家。请根据提供的最近期业务开销数据，撰写一份简洁有力的【高管费控洞察周报】。\n"
+            "要求：\n"
+            "1. 不要机械地罗列数字，要像人一样指出异常（例如：哪个部门花钱最多？打车费是不是异常高？）。\n"
+            "2. 给出 1 条削减成本的实操建议。\n"
+            "3. 纯 Markdown 格式输出，不超过 300 字。"
+        )
+
+        messages = [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": f"开销统计数据：\n{json.dumps(data, ensure_ascii=False, indent=2)}"}
+        ]
+
+        response = self.chat_service.complete(
+            messages,
+            max_tokens=800,
+            temperature=0.4
+        )
+        return response
+
+    def _deliver_report(self, report_markdown: str, log_id: str | None) -> None:
+        # TODO: 未来在这里接入企微/钉钉机器人或邮件发送接口
+        logger.info(f"\n================ Hermes Weekly Report [LogID: {log_id}] ================\n"
+                    f"{report_markdown}\n"
+                    f"==========================================================================")
--- a/server/src/app/services/hermes_risk_scanner.py
+++ b/server/src/app/services/hermes_risk_scanner.py
@@ -0,0 +1,135 @@
+from __future__ import annotations
+
+import json
+from datetime import datetime, timezone
+from typing import Any
+
+from sqlalchemy import or_, select
+from sqlalchemy.orm import Session
+
+from app.core.logging import get_logger
+from app.models.financial_record import ExpenseClaim
+from app.models.hermes_config import HermesTaskExecutionLog
+from app.models.hermes_report import HermesRiskReport
+from app.services.runtime_chat import RuntimeChatService
+
+logger = get_logger("app.services.hermes_risk_scanner")
+
+class HermesRiskScannerService:
+    def __init__(self, db: Session) -> None:
+        self.db = db
+        self.chat_service = RuntimeChatService(db)
+
+    def scan_global_risks(self, log_id: str | None = None) -> None:
+        logger.info("Starting global risk scan for Hermes...")
+        
+        # 1. Fetch unscanned claims
+        claims = self._fetch_unscanned_claims()
+        if not claims:
+            logger.info("No unscanned claims found. Aborting scan.")
+            return
+
+        logger.info(f"Fetched {len(claims)} claims to analyze.")
+        
+        # 2. Extract context for LLM
+        claims_context = []
+        for c in claims:
+            claims_context.append({
+                "claim_id": c.id,
+                "claim_no": c.claim_no,
+                "employee_name": c.employee_name,
+                "department_name": c.department_name,
+                "expense_type": c.expense_type,
+                "location": c.location,
+                "amount": float(c.amount),
+                "occurred_at": str(c.occurred_at) if c.occurred_at else None,
+                "reason": c.reason,
+            })
+            
+        # 3. Analyze with LLM
+        risk_results = self._analyze_claims_with_llm(claims_context)
+        
+        # 4. Process and persist results
+        detected_risk_count = 0
+        if risk_results:
+            for risk in risk_results:
+                claim_ids = risk.get("claim_ids", [])
+                if not claim_ids:
+                    continue
+                    
+                detected_risk_count += 1
+                for cid in claim_ids:
+                    report = HermesRiskReport(
+                        claim_id=cid,
+                        execution_log_id=log_id,
+                        risk_level=risk.get("risk_level", "medium"),
+                        risk_type=risk.get("risk_type", "unknown"),
+                        risk_description=risk.get("description", "No description provided"),
+                        related_claim_ids=claim_ids,
+                    )
+                    self.db.add(report)
+                    
+                    # Update claim flags
+                    claim_obj = next((c for c in claims if c.id == cid), None)
+                    if claim_obj:
+                        claim_obj.hermes_risk_flag = True
+
+        # 5. Mark all as scanned
+        now = datetime.now(timezone.utc)
+        for c in claims:
+            c.hermes_scanned_at = now
+            
+        self.db.commit()
+        logger.info(f"Hermes risk scan completed. Found {detected_risk_count} risks.")
+        
+    def _fetch_unscanned_claims(self) -> list[ExpenseClaim]:
+        stmt = select(ExpenseClaim).where(
+            ExpenseClaim.status.in_(["draft", "submitted", "review"]),
+            or_(
+                ExpenseClaim.hermes_scanned_at.is_(None),
+                ExpenseClaim.hermes_risk_flag.is_(False) # only rescan if it has no flags yet
+            )
+        ).limit(50) # Batch size to prevent Token overflow
+        
+        return list(self.db.scalars(stmt).all())
+        
+    def _analyze_claims_with_llm(self, claims_context: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        system_prompt = (
+            "你是 X-Financial 的 Hermes 内控审计智能体。请分析以下近期的报销单数据集合，寻找以下潜在风险：\n"
+            "1. 拆单行为 (split_billing)：同一人在相邻日期针对同一类目/商户提交多笔恰好贴近免审额度的小额单据。\n"
+            "2. 群体合谋 (collusion)：不同部门的员工在同一天去同一家非标准酒店类偏僻商户高额消费。\n"
+            "3. 异常频次 (frequency_anomaly)：某员工在短时间内的打车或招待频次极度不合理。\n"
+            "请严格以 JSON 数组格式返回结果，如果没有风险返回空数组 `[]`。\n"
+            "JSON 格式要求：\n"
+            "[\n"
+            "  {\n"
+            '    "risk_type": "split_billing",\n'
+            '    "risk_level": "high",\n'
+            '    "claim_ids": ["uuid-1", "uuid-2"],\n'
+            '    "description": "详细推理过程，为什么判定为拆单。"\n'
+            "  }\n"
+            "]\n"
+        )
+        
+        messages = [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": json.dumps(claims_context, ensure_ascii=False, indent=2)}
+        ]
+        
+        response_text = self.chat_service.complete(
+            messages,
+            max_tokens=1500,
+            temperature=0.1
+        )
+        
+        if not response_text:
+            logger.warning("LLM returned empty response for risk scan.")
+            return []
+            
+        # Clean markdown formatting if present
+        cleaned_text = response_text.replace("```json", "").replace("```", "").strip()
+        try:
+            return json.loads(cleaned_text)
+        except json.JSONDecodeError as e:
+            logger.error(f"Failed to parse LLM risk scan response as JSON: {e}\nResponse: {response_text}")
+            return []
--- a/server/src/app/services/hermes_scheduler.py
+++ b/server/src/app/services/hermes_scheduler.py
@@ -0,0 +1,131 @@
+import logging
+import threading
+import time
+from datetime import datetime, timezone
+import traceback
+
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from app.core.logging import get_logger
+from app.db.session import get_session_factory
+from app.models.hermes_config import HermesTaskConfig, HermesTaskExecutionLog
+from app.services.hermes_risk_scanner import HermesRiskScannerService
+from app.services.hermes_expense_report import HermesExpenseReportService
+
+logger = get_logger("app.services.hermes_scheduler")
+
+
+class HermesScheduler:
+    def __init__(self) -> None:
+        self._stop_event = threading.Event()
+        self._thread: threading.Thread | None = None
+        self._lock = threading.Lock()
+        self.session_factory = get_session_factory()
+
+    def start(self) -> None:
+        with self._lock:
+            if self._thread is not None and self._thread.is_alive():
+                return
+            self._stop_event.clear()
+            self._thread = threading.Thread(
+                target=self._run_loop,
+                name="hermes-agent-scheduler",
+                daemon=True,
+            )
+            self._thread.start()
+            logger.info("Hermes Agent Scheduler started.")
+
+    def shutdown(self) -> None:
+        with self._lock:
+            thread = self._thread
+            self._thread = None
+        self._stop_event.set()
+        if thread is not None and thread.is_alive():
+            thread.join(timeout=3)
+        logger.info("Hermes Agent Scheduler stopped.")
+
+    def _run_loop(self) -> None:
+        logger.info("Hermes background loop is now active. Polling interval: 60s.")
+        while not self._stop_event.is_set():
+            try:
+                self._check_and_run_tasks()
+            except Exception as e:
+                logger.error(f"Error in Hermes run loop: {e}", exc_info=True)
+            
+            # 睡眠一分钟，每分钟轮询一次
+            if self._stop_event.wait(60.0):
+                break
+
+    def _check_and_run_tasks(self) -> None:
+        db = self.session_factory()
+        try:
+            # 获取所有启用的任务配置
+            stmt = select(HermesTaskConfig).where(HermesTaskConfig.is_enabled == True)
+            configs = db.scalars(stmt).all()
+            
+            for config in configs:
+                if self._should_run_now(db, config):
+                    self._execute_task(db, config)
+                    
+        finally:
+            db.close()
+            
+    def _should_run_now(self, db: Session, config: HermesTaskConfig) -> bool:
+        # 简单策略：检查是否在过去24小时内运行过。
+        # 如果没有 croniter 库，我们暂时采用按天执行的简化逻辑
+        stmt = select(HermesTaskExecutionLog).where(
+            HermesTaskExecutionLog.config_id == config.id,
+            HermesTaskExecutionLog.status.in_(["success", "running"])
+        ).order_by(HermesTaskExecutionLog.started_at.desc()).limit(1)
+        
+        last_log = db.scalars(stmt).first()
+        
+        if not last_log:
+            return True # 从未执行过，立即执行
+            
+        now = datetime.now(timezone.utc)
+        elapsed_hours = (now - last_log.started_at).total_seconds() / 3600
+        
+        # 简化：只要距离上次成功执行超过了 23.5 小时，就认为该跑了（模拟每天跑一次）
+        if elapsed_hours >= 23.5:
+            return True
+            
+        return False
+
+    def _execute_task(self, db: Session, config: HermesTaskConfig) -> None:
+        logger.info(f"Triggering Hermes task: {config.task_type} (Config ID: {config.id})")
+        
+        # 创建执行日志，标记为 running
+        log_record = HermesTaskExecutionLog(
+            config_id=config.id,
+            status="running"
+        )
+        db.add(log_record)
+        db.commit()
+        db.refresh(log_record)
+        
+        try:
+            if config.task_type == "global_risk_scan":
+                scanner = HermesRiskScannerService(db)
+                scanner.scan_global_risks(log_id=log_record.id)
+            elif config.task_type == "weekly_expense_report":
+                reporter = HermesExpenseReportService(db)
+                reporter.generate_weekly_report(log_id=log_record.id)
+                
+            log_record.status = "success"
+            log_record.completed_at = datetime.now(timezone.utc)
+            log_record.result_summary = "Task executed successfully."
+            
+        except Exception as e:
+            logger.error(f"Failed to execute Hermes task {config.task_type}: {e}")
+            log_record.status = "failed"
+            log_record.completed_at = datetime.now(timezone.utc)
+            log_record.error_trace = traceback.format_exc()
+            
+        finally:
+            db.commit()
+
+
+# 全局单例
+hermes_scheduler = HermesScheduler()
--- a/server/src/app/services/knowledge_document_extractors.py
+++ b/server/src/app/services/knowledge_document_extractors.py
@@ -34,10 +34,104 @@ def _extract_docx_text(file_path: Path) -> str:
        return "当前 Word 文件解析失败。"

    root = ElementTree.fromstring(xml_content)
+    body = next((node for node in root.iter() if node.tag.endswith("}body")), root)
+    blocks: list[str] = []
+
+    for child in body:
+        if child.tag.endswith("}p"):
+            paragraph = _extract_docx_paragraph_text(child)
+            if paragraph:
+                blocks.append(paragraph)
+            continue
+        if child.tag.endswith("}tbl"):
+            table = _extract_docx_table_rows(child)
+            rendered = _build_docx_table_markdown(table)
+            if rendered:
+                blocks.append(rendered)
+
+    if blocks:
+        return "\n\n".join(blocks)
+
    texts = [node.text.strip() for node in root.iter() if node.tag.endswith("}t") and node.text]
    return "\n".join(texts)


+def _extract_docx_paragraph_text(node: ElementTree.Element) -> str:
+    parts: list[str] = []
+    for child in node.iter():
+        if child.tag.endswith("}t") and child.text:
+            parts.append(child.text)
+        elif child.tag.endswith("}tab"):
+            parts.append("\t")
+        elif child.tag.endswith("}br"):
+            parts.append("\n")
+    return _normalize_docx_cell_text("".join(parts))
+
+
+def _extract_docx_table_rows(table_node: ElementTree.Element) -> list[list[str]]:
+    rows: list[list[str]] = []
+    for row_node in table_node:
+        if not row_node.tag.endswith("}tr"):
+            continue
+
+        row: list[str] = []
+        for cell_node in row_node:
+            if not cell_node.tag.endswith("}tc"):
+                continue
+
+            cell_parts = [
+                _extract_docx_paragraph_text(paragraph)
+                for paragraph in cell_node
+                if paragraph.tag.endswith("}p")
+            ]
+            row.append(_normalize_docx_cell_text(" ".join(part for part in cell_parts if part)))
+
+        if any(row):
+            rows.append(row)
+    return rows
+
+
+def _build_docx_table_markdown(rows: list[list[str]]) -> str:
+    visible_rows = [
+        [_escape_markdown_cell(cell) for cell in row]
+        for row in rows
+        if any(str(cell or "").strip() for cell in row)
+    ]
+    if len(visible_rows) < 2:
+        return ""
+
+    column_count = max(len(row) for row in visible_rows)
+    normalized_rows = [row + [""] * (column_count - len(row)) for row in visible_rows]
+    header = [
+        cell or f"列{column_index + 1}" for column_index, cell in enumerate(normalized_rows[0])
+    ]
+    body_rows = normalized_rows[1:]
+    parts = [_format_markdown_table(header, body_rows)]
+
+    row_clues: list[str] = []
+    for row_number, row in enumerate(body_rows, start=2):
+        pairs = [
+            f"{header[column_index]}={value}"
+            for column_index, value in enumerate(row)
+            if value
+        ]
+        if pairs:
+            row_clues.append(f"- 表格第 {row_number} 行：" + "；".join(pairs))
+
+    if row_clues:
+        parts.append("### 表格行级检索线索")
+        parts.extend(row_clues)
+
+    return "\n\n".join(parts)
+
+
+def _normalize_docx_cell_text(value: str) -> str:
+    normalized = str(value or "").replace("\r\n", "\n").replace("\r", "\n")
+    normalized = re.sub(r"[ \t]*\n[ \t]*", " ", normalized)
+    normalized = re.sub(r"\s+", " ", normalized)
+    return normalized.strip()
+
+
 def _extract_document_text_from_path(
    *,
    file_path: Path,
--- a/server/src/app/services/knowledge_normalizer.py
+++ b/server/src/app/services/knowledge_normalizer.py
@@ -12,7 +12,7 @@ logger = get_logger("app.services.knowledge_normalizer")

 TABLE_MARKER_PATTERN = re.compile(r"表\s*(\d+)")
 SECTION_HEADING_PATTERN = re.compile(
-    r"^(第[一二三四五六七八九十百零0-9]+[章节]\s*.*|[一二三四五六七八九十]+、.*|（[一二三四五六七八九十]+）.*|\([一二三四五六七八九十]+\).*)$"
+    r"^(第[一二三四五六七八九十百零0-9]+[部分章节]\s*.*|[一二三四五六七八九十]+、.*|（[一二三四五六七八九十]+）.*|\([一二三四五六七八九十]+\).*)$"
 )
 LIST_ITEM_PATTERN = re.compile(r"^[-*•]\s+.+$")
 NUMBERED_ITEM_PATTERN = re.compile(r"^(?:\d+[.)、]|[①②③④⑤⑥⑦⑧⑨⑩])\s*.+$")
--- a/server/src/app/services/knowledge_rag.py
+++ b/server/src/app/services/knowledge_rag.py
@@ -50,6 +50,12 @@ QUERY_TERM_STOPWORDS = {
    "哪些人",
 }
 TABLE_OR_STANDARD_QUERY_HINTS = (
+    "表",
+    "表格",
+    "清单",
+    "明细",
+    "目录",
+    "科目",
    "标准",
    "金额",
    "限额",
@@ -61,6 +67,20 @@ TABLE_OR_STANDARD_QUERY_HINTS = (
    "档位",
    "额度",
 )
+QUERY_ANCHOR_TERMS = (
+    "财务基础知识手册",
+    "基础知识手册",
+    "会计科目",
+    "常用会计科目",
+    "财务报表",
+    "主要税种",
+    "税种",
+    "标准",
+    "清单",
+    "明细",
+    "流程",
+)
+GENERIC_TITLE_TERMS = {"远光软件", "股份有限", "有限公司"}
 STRUCTURED_APPENDIX_LEADING_MARKERS = (
    "# 章节导航",
    "# 重点章节摘录",
@@ -96,6 +116,10 @@ class KnowledgeRagService:
                "message": "请先输入要检索的知识库问题。",
            }

+        rewritten_query = normalized_query
+        if conversation_history:
+            rewritten_query = self._rewrite_query(normalized_query, conversation_history)
+
        workspace = (
            os.environ.get("LIGHTRAG_WORKSPACE", DEFAULT_LIGHTRAG_WORKSPACE).strip()
            or DEFAULT_LIGHTRAG_WORKSPACE
@@ -103,81 +127,102 @@ class KnowledgeRagService:
        local_result = query_local_text_chunks(
            lightrag_root=(self.storage_root / "knowledge" / ".lightrag").resolve(),
            workspace=workspace,
-            query=normalized_query,
+            query=rewritten_query,
            limit=limit,
        )
-        if local_result.confident:
-            return {
-                "result_type": "knowledge_search",
-                "query": normalized_query,
-                "record_count": len(local_result.hits),
-                "hits": local_result.hits,
-                "references": [
-                    str(item.get("code") or "").strip()
-                    for item in local_result.hits
-                    if str(item.get("code") or "").strip()
-                ],
-                "raw_references": [],
-                "metadata": {
-                    "retrieval_strategy": "local_text_chunks",
-                    "elapsed_seconds": round(local_result.elapsed_seconds, 4),
-                    "total_chunks": local_result.total_chunks,
-                    "best_score": local_result.best_score,
-                },
-                "message": f"已从本地知识块中检索到 {len(local_result.hits)} 条相关内容。",
-            }

+        runtime_hits: list[dict[str, Any]] = []
+        runtime_references: list[str] = []
        try:
            runtime = self._get_runtime()
-            raw = runtime.query_data(normalized_query, conversation_history=conversation_history)
+            raw = runtime.query_data(rewritten_query, conversation_history=conversation_history)
+            data = raw.get("data") if isinstance(raw, dict) else {}
+            chunks = list(data.get("chunks") or []) if isinstance(data, dict) else []
+            entities = list(data.get("entities") or []) if isinstance(data, dict) else []
+            runtime_references = list(data.get("references") or []) if isinstance(data, dict) else []
+            runtime_hits = self._build_hits_from_query_data(
+                query=rewritten_query,
+                chunks=chunks,
+                entities=entities,
+                limit=limit,
+            )
        except Exception as exc:
            logger.warning("Knowledge query failed: %s", exc)
+
+        all_hits: dict[str, dict[str, Any]] = {}
+        for hit in local_result.hits:
+            hit["score"] = int(hit.get("score") or 0)
+            all_hits[hit["code"]] = hit
+
+        for hit in runtime_hits:
+            code = hit["code"]
+            if code in all_hits:
+                all_hits[code]["score"] = max(all_hits[code]["score"], int(hit.get("score") or 0) + 20)
+                if not all_hits[code].get("tags") and hit.get("tags"):
+                    all_hits[code]["tags"] = hit["tags"]
+            else:
+                hit["score"] = int(hit.get("score") or 0)
+                all_hits[code] = hit
+
+        merged_hits = sorted(all_hits.values(), key=lambda x: int(x.get("score") or 0), reverse=True)[:max(1, limit)]
+
+        if not merged_hits:
            return {
                "result_type": "knowledge_search",
-                "query": normalized_query,
+                "query": rewritten_query,
                "record_count": 0,
                "hits": [],
                "references": [],
-                "message": f"知识库检索暂不可用：{exc}",
-            }
-
-        data = raw.get("data") if isinstance(raw, dict) else {}
-        chunks = list(data.get("chunks") or []) if isinstance(data, dict) else []
-        entities = list(data.get("entities") or []) if isinstance(data, dict) else []
-        references = list(data.get("references") or []) if isinstance(data, dict) else []
-        hits = self._build_hits_from_query_data(
-            query=normalized_query,
-            chunks=chunks,
-            entities=entities,
-            limit=limit,
-        )
-
-        if not hits:
-            return {
-                "result_type": "knowledge_search",
-                "query": normalized_query,
-                "record_count": 0,
-                "hits": [],
-                "references": [],
-                "raw_references": references,
+                "raw_references": runtime_references,
                "message": "当前知识库中没有检索到与本次问题直接匹配的内容。",
            }

        return {
            "result_type": "knowledge_search",
-            "query": normalized_query,
-            "record_count": len(hits),
-            "hits": hits,
+            "query": rewritten_query,
+            "record_count": len(merged_hits),
+            "hits": merged_hits,
            "references": [
                str(item.get("code") or "").strip()
-                for item in hits
+                for item in merged_hits
                if str(item.get("code") or "").strip()
            ],
-            "raw_references": references,
-            "metadata": raw.get("metadata") if isinstance(raw, dict) else {},
-            "message": f"已从知识库中检索到 {len(hits)} 条相关内容。",
+            "raw_references": runtime_references,
+            "metadata": {
+                "retrieval_strategy": "fusion",
+                "local_total_chunks": local_result.total_chunks,
+                "local_best_score": local_result.best_score,
+            },
+            "message": f"已从知识库中联合检索到 {len(merged_hits)} 条相关内容。",
        }

+    def _rewrite_query(self, query: str, conversation_history: list[dict[str, str]]) -> str:
+        if not self.db:
+            return query
+            
+        from app.services.runtime_chat import RuntimeChatService
+        try:
+            chat_service = RuntimeChatService(self.db)
+            messages: list[dict[str, Any]] = [{"role": "system", "content": "你是一个查询重写助手。你的任务是根据用户的多轮对话历史，将用户的最后一次提问重写为一句独立、完整的查询语句，以便于在知识库中进行向量检索。只输出重写后的句子，不要任何解释。"}]
+            for msg in conversation_history[-6:]:
+                messages.append({"role": msg.get("role", "user"), "content": msg.get("content", "")})
+            messages.append({"role": "user", "content": f"当前提问：{query}\n\n请重写当前提问。"})
+            
+            rewritten = chat_service.complete(
+                messages,
+                max_tokens=60,
+                temperature=0.1,
+                timeout_seconds=10,
+            )
+            
+            if rewritten and len(rewritten) > 2 and len(rewritten) < 80:
+                logger.info("Query rewritten: '%s' -> '%s'", query, rewritten)
+                return rewritten
+        except Exception as exc:
+            logger.warning("Query rewrite failed: %s", exc)
+            
+        return query
+
    def index_documents(
        self,
        *,
@@ -686,6 +731,24 @@ def _extract_query_terms(query: str) -> list[str]:
        remember(item)

    for block in re.findall(r"[\u4e00-\u9fff]{2,20}", normalized_query):
+        for marker in ("标准", "金额", "限额", "额度"):
+            marker_index = block.find(marker)
+            if marker_index <= 0:
+                continue
+            subject = block[:marker_index]
+            for width in (6, 4, 3, 2):
+                remember(subject[-width:])
+        for anchor in QUERY_ANCHOR_TERMS:
+            if anchor in block:
+                remember(anchor)
+        tail = block[-14:]
+        for size in (8, 7, 6, 5, 4):
+            for start in range(0, len(tail) - size + 1):
+                piece = tail[start : start + size]
+                if any(anchor in piece for anchor in QUERY_ANCHOR_TERMS):
+                    remember(piece)
+                    if len(terms) >= MAX_QUERY_TERMS:
+                        return terms
        if len(block) <= 4:
            remember(block)
            continue
@@ -715,6 +778,11 @@ def _score_knowledge_hit(
    matched_terms = [term for term in query_terms if term in haystack]
    score += len(matched_terms) * 8
    score += sum(1 for term in matched_terms if term in title) * 6
+    score += sum(
+        (len(term) - 3) * 12
+        for term in matched_terms
+        if len(term) >= 4 and term in title and term not in GENERIC_TITLE_TERMS
+    )

    leading_appendix_marker = _leading_structured_appendix_marker(content)
    if leading_appendix_marker == "# 章节导航":
--- a/server/src/app/services/knowledge_rag_local.py
+++ b/server/src/app/services/knowledge_rag_local.py
@@ -42,6 +42,12 @@ LOCAL_QUERY_STOPWORDS = {
    "问题",
 }
 LOCAL_TABLE_QUERY_HINTS = (
+    "表",
+    "表格",
+    "清单",
+    "明细",
+    "目录",
+    "科目",
    "标准",
    "金额",
    "限额",
@@ -53,6 +59,20 @@ LOCAL_TABLE_QUERY_HINTS = (
    "档位",
    "额度",
 )
+LOCAL_QUERY_ANCHOR_TERMS = (
+    "财务基础知识手册",
+    "基础知识手册",
+    "会计科目",
+    "常用会计科目",
+    "财务报表",
+    "主要税种",
+    "税种",
+    "标准",
+    "清单",
+    "明细",
+    "流程",
+)
+LOCAL_GENERIC_TITLE_TERMS = {"远光软件", "股份有限", "有限公司"}
 LOCAL_DOMAIN_TERMS = (
    "报销",
    "费用",
@@ -253,6 +273,8 @@ def _score_local_chunk(
        score += weight
        if term in lowered_title:
            score += max(4, weight)
+            if len(term) >= 4 and term not in LOCAL_GENERIC_TITLE_TERMS:
+                score += (len(term) - 3) * 12
        occurrences = lowered_content.count(term)
        if occurrences > 1:
            score += min(8, occurrences * 2)
@@ -299,6 +321,24 @@ def _extract_local_query_terms(query: str) -> list[str]:
            remember(item)

    for block in re.findall(r"[\u4e00-\u9fff]{2,24}", normalized_query):
+        for marker in ("标准", "金额", "限额", "额度"):
+            marker_index = block.find(marker)
+            if marker_index <= 0:
+                continue
+            subject = block[:marker_index]
+            for width in (6, 4, 3, 2):
+                remember(subject[-width:])
+        for anchor in LOCAL_QUERY_ANCHOR_TERMS:
+            if anchor in block:
+                remember(anchor)
+        tail = block[-14:]
+        for size in (8, 7, 6, 5, 4):
+            for start in range(0, len(tail) - size + 1):
+                piece = tail[start : start + size]
+                if any(anchor in piece for anchor in LOCAL_QUERY_ANCHOR_TERMS):
+                    remember(piece)
+                    if len(terms) >= MAX_LOCAL_QUERY_TERMS:
+                        return terms
        if len(block) <= 4:
            remember(block)
            continue
--- a/server/src/app/services/ontology.py
+++ b/server/src/app/services/ontology.py
@@ -102,7 +102,7 @@ class SemanticOntologyService(
        context_json = payload.context_json or {}
        reference = self._load_reference_catalog()
        compact_query = self._compact(query)
-        entities = self._extract_entities(query, compact_query, reference)
+        entities = self._extract_entities(query, compact_query, reference, context_json=context_json)
        rule_scenario, scenario_score = self._detect_scenario(compact_query)
        time_range, _time_score = self._extract_time_range(
            query,
@@ -111,9 +111,14 @@ class SemanticOntologyService(
        )
        session_scenario = self._resolve_session_type_scenario(context_json)
        context_scenario = self._resolve_context_scenario(context_json)
+        application_context = self._is_expense_application_context(context_json)
+        application_query = self._looks_like_expense_application(compact_query)
        if session_scenario == "knowledge":
            rule_scenario = "knowledge"
            scenario_score = max(scenario_score, 0.34)
+        if session_scenario != "knowledge" and (application_context or application_query):
+            rule_scenario = "expense"
+            scenario_score = max(scenario_score, 0.22)
        if rule_scenario == "unknown" and context_scenario is not None:
            rule_scenario = context_scenario
            scenario_score = max(scenario_score, 0.14)
@@ -138,6 +143,9 @@ class SemanticOntologyService(
            entities=entities,
            time_range=time_range,
        )
+        if session_scenario != "knowledge" and (application_context or application_query):
+            rule_intent = "draft"
+            intent_score = max(intent_score, 0.22)
        if session_scenario != "knowledge" and self._should_inherit_expense_draft(
            compact_query,
            scenario=rule_scenario,
--- a/server/src/app/services/ontology_detection.py
+++ b/server/src/app/services/ontology_detection.py
@@ -20,6 +20,8 @@ from app.services.ontology_rules import (
    COMPARE_KEYWORDS,
    DRAFT_FOLLOW_UP_KEYWORDS,
    DRAFT_KEYWORDS,
+    EXPENSE_APPLICATION_CONTEXT_TYPES,
+    EXPENSE_APPLICATION_KEYWORDS,
    EXPENSE_NARRATIVE_KEYWORDS,
    EXPENSE_REVIEW_ACTIONS,
    EXPLAIN_KEYWORDS,
@@ -71,6 +73,21 @@ EXPLICIT_ENTERTAINMENT_KEYWORDS = (


 class OntologyDetectionMixin:
+    @staticmethod
+    def _is_expense_application_context(context_json: dict[str, Any]) -> bool:
+        document_type = str(context_json.get("document_type") or "").strip()
+        application_stage = str(context_json.get("application_stage") or "").strip()
+        entry_source = str(context_json.get("entry_source") or "").strip()
+        return (
+            document_type in EXPENSE_APPLICATION_CONTEXT_TYPES
+            or application_stage in EXPENSE_APPLICATION_CONTEXT_TYPES
+            or entry_source in {"documents_application", "expense_application"}
+        )
+
+    @staticmethod
+    def _looks_like_expense_application(compact_query: str) -> bool:
+        return any(keyword in compact_query for keyword in EXPENSE_APPLICATION_KEYWORDS)
+
    def _detect_scenario(self, compact_query: str) -> tuple[str, float]:
        scores = {key: 0.0 for key in SCENARIO_KEYWORDS}
        for scenario, keywords in SCENARIO_KEYWORDS.items():
@@ -341,6 +358,9 @@ class OntologyDetectionMixin:
                "conversation_id": payload.context_json.get("conversation_id"),
                "conversation_scenario": payload.context_json.get("conversation_scenario"),
                "conversation_intent": payload.context_json.get("conversation_intent"),
+                "document_type": payload.context_json.get("document_type"),
+                "application_stage": payload.context_json.get("application_stage"),
+                "application_fields": payload.context_json.get("application_fields"),
                "draft_claim_id": payload.context_json.get("draft_claim_id"),
                "review_action": payload.context_json.get("review_action"),
                "review_form_values": payload.context_json.get("review_form_values"),
--- a/server/src/app/services/ontology_extraction.py
+++ b/server/src/app/services/ontology_extraction.py
@@ -18,7 +18,12 @@ from app.services.ontology_rules import (
    DATE_RANGE_PATTERN,
    EXPLICIT_DATE_PATTERN,
    EXPLICIT_MONTH_PATTERN,
+    EXPENSE_APPLICATION_ATTACHMENT_REQUIRED_TYPES,
+    EXPENSE_APPLICATION_CONTEXT_TYPES,
+    EXPENSE_APPLICATION_KEYWORDS,
+    EXPENSE_APPLICATION_REQUIRED_SLOT_KEYS,
    EXPENSE_TYPE_KEYWORDS,
+    GENERIC_EXPENSE_APPLICATION_PROMPTS,
    GENERIC_EXPENSE_PROMPTS,
    LOCATION_KEYWORDS,
    MONTH_DAY_PATTERN,
@@ -30,6 +35,21 @@ from app.services.ontology_rules import (


 class OntologyExtractionMixin:
+    @staticmethod
+    def _is_expense_application_context_value(context_json: dict[str, Any]) -> bool:
+        document_type = str(context_json.get("document_type") or "").strip()
+        application_stage = str(context_json.get("application_stage") or "").strip()
+        entry_source = str(context_json.get("entry_source") or "").strip()
+        return (
+            document_type in EXPENSE_APPLICATION_CONTEXT_TYPES
+            or application_stage in EXPENSE_APPLICATION_CONTEXT_TYPES
+            or entry_source in {"documents_application", "expense_application"}
+        )
+
+    @staticmethod
+    def _has_expense_application_signal(compact_query: str) -> bool:
+        return any(keyword in compact_query for keyword in EXPENSE_APPLICATION_KEYWORDS)
+
    def _infer_default_missing_slots(
        self,
        compact_query: str,
@@ -46,6 +66,44 @@ class OntologyExtractionMixin:
        entity_types = {item.type for item in entities}
        attachment_count = int(context_json.get("attachment_count") or 0)
        missing_slots: list[str] = []
+        application_mode = (
+            self._is_expense_application_context_value(context_json)
+            or self._has_expense_application_signal(compact_query)
+            or any(
+                item.type == "document_type" and item.normalized_value == "expense_application"
+                for item in entities
+            )
+        )
+
+        if application_mode:
+            form_values = context_json.get("review_form_values")
+            if not isinstance(form_values, dict):
+                form_values = {}
+            expense_type_codes = {
+                str(item.normalized_value or item.value or "").strip()
+                for item in entities
+                if item.type == "expense_type"
+            }
+            if "expense_type" not in entity_types and not str(form_values.get("expense_type") or "").strip():
+                missing_slots.append("expense_type")
+            if "amount" not in entity_types and not str(form_values.get("amount") or "").strip():
+                missing_slots.append("amount")
+            if not time_range.start_date and not (
+                str(form_values.get("time_range") or form_values.get("business_time") or "").strip()
+            ):
+                missing_slots.append("time_range")
+            reason_value = str(
+                form_values.get("reason")
+                or form_values.get("business_reason")
+                or form_values.get("reason_value")
+                or ""
+            ).strip()
+            if not reason_value and compact_query in GENERIC_EXPENSE_APPLICATION_PROMPTS:
+                missing_slots.append("reason")
+            if attachment_count <= 0 and expense_type_codes & EXPENSE_APPLICATION_ATTACHMENT_REQUIRED_TYPES:
+                missing_slots.append("attachments")
+            ordered_keys = [*EXPENSE_APPLICATION_REQUIRED_SLOT_KEYS, "attachments"]
+            return [item for item in ordered_keys if item in missing_slots]

        if self._is_generic_expense_prompt(compact_query):
            if "expense_type" not in entity_types:
@@ -98,14 +156,40 @@ class OntologyExtractionMixin:
        query: str,
        compact_query: str,
        reference: ReferenceCatalog,
+        *,
+        context_json: dict[str, Any] | None = None,
    ) -> list[OntologyEntity]:
        entities: dict[tuple[str, str], OntologyEntity] = {}
+        context_json = context_json or {}

        def upsert(entity: OntologyEntity) -> None:
            key = (entity.type, entity.normalized_value)
            if key not in entities:
                entities[key] = entity

+        if (
+            self._is_expense_application_context_value(context_json)
+            or self._has_expense_application_signal(compact_query)
+        ):
+            upsert(
+                self._make_entity(
+                    "document_type",
+                    "费用申请",
+                    "expense_application",
+                    role="target",
+                    confidence=0.94,
+                )
+            )
+            upsert(
+                self._make_entity(
+                    "workflow_stage",
+                    "前置申请",
+                    "pre_approval",
+                    role="target",
+                    confidence=0.9,
+                )
+            )
+
        for match in re.finditer(r"客户\s*([A-Za-z0-9一二三四五六七八九十]+)", query):
            suffix = match.group(1).strip()
            normalized = f"客户{suffix}".replace(" ", "")
@@ -510,6 +594,8 @@ class OntologyExtractionMixin:
                "project",
                "location",
                "expense_type",
+                "document_type",
+                "workflow_stage",
            }:
                upsert(
                    OntologyConstraint(
--- a/server/src/app/services/ontology_rules.py
+++ b/server/src/app/services/ontology_rules.py
@@ -173,6 +173,49 @@ GENERIC_EXPENSE_PROMPTS = {
    "发起报销",
    "提交报销",
 }
+EXPENSE_APPLICATION_CONTEXT_TYPES = {
+    "expense_application",
+    "application",
+    "pre_approval",
+    "preapproval",
+}
+EXPENSE_APPLICATION_KEYWORDS = (
+    "费用申请",
+    "申请单",
+    "发起申请",
+    "提交申请",
+    "提出申请",
+    "前置申请",
+    "报销申请",
+    "申请报销",
+    "差旅申请",
+    "出差申请",
+    "会务申请",
+    "会议申请",
+    "采购申请",
+    "培训申请",
+    "预算申请",
+)
+GENERIC_EXPENSE_APPLICATION_PROMPTS = {
+    "申请",
+    "费用申请",
+    "发起申请",
+    "提交申请",
+    "提出申请",
+    "申请报销",
+    "报销申请",
+}
+EXPENSE_APPLICATION_REQUIRED_SLOT_KEYS = (
+    "expense_type",
+    "amount",
+    "time_range",
+    "reason",
+)
+EXPENSE_APPLICATION_ATTACHMENT_REQUIRED_TYPES = {
+    "meeting",
+    "office",
+    "training",
+}
 MISSING_SLOT_LABELS = {
    "expense_type": "费用类型",
    "amount": "金额",
--- a/server/src/app/services/risk_rule_generation.py
+++ b/server/src/app/services/risk_rule_generation.py
@@ -14,6 +14,7 @@ from app.schemas.agent_asset import AgentAssetRiskRuleGenerateRequest
 from app.services.agent_asset_rule_library import AgentAssetRuleLibraryManager
 from app.services.agent_asset_spreadsheet import RISK_RULES_LIBRARY
 from app.services.audit import AuditLogService
+from app.services.expense_type_keywords import EXPENSE_TYPE_LABEL_BY_CODE
 from app.services.risk_rule_flow_diagram import (
    RiskRuleFlowDiagramField,
    RiskRuleFlowDiagramRenderer,
@@ -43,6 +44,24 @@ RISK_LEVEL_LABELS: dict[str, str] = {
    "high": "高风险",
 }

+EXPENSE_RISK_CATEGORY_CODES: tuple[str, ...] = (
+    "travel",
+    "hotel",
+    "transport",
+    "meal",
+    "meeting",
+    "office",
+    "training",
+    "communication",
+    "welfare",
+)
+EXPENSE_RISK_CATEGORY_LABELS: dict[str, str] = {
+    code: EXPENSE_TYPE_LABEL_BY_CODE[code] for code in EXPENSE_RISK_CATEGORY_CODES
+}
+EXPENSE_RISK_CATEGORY_ALIASES = {
+    "entertainment": "meal",
+}
+
 FIELD_ONTOLOGY: tuple[RiskRuleField, ...] = (
    RiskRuleField("claim.reason", "报销事由", "text", "claim", ("事由", "说明", "理由", "用途")),
    RiskRuleField(
@@ -156,17 +175,23 @@ class RiskRuleGenerationService:
        risk_level = str(body.risk_level or "medium").strip().lower()
        if risk_level not in RISK_LEVEL_LABELS:
            raise ValueError("风险等级仅支持 low、medium、high。")
+        requires_attachment = bool(body.requires_attachment)
+        expense_category = self._normalize_expense_category(body.expense_category, domain)
+        expense_category_label = EXPENSE_RISK_CATEGORY_LABELS.get(expense_category or "", "")

        created_at = datetime.now(UTC)
        fields = self._resolve_fields(natural_language, domain=domain)
        draft = self._compile_with_model(
            natural_language=natural_language,
            domain=domain,
+            expense_category=expense_category,
+            expense_category_label=expense_category_label,
            risk_level=risk_level,
            fields=fields,
        ) or self._build_fallback_draft(
            natural_language=natural_language,
            domain=domain,
+            expense_category_label=expense_category_label,
            risk_level=risk_level,
            fields=fields,
        )
@@ -179,10 +204,13 @@ class RiskRuleGenerationService:
            draft,
            natural_language=natural_language,
            domain=domain,
+            expense_category=expense_category,
+            expense_category_label=expense_category_label,
            risk_level=risk_level,
            fields=fields,
            created_at=created_at,
            actor=actor,
+            requires_attachment=requires_attachment,
        )
        rule_code = str(payload["rule_code"])
        file_name = f"{rule_code}.json"
@@ -209,8 +237,11 @@ class RiskRuleGenerationService:
            config_json={
                "severity": risk_level,
                "enabled": True,
+                "requires_attachment": requires_attachment,
                "tag": "风险规则",
                "detail_mode": "json_risk",
+                "expense_category": expense_category,
+                "expense_category_label": expense_category_label,
                "risk_category": payload.get("risk_category"),
                "rule_library": RISK_RULES_LIBRARY,
                "rule_document": {
@@ -241,7 +272,13 @@ class RiskRuleGenerationService:
            resource_type=AgentAssetType.RULE.value,
            resource_id=asset.id,
            before_json=None,
-            after_json={"rule_code": rule_code, "risk_level": risk_level, "domain": domain},
+            after_json={
+                "rule_code": rule_code,
+                "risk_level": risk_level,
+                "domain": domain,
+                "expense_category": expense_category,
+                "requires_attachment": requires_attachment,
+            },
            request_id=request_id,
        )
        self.db.refresh(asset)
@@ -252,6 +289,8 @@ class RiskRuleGenerationService:
        *,
        natural_language: str,
        domain: str,
+        expense_category: str | None,
+        expense_category_label: str,
        risk_level: str,
        fields: list[RiskRuleField],
    ) -> dict[str, Any] | None:
@@ -279,6 +318,8 @@ class RiskRuleGenerationService:
                    {
                        "business_domain": domain,
                        "business_domain_label": BUSINESS_DOMAIN_LABELS[domain],
+                        "expense_category": expense_category,
+                        "expense_category_label": expense_category_label,
                        "risk_level": risk_level,
                        "risk_level_label": RISK_LEVEL_LABELS[risk_level],
                        "natural_language": natural_language,
@@ -370,6 +411,7 @@ class RiskRuleGenerationService:
        *,
        natural_language: str,
        domain: str,
+        expense_category_label: str,
        risk_level: str,
        fields: list[RiskRuleField],
    ) -> dict[str, Any]:
@@ -381,8 +423,9 @@ class RiskRuleGenerationService:
            fields=fields,
        )
        name = self._infer_rule_name(natural_language)
+        business_label = expense_category_label or BUSINESS_DOMAIN_LABELS[domain]
        description = (
-            f"当{BUSINESS_DOMAIN_LABELS[domain]}业务满足“{natural_language}”时，系统会按"
+            f"当{business_label}业务满足“{natural_language}”时，系统会按"
            f"{RISK_LEVEL_LABELS[risk_level]}进行提示，并要求经办人或审核人补充核对依据。"
        )
        return {
@@ -393,7 +436,7 @@ class RiskRuleGenerationService:
            "condition_summary": condition_summary,
            "keywords": self._infer_keywords(natural_language),
            "flow": {
-                "start": f"{BUSINESS_DOMAIN_LABELS[domain]}单据提交",
+                "start": f"{business_label}单据提交",
                "evidence": "读取" + "、".join(item.label for item in fields[:3]),
                "decision": condition_summary,
                "pass": "未命中风险，继续业务流转",
@@ -407,14 +450,18 @@ class RiskRuleGenerationService:
        *,
        natural_language: str,
        domain: str,
+        expense_category: str | None,
+        expense_category_label: str,
        risk_level: str,
        fields: list[RiskRuleField],
        created_at: datetime,
        actor: str,
+        requires_attachment: bool,
    ) -> dict[str, Any]:
-        created_stamp = created_at.strftime("%Y%m%d%H%M%S")
+        created_stamp = created_at.strftime("%Y%m%d%H%M%S%f")
        domain_slug = {"expense": "expense", "ar": "ar", "ap": "ap"}[domain]
-        rule_code = f"risk.{domain_slug}.generated_{created_stamp}"
+        category_slug = f".{expense_category}" if expense_category else ""
+        rule_code = f"risk.{domain_slug}{category_slug}.generated_{created_stamp}"
        template_key = str(draft.get("template_key") or "field_required_v1").strip()
        field_keys = [
            str(item or "").strip()
@@ -424,7 +471,7 @@ class RiskRuleGenerationService:
        condition_summary = (
            self._clean_text(draft.get("condition_summary")) or "判断是否符合自然语言规则描述"
        )
-        risk_category = BUSINESS_DOMAIN_LABELS[domain]
+        risk_category = expense_category_label or BUSINESS_DOMAIN_LABELS[domain]
        keywords = list(draft.get("keywords") or [])
        field_by_key = {item.key: item for item in fields}
        params: dict[str, Any] = {
@@ -440,6 +487,9 @@ class RiskRuleGenerationService:
        if template_key == "keyword_match_v1":
            params["keywords"] = keywords
            params["search_fields"] = field_keys
+        applies_to: dict[str, Any] = {"domains": [domain]}
+        if expense_category:
+            applies_to["expense_categories"] = [expense_category]

        payload = {
            "schema_version": "2.0",
@@ -447,12 +497,13 @@ class RiskRuleGenerationService:
            "name": self._clean_text(draft.get("name")) or self._infer_rule_name(natural_language),
            "description": self._clean_text(draft.get("description")) or natural_language,
            "enabled": True,
+            "requires_attachment": requires_attachment,
            "risk_dimension": "natural_language_rule",
            "risk_category": risk_category,
            "ontology_signal": "natural_language_risk",
            "evaluator": "template_rule",
            "template_key": template_key,
-            "applies_to": {"domains": [domain]},
+            "applies_to": applies_to,
            "inputs": {
                "fields": [
                    {
@@ -478,6 +529,9 @@ class RiskRuleGenerationService:
                "source_ref": "自然语言风险规则",
                "created_at": created_at.isoformat(),
                "created_by": actor,
+                "requires_attachment": requires_attachment,
+                "expense_category": expense_category,
+                "expense_category_label": expense_category_label,
                "natural_language": natural_language,
                "business_explanation": self._clean_text(draft.get("description")),
                "condition_summary": condition_summary,
@@ -488,6 +542,7 @@ class RiskRuleGenerationService:
            payload,
            fields=[field_by_key[key] for key in field_keys if key in field_by_key],
            domain=domain,
+            domain_label=risk_category,
            risk_level=risk_level,
        )
        return payload
@@ -498,6 +553,7 @@ class RiskRuleGenerationService:
        *,
        fields: list[RiskRuleField],
        domain: str,
+        domain_label: str | None = None,
        risk_level: str,
    ) -> str:
        metadata = payload.get("metadata") if isinstance(payload.get("metadata"), dict) else {}
@@ -506,7 +562,7 @@ class RiskRuleGenerationService:
        return self.flow_diagram_renderer.render(
            RiskRuleFlowDiagramSpec(
                title=self._clean_text(payload.get("name")) or "风险规则判断流程",
-                domain_label=BUSINESS_DOMAIN_LABELS.get(domain, "业务"),
+                domain_label=domain_label or BUSINESS_DOMAIN_LABELS.get(domain, "业务"),
                severity=risk_level,
                severity_label=RISK_LEVEL_LABELS.get(risk_level, "中风险"),
                fields=tuple(
@@ -528,6 +584,21 @@ class RiskRuleGenerationService:
            )
        )

+    @staticmethod
+    def _normalize_expense_category(value: str | None, domain: str) -> str | None:
+        if domain != AgentAssetDomain.EXPENSE.value:
+            return None
+
+        normalized = str(value or "").strip().lower()
+        if not normalized:
+            return None
+
+        normalized = EXPENSE_RISK_CATEGORY_ALIASES.get(normalized, normalized)
+        if normalized not in EXPENSE_RISK_CATEGORY_LABELS:
+            allowed = "、".join(EXPENSE_RISK_CATEGORY_LABELS.values())
+            raise ValueError(f"费用领域仅支持：{allowed}。")
+        return normalized
+
    def _resolve_fields(self, text: str, *, domain: str) -> list[RiskRuleField]:
        prefixes = DOMAIN_FIELD_PREFIXES.get(domain, ())
        candidates = [field for field in FIELD_ONTOLOGY if field.key.startswith(prefixes)]
--- a/server/src/app/services/risk_rule_template_executor.py
+++ b/server/src/app/services/risk_rule_template_executor.py
@@ -172,8 +172,12 @@ class RiskRuleTemplateExecutor:
            if field_key == "ocr_text":
                values.extend([context.get("ocr_text"), context.get("ocr_summary")])
            if field_key in {"hotel_city", "route_cities"}:
-                values.extend(self._scan_document_values(document_info, field_key))
-                values.extend(self._scan_document_values(document_info, "city"))
+                specific_values = self._scan_document_values(document_info, field_key)
+                values.extend(
+                    specific_values
+                    if specific_values
+                    else self._scan_document_values(document_info, "city")
+                )
            else:
                values.extend(self._scan_document_values(document_info, field_key))
        return self._normalize_values(values)
@@ -203,8 +207,8 @@ class RiskRuleTemplateExecutor:
            "buyer_name": ("购买方", "抬头", "买方"),
            "goods_name": ("品名", "商品", "服务名称"),
            "issue_date": ("日期", "开票日期", "发票日期"),
-            "hotel_city": ("住宿城市", "酒店城市", "酒店地点"),
-            "route_cities": ("行程", "路线", "城市"),
+            "hotel_city": ("住宿城市", "酒店城市", "酒店地点", "住宿", "酒店"),
+            "route_cities": ("行程", "路线", "目的地", "出差城市"),
            "city": ("城市", "地点"),
        }
        return any(item in label for item in label_map.get(field_key, ()))
--- a/server/src/app/services/settings.py
+++ b/server/src/app/services/settings.py
@@ -16,6 +16,7 @@ from app.db.session import get_session_factory
 from app.models.system_model_setting import SystemModelSetting
 from app.models.system_setting import SystemSetting
 from app.models.system_setting_secret import SystemSettingSecret
+from app.models.hermes_config import HermesTaskConfig
 from app.repositories.settings import SETTINGS_ROW_ID, SettingsRepository
 from app.schemas.settings import SettingsRead, SettingsWrite
 from app.services.hermes_sync import (
@@ -183,6 +184,8 @@ class SettingsService:
                capability=config.capability,
                priority=config.priority,
                enabled=True,
+
+
                api_key_encrypted=str(getattr(secrets_row, config.legacy_secret_attr, "") or ""),
            )
            self.db.add(model_row)
@@ -199,12 +202,12 @@ class SettingsService:
    def get_settings_snapshot(self) -> SettingsRead:
        settings_row, secrets_row = self.ensure_settings_ready()
        model_rows = self.ensure_model_settings_ready(settings_row, secrets_row)
-        return self._serialize(settings_row, secrets_row, model_rows)
+        hermes_form = self._build_hermes_form_snapshot()
+        return self._serialize(settings_row, secrets_row, model_rows, hermes_form)

    def save_settings_snapshot(self, payload: SettingsWrite) -> SettingsRead:
        settings_row, secrets_row = self.ensure_settings_ready()
        model_rows = self.ensure_model_settings_ready(settings_row, secrets_row)
-
        if payload.adminForm.newPassword:
            if len(payload.adminForm.newPassword) < 5:
                raise ValueError("管理员密码至少需要 5 位。")
@@ -309,6 +312,8 @@ class SettingsService:

        hermes_snapshot = capture_hermes_config_snapshot()
        
+        self._save_hermes_form_snapshot(payload.hermesForm)
+
        try:
            sync_hermes_model_settings(
                primary_route=self._build_hermes_model_route(model_rows["main"]),
@@ -642,11 +647,75 @@ class SettingsService:

        return should_commit

+    def _build_hermes_form_snapshot(self) -> dict:
+        configs = self.db.query(HermesTaskConfig).all()
+        capabilities = {}
+        schedules = {}
+        master_enabled = True # 这里假设只要有一个开启，主开关就是开启的（为简单起见）
+
+        for config in configs:
+            task_type = config.task_type
+            capabilities[task_type] = config.is_enabled
+            
+            # 简化解析 cron_expression 到 time (假设 cron 为 "0 9 * * 1" 这种形式)
+            time_str = "00:00"
+            if config.cron_expression:
+                parts = config.cron_expression.split(" ")
+                if len(parts) >= 2:
+                    minute, hour = parts[0], parts[1]
+                    try:
+                        time_str = f"{int(hour):02d}:{int(minute):02d}"
+                    except ValueError:
+                        pass
+                        
+            schedules[task_type] = {
+                "enabled": config.is_enabled,
+                "time": time_str
+            }
+            
+        return {
+            "masterEnabled": master_enabled,
+            "notifyOnFailure": True,
+            "capabilities": capabilities,
+            "schedules": schedules
+        }
+        
+    def _save_hermes_form_snapshot(self, hermes_form: dict) -> None:
+        if not hermes_form:
+            return
+            
+        schedules = hermes_form.get("schedules", {})
+        capabilities = hermes_form.get("capabilities", {})
+        master_enabled = hermes_form.get("masterEnabled", True)
+        
+        for task_type, schedule in schedules.items():
+            config = self.db.query(HermesTaskConfig).filter_by(task_type=task_type).first()
+            if not config:
+                config = HermesTaskConfig(task_type=task_type)
+                self.db.add(config)
+                
+            task_enabled = schedule.get("enabled", False) and capabilities.get(task_type, False) and master_enabled
+            config.is_enabled = task_enabled
+            
+            # 从 time 构建简单的 cron expression
+            time_str = schedule.get("time", "00:00")
+            parts = time_str.split(":")
+            if len(parts) == 2:
+                # 简单映射：把时分放进去，后面保留为 * * * （或者保留旧的后半段）
+                # 这里偷个懒，风险扫描每天跑，周报每周一跑
+                if task_type == "global_risk_scan":
+                    config.cron_expression = f"{int(parts[1])} {int(parts[0])} * * *"
+                elif task_type == "weekly_expense_report":
+                    config.cron_expression = f"{int(parts[1])} {int(parts[0])} * * 1"
+                else:
+                    config.cron_expression = f"{int(parts[1])} {int(parts[0])} * * *"
+
    @staticmethod
    def _serialize(
        settings_row: SystemSetting,
        secrets_row: SystemSettingSecret,
        model_rows: dict[str, SystemModelSetting],
+        hermes_form: dict,
    ) -> SettingsRead:
        main_model = model_rows["main"]
        backup_model = model_rows["backup"]
@@ -676,12 +745,9 @@ class SettingsService:
            sessionForm={
                "conversationRetentionDays": settings_row.conversation_retention_days,
            },
+            hermesForm=hermes_form,
            llmForm={
                "mainProvider": main_model.provider,
-                "mainModel": main_model.model_name,
-                "mainEndpoint": main_model.endpoint,
-                "mainApiKey": "",
-                "mainApiKeyConfigured": bool(main_model.api_key_encrypted),
                "backupProvider": backup_model.provider,
                "backupModel": backup_model.model_name,
                "backupEndpoint": backup_model.endpoint,
--- a/server/src/app/services/user_agent_constants.py
+++ b/server/src/app/services/user_agent_constants.py
@@ -71,8 +71,8 @@ EXPENSE_SCENE_SELECTION_OPTIONS = (
    ("other", "其他费用", "暂不属于以上分类的报销场景。"),
 )

-KNOWLEDGE_MODEL_MAIN_TIMEOUT_SECONDS = 3
-KNOWLEDGE_MODEL_BACKUP_TIMEOUT_SECONDS = 5
+KNOWLEDGE_MODEL_MAIN_TIMEOUT_SECONDS = 20
+KNOWLEDGE_MODEL_BACKUP_TIMEOUT_SECONDS = 30
 KNOWLEDGE_MODEL_TIMEOUT_SECONDS = KNOWLEDGE_MODEL_BACKUP_TIMEOUT_SECONDS

 EXPENSE_STATUS_LABELS = {
--- a/server/src/app/services/user_agent_knowledge.py
+++ b/server/src/app/services/user_agent_knowledge.py
@@ -86,6 +86,7 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
        *,
        citations: list[UserAgentCitation],
    ) -> str | None:
+        return None
        if payload.ontology.scenario != "knowledge":
            return None
        if str(payload.tool_payload.get("result_type") or "").strip() != "knowledge_search":
@@ -583,20 +584,23 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
        evidence_lines: list[str] = []
        for item in evidence_items[:3]:
            heading = str(item.get("heading") or "").strip()
-            heading_text = f" > {heading}" if heading else ""
+            if "表格行级检索线索" in heading:
+                heading = heading.replace("表格行级检索线索", "").strip(" >")
+            heading_text = f"（{heading}）" if heading else ""
+            item_title = item.get("title") or title
            if str(item.get("kind") or "") == "table":
                preview = self._extract_relevant_table_preview(
                    str(item.get("content") or ""),
                    self._extract_knowledge_query_terms(self._resolve_knowledge_question(payload)),
                )
-                evidence_lines.append(f"- 《{item.get('title') or title}》{heading_text}：\n{preview}")
+                evidence_lines.append(f"- **《{item_title}》** {heading_text}\n{preview}")
                continue
            rendered = self._render_knowledge_evidence_text(item)
            if rendered:
                if "\n" in rendered:
-                    evidence_lines.append(f"- 《{item.get('title') or title}》{heading_text}：\n{rendered}")
+                    evidence_lines.append(f"- **《{item_title}》** {heading_text}\n{rendered}")
                else:
-                    evidence_lines.append(f"- 《{item.get('title') or title}》{heading_text}：{rendered}")
+                    evidence_lines.append(f"- **《{item_title}》** {heading_text}\n  {rendered}")

        if not evidence_lines:
            for item in hits[:2]:
@@ -607,21 +611,22 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
                )
                if not excerpt:
                    continue
-                evidence_lines.append(f"- 《{item_title}》：{excerpt}")
+                evidence_lines.append(f"- **《{item_title}》**：{excerpt}")

        if not evidence_lines:
            return (
-                f"{prefix}我已经从《{title}》中检索到与你这次问题相关的制度依据，"
-                "但本次答案生成环节暂时没有成功返回。请稍后重试一次；如果仍然失败，"
-                "建议先检查主对话模型的连通性。"
+                f"{prefix}当前《{title}》里可用于回答的关键条款还不够明确。"
+                "请补充费用类型、适用地区、职级或具体业务场景，我再继续帮你缩小范围。"
            )

        return "\n".join(
            [
-                f"{prefix}我已经命中与你这次问题最相关的制度依据，但答案整理阶段本轮没有及时返回。",
-                "先给你当前最直接的依据：",
+                f"{prefix}我先根据当前制度依据给出可以确认的部分。",
+                "",
+                "**依据**：",
                *evidence_lines,
-                "如果你希望我继续把这些依据整理成更完整的结论、步骤或对比说明，可以继续缩小问题范围后再问一次。",
+                "",
+                "**说明**：以上只使用当前命中的知识库证据；没有在证据中出现的适用条件或金额，我不会替你默认补齐。",
            ]
        ).strip()

--- a/server/src/app/services/user_agent_knowledge_constants.py
+++ b/server/src/app/services/user_agent_knowledge_constants.py
@@ -4,6 +4,9 @@ import re

 KNOWLEDGE_DIRECT_ANSWER_HINTS = (
    "是什么",
+    "介绍",
+    "说明",
+    "概述",
    "标准",
    "限额",
    "流程",
@@ -45,7 +48,7 @@ MAX_KNOWLEDGE_QUERY_TERMS = 12
 MAX_KNOWLEDGE_DIRECT_EVIDENCE = 4
 MAX_KNOWLEDGE_MODEL_HITS = 5
 KNOWLEDGE_SECTION_HEADING_PATTERN = re.compile(
-    r"^(#\s*.+|##\s*.+|###\s*.+|第[一二三四五六七八九十百零0-9]+[章节条]\s*.*|[一二三四五六七八九十]+、.*|（[一二三四五六七八九十]+）.*|\([一二三四五六七八九十]+\).*)$"
+    r"^(#\s*.+|##\s*.+|###\s*.+|第[一二三四五六七八九十百零0-9]+[部分章节条]\s*.*|[一二三四五六七八九十]+、.*|（[一二三四五六七八九十]+）.*|\([一二三四五六七八九十]+\).*)$"
 )
 KNOWLEDGE_LIST_ITEM_PATTERN = re.compile(r"^[-*•]\s+.+$")
 KNOWLEDGE_NUMBERED_ITEM_PATTERN = re.compile(
--- a/server/src/app/services/user_agent_knowledge_helpers.py
+++ b/server/src/app/services/user_agent_knowledge_helpers.py
@@ -15,6 +15,20 @@ from app.services.user_agent_knowledge_constants import (


 class UserAgentKnowledgeHelpersMixin:
+    GENERIC_KNOWLEDGE_TITLE_TERMS = {"远光软件", "股份有限", "有限公司"}
+    KNOWLEDGE_QUERY_ANCHOR_TERMS = (
+        "财务基础知识手册",
+        "基础知识手册",
+        "会计科目",
+        "常用会计科目",
+        "财务报表",
+        "主要税种",
+        "税种",
+        "标准",
+        "清单",
+        "明细",
+        "流程",
+    )

    @staticmethod
    def _select_knowledge_model_hits(
@@ -26,7 +40,7 @@ class UserAgentKnowledgeHelpersMixin:
            item
            for item in list(tool_payload.get("hits") or [])
            if isinstance(item, dict)
-        ][: max(MAX_KNOWLEDGE_MODEL_HITS + 1, 6)]
+        ][: max(MAX_KNOWLEDGE_MODEL_HITS + 3, 8)]
        if not raw_hits:
            return []

@@ -64,7 +78,16 @@ class UserAgentKnowledgeHelpersMixin:
        matched_terms = [term for term in query_terms if term in haystack]
        score = max(1, 48 - rank_index * 4)
        score += len(matched_terms) * 10
+        score += sum(max(0, len(term) - 4) * 8 for term in matched_terms)
        score += sum(1 for term in matched_terms if term in title) * 8
+        score += sum(max(0, len(term) - 4) * 6 for term in matched_terms if term in title)
+        score += sum(
+            (len(term) - 3) * 10
+            for term in matched_terms
+            if len(term) >= 4
+            and term in title
+            and term not in UserAgentKnowledgeHelpersMixin.GENERIC_KNOWLEDGE_TITLE_TERMS
+        )

        leading_marker = UserAgentKnowledgeHelpersMixin._leading_knowledge_appendix_marker(content)
        if leading_marker == "# 章节导航":
@@ -149,6 +172,40 @@ class UserAgentKnowledgeHelpersMixin:
        return ""


+    @staticmethod
+    def _knowledge_list_marker_sort_key(content: str) -> int:
+        normalized = str(content or "").strip()
+        match = re.match(r"^[（(]([一二三四五六七八九十百零0-9]+)[)）]", normalized)
+        if not match:
+            return 999
+        marker = match.group(1)
+        if marker.isdigit():
+            return int(marker)
+        values = {
+            "零": 0,
+            "一": 1,
+            "二": 2,
+            "三": 3,
+            "四": 4,
+            "五": 5,
+            "六": 6,
+            "七": 7,
+            "八": 8,
+            "九": 9,
+            "十": 10,
+        }
+        if marker in values:
+            return values[marker]
+        if marker.startswith("十") and len(marker) == 2:
+            return 10 + values.get(marker[1], 0)
+        if marker.endswith("十") and len(marker) == 2:
+            return values.get(marker[0], 0) * 10
+        if "十" in marker:
+            left, right = marker.split("十", 1)
+            return values.get(left, 1) * 10 + values.get(right, 0)
+        return 999
+
+

    @staticmethod
    def _format_knowledge_heading_label(heading: str) -> str:
@@ -156,6 +213,169 @@ class UserAgentKnowledgeHelpersMixin:
        return " / ".join(parts)


+    @staticmethod
+    def _has_inline_numbered_knowledge_items(content: str) -> bool:
+        return len(
+            re.findall(
+                r"[（(][一二三四五六七八九十百零0-9]+[)）]",
+                str(content or ""),
+            )
+        ) >= 2
+
+
+    @staticmethod
+    def _split_inline_numbered_knowledge_items(content: str) -> list[str]:
+        normalized = str(content or "").strip()
+        if not UserAgentKnowledgeHelpersMixin._has_inline_numbered_knowledge_items(normalized):
+            return [normalized] if normalized else []
+
+        marker_pattern = r"[（(][一二三四五六七八九十百零0-9]+[)）]"
+        first_marker = re.search(marker_pattern, normalized)
+        if first_marker is None:
+            return [normalized] if normalized else []
+
+        prefix = normalized[: first_marker.start()].strip(" ：:")
+        tail = normalized[first_marker.start() :].strip()
+        item_pattern = (
+            r"([（(][一二三四五六七八九十百零0-9]+[)）]\s*.*?"
+            r"(?=\s*[（(][一二三四五六七八九十百零0-9]+[)）]|\s*$))"
+        )
+        items = [item.strip() for item in re.findall(item_pattern, tail) if item.strip()]
+        if prefix:
+            return [prefix, *items]
+        return items or [normalized]
+
+
+    @staticmethod
+    def _focus_knowledge_segment_content(content: str, query_terms: list[str]) -> str:
+        normalized = re.sub(r"\s+", " ", str(content or "").strip())
+        if not normalized:
+            return ""
+
+        anchor_terms = sorted(
+            {
+                str(term or "").strip()
+                for term in query_terms
+                if len(str(term or "").strip()) >= 3
+            },
+            key=len,
+            reverse=True,
+        )
+        anchor_index = -1
+        for term in anchor_terms:
+            anchor_index = normalized.lower().find(term.lower())
+            if anchor_index >= 0:
+                break
+        if anchor_index < 0:
+            return normalized
+
+        prefix_window = normalized[max(0, anchor_index - 40) : anchor_index]
+        marker_match = None
+        for match in re.finditer(
+            r"(?:第[一二三四五六七八九十百零0-9]+[部分章节条]|[一二三四五六七八九十]+、|[（(][一二三四五六七八九十百零0-9]+[)）])",
+            prefix_window,
+        ):
+            marker_match = match
+        start = anchor_index
+        if marker_match is not None:
+            start = max(0, anchor_index - len(prefix_window) + marker_match.start())
+
+        return normalized[start : start + 700].strip()
+
+
+    @staticmethod
+    def _split_markdown_table_cells(line: str) -> list[str]:
+        stripped = str(line or "").strip()
+        if stripped.startswith("|"):
+            stripped = stripped[1:]
+        if stripped.endswith("|"):
+            stripped = stripped[:-1]
+        return [
+            re.sub(r"\s+", " ", cell.replace("**", "").strip())
+            for cell in stripped.split("|")
+        ]
+
+
+    @classmethod
+    def _summarize_knowledge_table_preview(cls, preview: str) -> str:
+        rows: list[list[str]] = []
+        for line in str(preview or "").splitlines():
+            if line.count("|") < 2:
+                continue
+            cells = cls._split_markdown_table_cells(line)
+            if not cells or all(re.fullmatch(r":?-{2,}:?", cell.replace(" ", "")) for cell in cells):
+                continue
+            rows.append(cells)
+
+        if len(rows) < 2:
+            return "可直接参考的标准表如下。"
+
+        header = rows[0]
+        data_rows = [row for row in rows[1:] if len(row) == len(header)]
+        if len(data_rows) == 1 and len(header) >= 2:
+            row = data_rows[0]
+            subject = row[0] or "该项目"
+            pairs = [
+                f"{label}：{value}"
+                for label, value in zip(header[1:], row[1:])
+                if label and value and value not in {"-", "—"}
+            ]
+            if pairs:
+                return f"{subject}的标准为：{'；'.join(pairs)}。"
+
+        return "相关标准项如下，请按表头和行内容对应使用。"
+
+
+    def _summarize_knowledge_lines_conclusion(
+        self,
+        lines: list[str],
+        *,
+        heading: str = "",
+    ) -> str:
+        clean_lines = [
+            self._clean_knowledge_segment_text(line)
+            for line in lines
+            if self._clean_knowledge_segment_text(line)
+        ]
+        if not clean_lines:
+            return ""
+
+        clean_heading = str(heading or "").strip()
+        if not clean_heading and clean_lines and "：" not in clean_lines[0] and ":" not in clean_lines[0]:
+            clean_heading = clean_lines[0]
+        clean_heading = re.sub(
+            r"^[一二三四五六七八九十百零0-9]+、\s*",
+            "",
+            clean_heading,
+        )
+        item_labels: list[str] = []
+        for line in clean_lines:
+            if "：" not in line and ":" not in line:
+                continue
+            label = re.split(r"[：:]", line, maxsplit=1)[0].strip()
+            if 1 <= len(label) <= 24:
+                item_labels.append(label)
+
+        if clean_heading and len(item_labels) >= 2:
+            return f"{clean_heading}包括：{'、'.join(item_labels[:6])}。"
+        if item_labels:
+            return f"{item_labels[0]}：{clean_lines[0].split('：', 1)[-1].strip()}"
+        return clean_lines[0]
+
+
+    @staticmethod
+    def _knowledge_lines_have_multiple_labeled_items(lines: list[str]) -> bool:
+        labeled_count = 0
+        for line in lines:
+            normalized = str(line or "").strip()
+            if "：" not in normalized and ":" not in normalized:
+                continue
+            label = re.split(r"[：:]", normalized, maxsplit=1)[0].strip()
+            if 1 <= len(label) <= 24:
+                labeled_count += 1
+        return labeled_count >= 2
+
+

    def _score_knowledge_evidence_candidate(
        self,
@@ -169,10 +389,14 @@ class UserAgentKnowledgeHelpersMixin:

        matched_terms = [term for term in query_terms if term in haystack]
        score = len(matched_terms) * 10
+        score += sum(max(0, len(term) - 4) * 8 for term in matched_terms)
        score += sum(1 for term in matched_terms if term in heading) * 6
+        score += sum(max(0, len(term) - 4) * 6 for term in matched_terms if term in heading)

        if kind == "table":
            score += 10
+            if content.count("\n") < 2:
+                score -= 24
        elif kind in {"kv", "clause", "list"}:
            score += 8
        elif kind == "paragraph":
@@ -220,6 +444,30 @@ class UserAgentKnowledgeHelpersMixin:
            remember(item)

        for block in re.findall(r"[\u4e00-\u9fff]{2,20}", normalized_question):
+            remember(block)
+            if len(terms) >= MAX_KNOWLEDGE_QUERY_TERMS:
+                return terms
+            for marker in ("标准", "金额", "限额", "额度"):
+                marker_index = block.find(marker)
+                if marker_index <= 0:
+                    continue
+                subject = block[:marker_index]
+                for width in (6, 4, 3, 2):
+                    remember(subject[-width:])
+            for anchor in UserAgentKnowledgeHelpersMixin.KNOWLEDGE_QUERY_ANCHOR_TERMS:
+                if anchor in block:
+                    remember(anchor)
+            tail = block[-14:]
+            for size in (8, 7, 6, 5, 4):
+                for start in range(0, len(tail) - size + 1):
+                    piece = tail[start : start + size]
+                    if any(
+                        anchor in piece
+                        for anchor in UserAgentKnowledgeHelpersMixin.KNOWLEDGE_QUERY_ANCHOR_TERMS
+                    ):
+                        remember(piece)
+                        if len(terms) >= MAX_KNOWLEDGE_QUERY_TERMS:
+                            return terms
            if len(block) <= 4:
                remember(block)
                continue
@@ -276,7 +524,14 @@ class UserAgentKnowledgeHelpersMixin:


    @staticmethod
-    def _extract_relevant_table_preview(content: str, query_terms: list[str]) -> str:
+    def _extract_relevant_table_preview(
+        content: str,
+        query_terms: list[str],
+        *,
+        preferred_terms: list[str] | None = None,
+        max_rows: int = 3,
+        fallback_rows: int = 2,
+    ) -> str:
        lines = [line.strip() for line in str(content or "").splitlines() if line.strip()]
        if len(lines) <= 3:
            return "\n".join(lines)
@@ -285,12 +540,39 @@ class UserAgentKnowledgeHelpersMixin:
        divider = lines[1] if len(lines) > 1 else ""
        body = lines[2:] if divider.count("|") >= 2 else lines[1:]

+        preferred = [
+            str(term or "").strip().lower()
+            for term in list(preferred_terms or [])
+            if str(term or "").strip()
+        ]
+        base_terms = preferred + [
+            str(term or "").strip().lower()
+            for term in query_terms
+            if str(term or "").strip().lower() not in preferred
+        ]
+        derived_terms: list[str] = []
+        for term in base_terms:
+            for marker in ("标准", "金额", "限额", "额度", "是多少"):
+                marker_index = term.find(marker)
+                if marker_index <= 0:
+                    continue
+                subject = term[:marker_index].strip()
+                if len(subject) < 2:
+                    continue
+                for width in (6, 4, 3, 2):
+                    derived_terms.append(subject[-width:])
+
+        search_terms: list[str] = []
+        for term in [*preferred, *derived_terms, *base_terms]:
+            if term and term not in search_terms:
+                search_terms.append(term)
+
        matched_rows = [
            row
            for row in body
-            if any(term in row.lower() for term in query_terms)
+            if any(term in row.lower() for term in search_terms)
        ]
-        selected_rows = matched_rows[:3] or body[:2]
+        selected_rows = matched_rows[:max_rows] or body[:fallback_rows]
        preview_lines = [header]
        if divider:
            preview_lines.append(divider)
@@ -298,6 +580,18 @@ class UserAgentKnowledgeHelpersMixin:
        return "\n".join(preview_lines).strip()


+    @staticmethod
+    def _question_requests_broad_knowledge_table(question: str) -> bool:
+        normalized = str(question or "").strip()
+        if not normalized:
+            return False
+        broad_hints = ("有哪些", "是什么", "介绍", "说明", "列表", "清单", "全部", "完整")
+        table_subject_hints = ("科目", "目录", "清单", "列表", "表", "明细")
+        return any(hint in normalized for hint in broad_hints) and any(
+            hint in normalized for hint in table_subject_hints
+        )
+
+

    @staticmethod
    def _question_requires_explicit_condition(question: str) -> bool:
--- a/server/src/app/services/user_agent_response.py
+++ b/server/src/app/services/user_agent_response.py
@@ -261,7 +261,6 @@ class UserAgentResponseMixin:
            "draft_payload": draft_payload.model_dump(mode="json") if draft_payload is not None else None,
            "selected_capability_codes": payload.selected_capability_codes,
            "requires_confirmation": payload.requires_confirmation,
-            "fallback_answer": fallback_answer,
        }
        if payload.ontology.scenario == "knowledge":
            facts["knowledge_evidence_blocks"] = self._build_knowledge_evidence_blocks(
--- a/server/storage/knowledge/.index.json
+++ b/server/storage/knowledge/.index.json
@@ -57,8 +57,8 @@
      "uploaded_by": "系统导入",
      "version_number": 1,
      "ingest_status": 3,
-      "ingest_status_updated_at": "2026-05-22T09:22:52.110824+00:00",
-      "ingest_completed_at": "2026-05-22T09:22:52.110824+00:00",
+      "ingest_status_updated_at": "2026-05-23T14:30:33.605531+00:00",
+      "ingest_completed_at": "2026-05-23T14:30:33.605531+00:00",
      "ingest_document_name": "远光软件财务基础知识手册.docx",
      "ingest_document_updated_at": "2026-05-22T07:00:22.011016+00:00",
      "ingest_document_sha256": "",
--- a/server/storage/knowledge/.lightrag/x_financial_knowledge/graph_chunk_entity_relation.graphml
+++ b/server/storage/knowledge/.lightrag/x_financial_knowledge/graph_chunk_entity_relation.graphml
--- a/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_doc_status.json
+++ b/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_doc_status.json
@@ -82,23 +82,6 @@
      "processing_end_time": 1779441745
    }
  },
-  "b0277cd76034437997fbf5219662725a": {
-    "status": "processed",
-    "chunks_count": 1,
-    "chunks_list": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "content_summary": "远光软件股份有限公司\n财务基础知识手册\n第一部分  会计基础知识\n一、会计要素\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。\n会计恒等式：资产 = 负债 + 所有者权益\n二、常用会计科目\n科目类别\n科目名称\n说明\n资产类\n库存现金\n公司持有的现金\n资产类\n银行存款\n存放在银行的资金\n资产类\n应收账款\n因销售商品或提供劳务应收的款项\n资产类\n固定资产\n使用年限超过一年的有形资产\n负债类\n应付账款\n因购买商品或接受劳务应付的款项\n负债类\n应交税费\n应缴纳的各种税费\n负债类\n应付职工薪酬\n...",
-    "content_length": 1082,
-    "created_at": "2026-05-22T09:22:31.538281+00:00",
-    "updated_at": "2026-05-22T09:22:52.110824+00:00",
-    "file_path": "/app/server/storage/knowledge/财务知识库/远光软件财务基础知识手册.docx",
-    "track_id": "insert_20260522_092231_e1b9d415",
-    "metadata": {
-      "processing_start_time": 1779441751,
-      "processing_end_time": 1779441772
-    }
-  },
  "23f56f159a3e4bc3b2338056544120dd": {
    "status": "processed",
    "chunks_count": 1,
@@ -688,5 +671,23 @@
      "processing_start_time": 1779467725,
      "processing_end_time": 1779467727
    }
+  },
+  "b0277cd76034437997fbf5219662725a": {
+    "status": "processed",
+    "chunks_count": 2,
+    "chunks_list": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b",
+      "chunk-f894acfbb6c681d00f75cf9c486d491b"
+    ],
+    "content_summary": "远光软件股份有限公司 财务基础知识手册\n\n第一部分 会计基础知识\n\n一、会计要素\n\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。\n\n会计恒等式：资产 = 负债 + 所有者权益\n\n二、常用会计科目\n\n| 科目类别 | 科目名称 | 说明 |\n| --- | --- | --- |\n| 资产类 | 库存现金 | 公司持有的现金 |\n| 资产类 | 银行存款 | 存放在银行的资金 |\n| 资产类 | 应收账款 | 因销售商品或提供劳务应收的款项 |\n| 资产类 | 固定资产 | 使用年限超...",
+    "content_length": 1641,
+    "created_at": "2026-05-23T14:29:37.233751+00:00",
+    "updated_at": "2026-05-23T14:30:33.605531+00:00",
+    "file_path": "/app/server/storage/knowledge/财务知识库/远光软件财务基础知识手册.docx",
+    "track_id": "insert_20260523_142937_5cd25327",
+    "metadata": {
+      "processing_start_time": 1779546577,
+      "processing_end_time": 1779546633
+    }
  }
 }
--- a/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_entity_chunks.json
+++ b/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_entity_chunks.json
@@ -6,7 +6,6 @@
      "chunk-dd87aa5bc62cc9587ecb4c26d35a5263",
      "chunk-31ff57cf79d009c378478f065eda9d4d",
      "chunk-e726f44fb0287c5192cf61b350f18abb",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43",
      "chunk-2ee7e2a66cb544bdfe1b09e133863ad1",
      "chunk-2c8384b328272063de4dac306a52d21e",
      "chunk-6c549250b13b7728acb37eb6082bc178",
@@ -44,8 +43,9 @@
      "chunk-6175768b05adf2e7229c16f13ee7cffd",
      "chunk-cfac1ddf5942f8fe2d5a296380818faf"
    ],
-    "count": 42,
-    "update_time": 1779467739,
+    "count": 41,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "远光软件股份有限公司"
  },
  "第一章总则": {
@@ -3504,31 +3504,31 @@
  },
  "库存现金": {
    "chunk_ids": [
-      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
+      "chunk-31ff57cf79d009c378478f065eda9d4d"
    ],
-    "count": 2,
-    "update_time": 1779441756,
+    "count": 1,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "库存现金"
  },
  "银行存款": {
    "chunk_ids": [
-      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
+      "chunk-31ff57cf79d009c378478f065eda9d4d"
    ],
-    "count": 2,
-    "update_time": 1779441756,
+    "count": 1,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "银行存款"
  },
  "应收账款": {
    "chunk_ids": [
      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43",
      "chunk-6f1d6991d45799bc8ff24afaed39244d",
-      "chunk-af56151a803634f02e294f2d692fc1f0"
+      "chunk-af56151a803634f02e294f2d692fc1f0",
+      "chunk-f894acfbb6c681d00f75cf9c486d491b"
    ],
    "count": 4,
-    "update_time": 1779467128,
+    "update_time": 1779546632,
    "_id": "应收账款"
  },
  "其他应收款": {
@@ -3552,12 +3552,12 @@
  "固定资产": {
    "chunk_ids": [
      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43",
      "chunk-4e95fc3e38b2bf65fcb3f6f0664fd9df",
      "chunk-4287121b009a169fe4155526bfe413ea"
    ],
-    "count": 4,
-    "update_time": 1779465910,
+    "count": 3,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "固定资产"
  },
  "累计折旧": {
@@ -3572,39 +3572,39 @@
  "应付账款": {
    "chunk_ids": [
      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43",
      "chunk-6f1d6991d45799bc8ff24afaed39244d",
      "chunk-af56151a803634f02e294f2d692fc1f0"
    ],
-    "count": 4,
-    "update_time": 1779467128,
+    "count": 3,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "应付账款"
  },
  "应交税费": {
    "chunk_ids": [
-      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
+      "chunk-31ff57cf79d009c378478f065eda9d4d"
    ],
-    "count": 2,
-    "update_time": 1779441756,
+    "count": 1,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "应交税费"
  },
  "应付职工薪酬": {
    "chunk_ids": [
-      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
+      "chunk-31ff57cf79d009c378478f065eda9d4d"
    ],
-    "count": 2,
-    "update_time": 1779441756,
+    "count": 1,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "应付职工薪酬"
  },
  "主营业务收入": {
    "chunk_ids": [
-      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
+      "chunk-31ff57cf79d009c378478f065eda9d4d"
    ],
-    "count": 2,
-    "update_time": 1779441756,
+    "count": 1,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "主营业务收入"
  },
  "主营业务成本": {
@@ -3618,20 +3618,20 @@
  },
  "管理费用": {
    "chunk_ids": [
-      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
+      "chunk-31ff57cf79d009c378478f065eda9d4d"
    ],
-    "count": 2,
-    "update_time": 1779441756,
+    "count": 1,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "管理费用"
  },
  "销售费用": {
    "chunk_ids": [
-      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
+      "chunk-31ff57cf79d009c378478f065eda9d4d"
    ],
-    "count": 2,
-    "update_time": 1779441756,
+    "count": 1,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "销售费用"
  },
  "财务费用": {
@@ -3667,11 +3667,11 @@
  "资产类": {
    "chunk_ids": [
      "chunk-31ff57cf79d009c378478f065eda9d4d",
-      "chunk-e726f44fb0287c5192cf61b350f18abb"
+      "chunk-e726f44fb0287c5192cf61b350f18abb",
+      "chunk-f894acfbb6c681d00f75cf9c486d491b"
    ],
-    "count": 2,
-    "create_time": 1779441742,
-    "update_time": 1779441742,
+    "count": 3,
+    "update_time": 1779546632,
    "_id": "资产类"
  },
  "负债类": {
@@ -3811,192 +3811,48 @@
    "update_time": 1779441742,
    "_id": "1221其他应收款"
  },
-  "财务基础知识手册": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441755,
-    "update_time": 1779441755,
-    "_id": "财务基础知识手册"
-  },
-  "会计要素": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441755,
-    "update_time": 1779441755,
-    "_id": "会计要素"
-  },
-  "资产": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441755,
-    "update_time": 1779441755,
-    "_id": "资产"
-  },
-  "负债": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441755,
-    "update_time": 1779441755,
-    "_id": "负债"
-  },
-  "所有者权益": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441755,
-    "update_time": 1779441755,
-    "_id": "所有者权益"
-  },
-  "收入": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441755,
-    "update_time": 1779441755,
-    "_id": "收入"
-  },
-  "费用": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441755,
-    "update_time": 1779441755,
-    "_id": "费用"
-  },
-  "利润": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441756,
-    "update_time": 1779441756,
-    "_id": "利润"
-  },
-  "会计恒等式": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441756,
-    "update_time": 1779441756,
-    "_id": "会计恒等式"
-  },
  "增值税": {
    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43",
      "chunk-2c8384b328272063de4dac306a52d21e",
      "chunk-f61c91e28e8d0f773f83e3daf161ab1c",
      "chunk-570642e8a00db7819c2b4048ebf1b279",
      "chunk-89afdbbf904b60cf6494cba2638e08a8"
    ],
-    "count": 5,
-    "update_time": 1779467303,
-    "_id": "增值税"
-  },
-  "企业所得税": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43",
-      "chunk-2c8384b328272063de4dac306a52d21e",
-      "chunk-f61c91e28e8d0f773f83e3daf161ab1c"
-    ],
-    "count": 3,
-    "update_time": 1779466482,
-    "_id": "企业所得税"
-  },
-  "个人所得税": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43",
-      "chunk-f61c91e28e8d0f773f83e3daf161ab1c",
-      "chunk-570642e8a00db7819c2b4048ebf1b279",
-      "chunk-89afdbbf904b60cf6494cba2638e08a8"
-    ],
    "count": 4,
-    "update_time": 1779467303,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
+    "_id": "增值税"
+  },
+  "企业所得税": {
+    "chunk_ids": [
+      "chunk-2c8384b328272063de4dac306a52d21e",
+      "chunk-f61c91e28e8d0f773f83e3daf161ab1c"
+    ],
+    "count": 2,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
+    "_id": "企业所得税"
+  },
+  "个人所得税": {
+    "chunk_ids": [
+      "chunk-f61c91e28e8d0f773f83e3daf161ab1c",
+      "chunk-570642e8a00db7819c2b4048ebf1b279",
+      "chunk-89afdbbf904b60cf6494cba2638e08a8"
+    ],
+    "count": 3,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "个人所得税"
  },
  "印花税": {
    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43",
      "chunk-f61c91e28e8d0f773f83e3daf161ab1c"
    ],
-    "count": 2,
-    "update_time": 1779466482,
+    "count": 1,
+    "updated_at": 1779546575,
+    "update_time": 1779546575,
    "_id": "印花税"
  },
-  "三大财务报表": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441756,
-    "update_time": 1779441756,
-    "_id": "三大财务报表"
-  },
-  "资产负债表": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441756,
-    "update_time": 1779441756,
-    "_id": "资产负债表"
-  },
-  "利润表": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441756,
-    "update_time": 1779441756,
-    "_id": "利润表"
-  },
-  "现金流量表": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441756,
-    "update_time": 1779441756,
-    "_id": "现金流量表"
-  },
-  "会计基础知识": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441756,
-    "update_time": 1779441756,
-    "_id": "会计基础知识"
-  },
-  "税务基础知识": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441757,
-    "update_time": 1779441757,
-    "_id": "税务基础知识"
-  },
-  "财务报表解读": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441757,
-    "update_time": 1779441757,
-    "_id": "财务报表解读"
-  },
  "财务术语解释手册": {
    "chunk_ids": [
      "chunk-2ee7e2a66cb544bdfe1b09e133863ad1"
@@ -4181,10 +4037,11 @@
  "Corporate Income Tax": {
    "chunk_ids": [
      "chunk-93d2389cdb74257e90201dccbc3f6539",
-      "chunk-bdfd18ae478b23604f1318623e8e9508"
+      "chunk-bdfd18ae478b23604f1318623e8e9508",
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
    ],
-    "count": 2,
-    "update_time": 1779466433,
+    "count": 3,
+    "update_time": 1779546632,
    "_id": "Corporate Income Tax"
  },
  "Venture Capital Deduction": {
@@ -5896,11 +5753,11 @@
  },
  "Yuan Guang Software Co., Ltd.": {
    "chunk_ids": [
-      "chunk-96ab661ad24e0cb4c468128a58a76b6d"
+      "chunk-96ab661ad24e0cb4c468128a58a76b6d",
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
    ],
-    "count": 1,
-    "create_time": 1779465962,
-    "update_time": 1779465962,
+    "count": 2,
+    "update_time": 1779546631,
    "_id": "Yuan Guang Software Co., Ltd."
  },
  "Procurement Management Methods": {
@@ -12387,5 +12244,302 @@
    "create_time": 1779467740,
    "update_time": 1779467740,
    "_id": "第三部分发票问题"
+  },
+  "Accounting Elements": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Accounting Elements"
+  },
+  "Assets": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Assets"
+  },
+  "Liabilities": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Liabilities"
+  },
+  "Owner's Equity": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Owner's Equity"
+  },
+  "Revenue": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Revenue"
+  },
+  "Expenses": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Expenses"
+  },
+  "Profit": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Profit"
+  },
+  "Accounting Equation": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Accounting Equation"
+  },
+  "Common Accounting Items": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Common Accounting Items"
+  },
+  "Assets Category": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Assets Category"
+  },
+  "Liabilities Category": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Liabilities Category"
+  },
+  "Income And Expense Category": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Income And Expense Category"
+  },
+  "Cash": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Cash"
+  },
+  "Bank Deposits": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Bank Deposits"
+  },
+  "Accounts Receivable": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Accounts Receivable"
+  },
+  "Fixed Assets": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Fixed Assets"
+  },
+  "Accounts Payable": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Accounts Payable"
+  },
+  "Taxes Payable": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Taxes Payable"
+  },
+  "Employee Compensation Payable": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Employee Compensation Payable"
+  },
+  "Main Business Revenue": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Main Business Revenue"
+  },
+  "Management Expenses": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Management Expenses"
+  },
+  "Sales Expenses": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Sales Expenses"
+  },
+  "Tax Fundamentals": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546631,
+    "update_time": 1779546631,
+    "_id": "Tax Fundamentals"
+  },
+  "Major Tax Types": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "Major Tax Types"
+  },
+  "Value Added Tax": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "Value Added Tax"
+  },
+  "Individual Income Tax": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "Individual Income Tax"
+  },
+  "Stamp Duty": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "Stamp Duty"
+  },
+  "Software Services Tax Rate 6%": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "Software Services Tax Rate 6%"
+  },
+  "Software Product Sales Tax Rate 13%": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "Software Product Sales Tax Rate 13%"
+  },
+  "Corporate Income Tax Rate 25%": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "Corporate Income Tax Rate 25%"
+  },
+  "High-Tech Enterprise Preferential Tax Rate 15%": {
+    "chunk_ids": [
+      "chunk-6fdc554482754c7c662adc7804d3cf0b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "High-Tech Enterprise Preferential Tax Rate 15%"
+  },
+  "所有者权益": {
+    "chunk_ids": [
+      "chunk-f894acfbb6c681d00f75cf9c486d491b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "所有者权益"
+  },
+  "常用会计科目": {
+    "chunk_ids": [
+      "chunk-f894acfbb6c681d00f75cf9c486d491b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "常用会计科目"
  }
 }
--- a/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_full_docs.json
+++ b/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_full_docs.json
@@ -20,13 +20,6 @@
    "update_time": 1779441661,
    "_id": "c7601043d9944ef2bcf4d3f67ed253f7"
  },
-  "b0277cd76034437997fbf5219662725a": {
-    "content": "远光软件股份有限公司\n财务基础知识手册\n第一部分  会计基础知识\n一、会计要素\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。\n会计恒等式：资产 = 负债 + 所有者权益\n二、常用会计科目\n科目类别\n科目名称\n说明\n资产类\n库存现金\n公司持有的现金\n资产类\n银行存款\n存放在银行的资金\n资产类\n应收账款\n因销售商品或提供劳务应收的款项\n资产类\n固定资产\n使用年限超过一年的有形资产\n负债类\n应付账款\n因购买商品或接受劳务应付的款项\n负债类\n应交税费\n应缴纳的各种税费\n负债类\n应付职工薪酬\n应付给职工的工资、福利等\n损益类\n主营业务收入\n主要经营业务产生的收入\n损益类\n管理费用\n为管理生产经营发生的费用\n损益类\n销售费用\n为销售产品发生的费用\n第二部分  税务基础知识\n三、主要税种介绍\n（一）增值税：公司为一般纳税人，软件服务适用6%税率，软件产品销售适用13%税率。\n（二）企业所得税：税率为25%，高新技术企业享受15%优惠税率。\n（三）个人所得税：按累进税率3%-45%，由公司代扣代缴。\n（四）印花税：对经济活动中的应税凭证征收。\n第三部分  财务报表解读\n四、三大财务报表\n（一）资产负债表：反映企业在某一特定日期的财务状况。\n（二）利润表：反映企业在一定期间的经营成果。\n（三）现金流量表：反映企业在一定期间现金和现金等价物的流入和流出。\n\n# 章节导航\n\n以下内容由入库阶段从制度原文中提取，供检索时优先理解制度层级、条目和标准所在章节。\n\n- 一、会计要素\n- 二、常用会计科目\n- （四）印花税：对经济活动中的应税凭证征收。\n\n# 重点章节摘录\n\n## 一、会计要素\n\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。；会计恒等式：资产 = 负债 + 所有者权益\n\n## 二、常用会计科目\n\n科目类别；科目名称；说明\n\n## （四）印花税：对经济活动中的应税凭证征收。\n\n第三部分  财务报表解读\n\n# 问答线索补充\n\n以下内容由入库阶段根据章节标题、条款、列表、键值对与相邻正文提炼，供问答检索时优先命中更短、更直接的制度依据。\n\n- 一、会计要素：会计要素包括：资产、负债、所有者权益、收入、费用和利润\n- 一、会计要素：会计恒等式：资产 = 负债 + 所有者权益\n- 二、常用会计科目：因销售商品或提供劳务应收的款项\n- 二、常用会计科目：因购买商品或接受劳务应付的款项\n- 二、常用会计科目：应缴纳的各种税费\n- 二、常用会计科目：应付职工薪酬\n- （四）印花税：对经济活动中的应税凭证征收。：第三部分 财务报表解读",
-    "file_path": "/app/server/storage/knowledge/财务知识库/远光软件财务基础知识手册.docx",
-    "create_time": 1779441751,
-    "update_time": 1779441751,
-    "_id": "b0277cd76034437997fbf5219662725a"
-  },
  "23f56f159a3e4bc3b2338056544120dd": {
    "content": "远光软件股份有限公司\n财务术语解释手册\n权责发生制\n以权利和责任的发生来决定收入和费用归属期的会计基础。即凡是当期已经实现的收入和已经发生或应当负担的费用，不论款项是否收付，都应当作为当期的收入和费用。\n收付实现制\n以现金收到或付出为标准来记录收入的实现和费用的发生。即凡是当期收到和支付的现金，都作为当期的收入和费用。\n固定资产折旧\n固定资产在使用过程中因磨损而逐渐转移的价值。公司采用年限平均法计提折旧。\n摊销\n将无形资产或长期待摊费用按照规定期限分期计入当期损益的过程。\n增值税进项税额\n企业购进货物、接受应税劳务或应税服务支付的增值税额，可以从销项税额中抵扣。\n增值税销项税额\n企业销售货物、提供应税劳务或应税服务收取的增值税额。\n预算\n企业对未来一定时期内经营活动的数量化计划，包括收入预算、支出预算、资本预算等。\n现金流\n企业在一定期间内现金和现金等价物流入和流出的数量。\n毛利率\n毛利润占营业收入的百分比，反映企业产品或服务的初始盈利能力。计算公式：毛利率 = （营业收入 - 营业成本）/ 营业收入 × 100%\n净资产收益率（ROE）\n净利润占股东权益的百分比，反映股东投入资金的获利能力。计算公式：ROE = 净利润 / 股东权益 × 100%\n成本中心\n企业内部只发生成本费用而不产生收入的组织单位，用于成本核算和控制。\n利润中心\n企业内部既发生成本费用又产生收入的组织单位，用于考核盈利能力。\n\n# 问答线索补充\n\n以下内容由入库阶段根据章节标题、条款、列表、键值对与相邻正文提炼，供问答检索时优先命中更短、更直接的制度依据。\n\n- 正文：以权利和责任的发生来决定收入和费用归属期的会计基础\n- 正文：即凡是当期已经实现的收入和已经发生或应当负担的费用，不论款项是否收付，都应当作为当期的收入和费用\n- 正文：以现金收到或付出为标准来记录收入的实现和费用的发生\n- 正文：即凡是当期收到和支付的现金，都作为当期的收入和费用\n- 正文：企业购进货物、接受应税劳务或应税服务支付的增值税额，可以从销项税额中抵扣\n- 正文：企业销售货物、提供应税劳务或应税服务收取的增值税额\n- 正文：毛利润占营业收入的百分比，反映企业产品或服务的初始盈利能力\n- 正文：计算公式：毛利率 = （营业收入 - 营业成本）/ 营业收入 × 100%\n- 正文：净利润占股东权益的百分比，反映股东投入资金的获利能力\n- 正文：计算公式：ROE = 净利润 / 股东权益 × 100%",
    "file_path": "/app/server/storage/knowledge/财务知识库/远光软件财务术语解释手册.docx",
@@ -257,5 +250,12 @@
    "create_time": 1779467725,
    "update_time": 1779467725,
    "_id": "3acd9c2df63b4a438c7eab876269b25d"
+  },
+  "b0277cd76034437997fbf5219662725a": {
+    "content": "远光软件股份有限公司 财务基础知识手册\n\n第一部分 会计基础知识\n\n一、会计要素\n\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。\n\n会计恒等式：资产 = 负债 + 所有者权益\n\n二、常用会计科目\n\n| 科目类别 | 科目名称 | 说明 |\n| --- | --- | --- |\n| 资产类 | 库存现金 | 公司持有的现金 |\n| 资产类 | 银行存款 | 存放在银行的资金 |\n| 资产类 | 应收账款 | 因销售商品或提供劳务应收的款项 |\n| 资产类 | 固定资产 | 使用年限超过一年的有形资产 |\n| 负债类 | 应付账款 | 因购买商品或接受劳务应付的款项 |\n| 负债类 | 应交税费 | 应缴纳的各种税费 |\n| 负债类 | 应付职工薪酬 | 应付给职工的工资、福利等 |\n| 损益类 | 主营业务收入 | 主要经营业务产生的收入 |\n| 损益类 | 管理费用 | 为管理生产经营发生的费用 |\n| 损益类 | 销售费用 | 为销售产品发生的费用 |\n\n### 表格行级检索线索\n\n- 表格第 2 行：科目类别=资产类；科目名称=库存现金；说明=公司持有的现金\n\n- 表格第 3 行：科目类别=资产类；科目名称=银行存款；说明=存放在银行的资金\n\n- 表格第 4 行：科目类别=资产类；科目名称=应收账款；说明=因销售商品或提供劳务应收的款项\n\n- 表格第 5 行：科目类别=资产类；科目名称=固定资产；说明=使用年限超过一年的有形资产\n\n- 表格第 6 行：科目类别=负债类；科目名称=应付账款；说明=因购买商品或接受劳务应付的款项\n\n- 表格第 7 行：科目类别=负债类；科目名称=应交税费；说明=应缴纳的各种税费\n\n- 表格第 8 行：科目类别=负债类；科目名称=应付职工薪酬；说明=应付给职工的工资、福利等\n\n- 表格第 9 行：科目类别=损益类；科目名称=主营业务收入；说明=主要经营业务产生的收入\n\n- 表格第 10 行：科目类别=损益类；科目名称=管理费用；说明=为管理生产经营发生的费用\n\n- 表格第 11 行：科目类别=损益类；科目名称=销售费用；说明=为销售产品发生的费用\n\n第二部分 税务基础知识\n\n三、主要税种介绍\n\n（一）增值税：公司为一般纳税人，软件服务适用6%税率，软件产品销售适用13%税率。\n\n（二）企业所得税：税率为25%，高新技术企业享受15%优惠税率。\n\n（三）个人所得税：按累进税率3%-45%，由公司代扣代缴。\n\n（四）印花税：对经济活动中的应税凭证征收。\n\n第三部分 财务报表解读\n\n四、三大财务报表\n\n（一）资产负债表：反映企业在某一特定日期的财务状况。\n\n（二）利润表：反映企业在一定期间的经营成果。\n\n（三）现金流量表：反映企业在一定期间现金和现金等价物的流入和流出。\n\n# 章节导航\n\n以下内容由入库阶段从制度原文中提取，供检索时优先理解制度层级、条目和标准所在章节。\n\n- 一、会计要素\n- 二、常用会计科目\n\n# 重点章节摘录\n\n## 一、会计要素\n\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。；会计恒等式：资产 = 负债 + 所有者权益\n\n## 二、常用会计科目\n\n| 科目类别 | 科目名称 | 说明 |；| --- | --- | --- |；| 资产类 | 库存现金 | 公司持有的现金 |\n\n# 问答线索补充\n\n以下内容由入库阶段根据章节标题、条款、列表、键值对与相邻正文提炼，供问答检索时优先命中更短、更直接的制度依据。\n\n- 一、会计要素：会计要素包括：资产、负债、所有者权益、收入、费用和利润\n- 一、会计要素：会计恒等式：资产 = 负债 + 所有者权益\n- 二、常用会计科目：表格第 2 行：科目类别=资产类\n- 二、常用会计科目：表格第 3 行：科目类别=资产类\n- 二、常用会计科目：表格第 4 行：科目类别=资产类\n- 二、常用会计科目：科目名称=应收账款",
+    "file_path": "/app/server/storage/knowledge/财务知识库/远光软件财务基础知识手册.docx",
+    "create_time": 1779546577,
+    "update_time": 1779546577,
+    "_id": "b0277cd76034437997fbf5219662725a"
  }
 }
--- a/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_full_entities.json
+++ b/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_full_entities.json
@@ -432,45 +432,6 @@
    "update_time": 1779441745,
    "_id": "c7601043d9944ef2bcf4d3f67ed253f7"
  },
-  "b0277cd76034437997fbf5219662725a": {
-    "entity_names": [
-      "固定资产",
-      "财务报表解读",
-      "银行存款",
-      "收入",
-      "负债",
-      "现金流量表",
-      "企业所得税",
-      "三大财务报表",
-      "会计恒等式",
-      "库存现金",
-      "所有者权益",
-      "费用",
-      "财务基础知识手册",
-      "应付账款",
-      "利润表",
-      "会计基础知识",
-      "应收账款",
-      "应交税费",
-      "主营业务收入",
-      "资产",
-      "管理费用",
-      "税务基础知识",
-      "应付职工薪酬",
-      "销售费用",
-      "印花税",
-      "资产负债表",
-      "个人所得税",
-      "会计要素",
-      "远光软件股份有限公司",
-      "利润",
-      "增值税"
-    ],
-    "count": 31,
-    "create_time": 1779441772,
-    "update_time": 1779441772,
-    "_id": "b0277cd76034437997fbf5219662725a"
-  },
  "23f56f159a3e4bc3b2338056544120dd": {
    "entity_names": [
      "净利润",
@@ -1890,5 +1851,50 @@
    "create_time": 1779467727,
    "update_time": 1779467727,
    "_id": "3acd9c2df63b4a438c7eab876269b25d"
+  },
+  "b0277cd76034437997fbf5219662725a": {
+    "entity_names": [
+      "Fixed Assets",
+      "Stamp Duty",
+      "Liabilities",
+      "Revenue",
+      "Management Expenses",
+      "Accounts Receivable",
+      "Profit",
+      "Income And Expense Category",
+      "Software Product Sales Tax Rate 13%",
+      "Expenses",
+      "High-Tech Enterprise Preferential Tax Rate 15%",
+      "Software Services Tax Rate 6%",
+      "应收账款",
+      "所有者权益",
+      "Accounting Equation",
+      "资产类",
+      "Main Business Revenue",
+      "Accounts Payable",
+      "Employee Compensation Payable",
+      "Individual Income Tax",
+      "Liabilities Category",
+      "Value Added Tax",
+      "Bank Deposits",
+      "Common Accounting Items",
+      "常用会计科目",
+      "Owner's Equity",
+      "Accounting Elements",
+      "Assets Category",
+      "Corporate Income Tax Rate 25%",
+      "Yuan Guang Software Co., Ltd.",
+      "Assets",
+      "Taxes Payable",
+      "Corporate Income Tax",
+      "Tax Fundamentals",
+      "Cash",
+      "Sales Expenses",
+      "Major Tax Types"
+    ],
+    "count": 37,
+    "create_time": 1779546633,
+    "update_time": 1779546633,
+    "_id": "b0277cd76034437997fbf5219662725a"
  }
 }
--- a/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_full_relations.json
+++ b/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_full_relations.json
@@ -355,34 +355,6 @@
    "update_time": 1779441745,
    "_id": "c7601043d9944ef2bcf4d3f67ed253f7"
  },
-  "b0277cd76034437997fbf5219662725a": {
-    "relation_pairs": [
-      [
-        "会计要素",
-        "资产"
-      ],
-      [
-        "财务基础知识手册",
-        "远光软件股份有限公司"
-      ],
-      [
-        "财务基础知识手册",
-        "财务报表解读"
-      ],
-      [
-        "税务基础知识",
-        "财务基础知识手册"
-      ],
-      [
-        "会计基础知识",
-        "财务基础知识手册"
-      ]
-    ],
-    "count": 5,
-    "create_time": 1779441772,
-    "update_time": 1779441772,
-    "_id": "b0277cd76034437997fbf5219662725a"
-  },
  "23f56f159a3e4bc3b2338056544120dd": {
    "relation_pairs": [
      [
@@ -2194,5 +2166,21 @@
    "create_time": 1779467727,
    "update_time": 1779467727,
    "_id": "3acd9c2df63b4a438c7eab876269b25d"
+  },
+  "b0277cd76034437997fbf5219662725a": {
+    "relation_pairs": [
+      [
+        "应收账款",
+        "资产类"
+      ],
+      [
+        "常用会计科目",
+        "资产类"
+      ]
+    ],
+    "count": 2,
+    "create_time": 1779546633,
+    "update_time": 1779546633,
+    "_id": "b0277cd76034437997fbf5219662725a"
  }
 }
--- a/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_relation_chunks.json
+++ b/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_relation_chunks.json
@@ -746,51 +746,6 @@
    "update_time": 1779441745,
    "_id": "1221其他应收款<SEP>远光软件股份有限公司"
  },
-  "财务基础知识手册<SEP>远光软件股份有限公司": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441770,
-    "update_time": 1779441770,
-    "_id": "财务基础知识手册<SEP>远光软件股份有限公司"
-  },
-  "会计要素<SEP>资产": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441770,
-    "update_time": 1779441770,
-    "_id": "会计要素<SEP>资产"
-  },
-  "会计基础知识<SEP>财务基础知识手册": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441771,
-    "update_time": 1779441771,
-    "_id": "会计基础知识<SEP>财务基础知识手册"
-  },
-  "税务基础知识<SEP>财务基础知识手册": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441771,
-    "update_time": 1779441771,
-    "_id": "税务基础知识<SEP>财务基础知识手册"
-  },
-  "财务基础知识手册<SEP>财务报表解读": {
-    "chunk_ids": [
-      "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-    ],
-    "count": 1,
-    "create_time": 1779441771,
-    "update_time": 1779441771,
-    "_id": "财务基础知识手册<SEP>财务报表解读"
-  },
  "财务术语解释手册<SEP>远光软件股份有限公司": {
    "chunk_ids": [
      "chunk-2ee7e2a66cb544bdfe1b09e133863ad1"
@@ -4230,5 +4185,23 @@
    "create_time": 1779467726,
    "update_time": 1779467726,
    "_id": "第三部分发票问题<SEP>财务报销常见问题解答"
+  },
+  "常用会计科目<SEP>资产类": {
+    "chunk_ids": [
+      "chunk-f894acfbb6c681d00f75cf9c486d491b"
+    ],
+    "count": 1,
+    "create_time": 1779546632,
+    "update_time": 1779546632,
+    "_id": "常用会计科目<SEP>资产类"
+  },
+  "应收账款<SEP>资产类": {
+    "chunk_ids": [
+      "chunk-f894acfbb6c681d00f75cf9c486d491b"
+    ],
+    "count": 1,
+    "create_time": 1779546633,
+    "update_time": 1779546633,
+    "_id": "应收账款<SEP>资产类"
  }
 }
--- a/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_text_chunks.json
+++ b/server/storage/knowledge/.lightrag/x_financial_knowledge/kv_store_text_chunks.json
@@ -197,17 +197,6 @@
    "update_time": 1779441661,
    "_id": "chunk-e726f44fb0287c5192cf61b350f18abb"
  },
-  "chunk-78edb0c8ccc8238159196ecaeeb08d43": {
-    "tokens": 839,
-    "content": "远光软件股份有限公司\n财务基础知识手册\n第一部分  会计基础知识\n一、会计要素\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。\n会计恒等式：资产 = 负债 + 所有者权益\n二、常用会计科目\n科目类别\n科目名称\n说明\n资产类\n库存现金\n公司持有的现金\n资产类\n银行存款\n存放在银行的资金\n资产类\n应收账款\n因销售商品或提供劳务应收的款项\n资产类\n固定资产\n使用年限超过一年的有形资产\n负债类\n应付账款\n因购买商品或接受劳务应付的款项\n负债类\n应交税费\n应缴纳的各种税费\n负债类\n应付职工薪酬\n应付给职工的工资、福利等\n损益类\n主营业务收入\n主要经营业务产生的收入\n损益类\n管理费用\n为管理生产经营发生的费用\n损益类\n销售费用\n为销售产品发生的费用\n第二部分  税务基础知识\n三、主要税种介绍\n（一）增值税：公司为一般纳税人，软件服务适用6%税率，软件产品销售适用13%税率。\n（二）企业所得税：税率为25%，高新技术企业享受15%优惠税率。\n（三）个人所得税：按累进税率3%-45%，由公司代扣代缴。\n（四）印花税：对经济活动中的应税凭证征收。\n第三部分  财务报表解读\n四、三大财务报表\n（一）资产负债表：反映企业在某一特定日期的财务状况。\n（二）利润表：反映企业在一定期间的经营成果。\n（三）现金流量表：反映企业在一定期间现金和现金等价物的流入和流出。\n\n# 章节导航\n\n以下内容由入库阶段从制度原文中提取，供检索时优先理解制度层级、条目和标准所在章节。\n\n- 一、会计要素\n- 二、常用会计科目\n- （四）印花税：对经济活动中的应税凭证征收。\n\n# 重点章节摘录\n\n## 一、会计要素\n\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。；会计恒等式：资产 = 负债 + 所有者权益\n\n## 二、常用会计科目\n\n科目类别；科目名称；说明\n\n## （四）印花税：对经济活动中的应税凭证征收。\n\n第三部分  财务报表解读\n\n# 问答线索补充\n\n以下内容由入库阶段根据章节标题、条款、列表、键值对与相邻正文提炼，供问答检索时优先命中更短、更直接的制度依据。\n\n- 一、会计要素：会计要素包括：资产、负债、所有者权益、收入、费用和利润\n- 一、会计要素：会计恒等式：资产 = 负债 + 所有者权益\n- 二、常用会计科目：因销售商品或提供劳务应收的款项\n- 二、常用会计科目：因购买商品或接受劳务应付的款项\n- 二、常用会计科目：应缴纳的各种税费\n- 二、常用会计科目：应付职工薪酬\n- （四）印花税：对经济活动中的应税凭证征收。：第三部分 财务报表解读",
-    "chunk_order_index": 0,
-    "full_doc_id": "b0277cd76034437997fbf5219662725a",
-    "file_path": "/app/server/storage/knowledge/财务知识库/远光软件财务基础知识手册.docx",
-    "llm_cache_list": [],
-    "create_time": 1779441751,
-    "update_time": 1779441751,
-    "_id": "chunk-78edb0c8ccc8238159196ecaeeb08d43"
-  },
  "chunk-2ee7e2a66cb544bdfe1b09e133863ad1": {
    "tokens": 760,
    "content": "远光软件股份有限公司\n财务术语解释手册\n权责发生制\n以权利和责任的发生来决定收入和费用归属期的会计基础。即凡是当期已经实现的收入和已经发生或应当负担的费用，不论款项是否收付，都应当作为当期的收入和费用。\n收付实现制\n以现金收到或付出为标准来记录收入的实现和费用的发生。即凡是当期收到和支付的现金，都作为当期的收入和费用。\n固定资产折旧\n固定资产在使用过程中因磨损而逐渐转移的价值。公司采用年限平均法计提折旧。\n摊销\n将无形资产或长期待摊费用按照规定期限分期计入当期损益的过程。\n增值税进项税额\n企业购进货物、接受应税劳务或应税服务支付的增值税额，可以从销项税额中抵扣。\n增值税销项税额\n企业销售货物、提供应税劳务或应税服务收取的增值税额。\n预算\n企业对未来一定时期内经营活动的数量化计划，包括收入预算、支出预算、资本预算等。\n现金流\n企业在一定期间内现金和现金等价物流入和流出的数量。\n毛利率\n毛利润占营业收入的百分比，反映企业产品或服务的初始盈利能力。计算公式：毛利率 = （营业收入 - 营业成本）/ 营业收入 × 100%\n净资产收益率（ROE）\n净利润占股东权益的百分比，反映股东投入资金的获利能力。计算公式：ROE = 净利润 / 股东权益 × 100%\n成本中心\n企业内部只发生成本费用而不产生收入的组织单位，用于成本核算和控制。\n利润中心\n企业内部既发生成本费用又产生收入的组织单位，用于考核盈利能力。\n\n# 问答线索补充\n\n以下内容由入库阶段根据章节标题、条款、列表、键值对与相邻正文提炼，供问答检索时优先命中更短、更直接的制度依据。\n\n- 正文：以权利和责任的发生来决定收入和费用归属期的会计基础\n- 正文：即凡是当期已经实现的收入和已经发生或应当负担的费用，不论款项是否收付，都应当作为当期的收入和费用\n- 正文：以现金收到或付出为标准来记录收入的实现和费用的发生\n- 正文：即凡是当期收到和支付的现金，都作为当期的收入和费用\n- 正文：企业购进货物、接受应税劳务或应税服务支付的增值税额，可以从销项税额中抵扣\n- 正文：企业销售货物、提供应税劳务或应税服务收取的增值税额\n- 正文：毛利润占营业收入的百分比，反映企业产品或服务的初始盈利能力\n- 正文：计算公式：毛利率 = （营业收入 - 营业成本）/ 营业收入 × 100%\n- 正文：净利润占股东权益的百分比，反映股东投入资金的获利能力\n- 正文：计算公式：ROE = 净利润 / 股东权益 × 100%",
@@ -889,5 +878,27 @@
    "create_time": 1779467725,
    "update_time": 1779467725,
    "_id": "chunk-cfac1ddf5942f8fe2d5a296380818faf"
+  },
+  "chunk-6fdc554482754c7c662adc7804d3cf0b": {
+    "tokens": 1195,
+    "content": "远光软件股份有限公司 财务基础知识手册\n\n第一部分 会计基础知识\n\n一、会计要素\n\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。\n\n会计恒等式：资产 = 负债 + 所有者权益\n\n二、常用会计科目\n\n| 科目类别 | 科目名称 | 说明 |\n| --- | --- | --- |\n| 资产类 | 库存现金 | 公司持有的现金 |\n| 资产类 | 银行存款 | 存放在银行的资金 |\n| 资产类 | 应收账款 | 因销售商品或提供劳务应收的款项 |\n| 资产类 | 固定资产 | 使用年限超过一年的有形资产 |\n| 负债类 | 应付账款 | 因购买商品或接受劳务应付的款项 |\n| 负债类 | 应交税费 | 应缴纳的各种税费 |\n| 负债类 | 应付职工薪酬 | 应付给职工的工资、福利等 |\n| 损益类 | 主营业务收入 | 主要经营业务产生的收入 |\n| 损益类 | 管理费用 | 为管理生产经营发生的费用 |\n| 损益类 | 销售费用 | 为销售产品发生的费用 |\n\n### 表格行级检索线索\n\n- 表格第 2 行：科目类别=资产类；科目名称=库存现金；说明=公司持有的现金\n\n- 表格第 3 行：科目类别=资产类；科目名称=银行存款；说明=存放在银行的资金\n\n- 表格第 4 行：科目类别=资产类；科目名称=应收账款；说明=因销售商品或提供劳务应收的款项\n\n- 表格第 5 行：科目类别=资产类；科目名称=固定资产；说明=使用年限超过一年的有形资产\n\n- 表格第 6 行：科目类别=负债类；科目名称=应付账款；说明=因购买商品或接受劳务应付的款项\n\n- 表格第 7 行：科目类别=负债类；科目名称=应交税费；说明=应缴纳的各种税费\n\n- 表格第 8 行：科目类别=负债类；科目名称=应付职工薪酬；说明=应付给职工的工资、福利等\n\n- 表格第 9 行：科目类别=损益类；科目名称=主营业务收入；说明=主要经营业务产生的收入\n\n- 表格第 10 行：科目类别=损益类；科目名称=管理费用；说明=为管理生产经营发生的费用\n\n- 表格第 11 行：科目类别=损益类；科目名称=销售费用；说明=为销售产品发生的费用\n\n第二部分 税务基础知识\n\n三、主要税种介绍\n\n（一）增值税：公司为一般纳税人，软件服务适用6%税率，软件产品销售适用13%税率。\n\n（二）企业所得税：税率为25%，高新技术企业享受15%优惠税率。\n\n（三）个人所得税：按累进税率3%-45%，由公司代扣代缴。\n\n（四）印花税：对经济活动中的应税凭证征收。\n\n第三部分 财务报表解读\n\n四、三大财务报表\n\n（一）资产负债表：反映企业在某一特定日期的财务状况。\n\n（二）利润表：反映企业在一定期间的经营成果。\n\n（三）现金流量表：反映企业在一定期间现金和现金等价物的流入和流出。\n\n# 章节导航\n\n以下内容由入库阶段从制度原文中提取，供检索时优先理解制度层级、条目和标准所在章节。\n\n- 一、会计要素\n- 二、常用会计科目\n\n# 重点章节摘录\n\n## 一、会计要素\n\n会计要素包括：资产、负债、所有者权益、收入、费用和利润。；会计恒等式：资产 = 负债 + 所有者权益\n\n## 二、常用会计科目\n\n| 科目类别 | 科目名称 | 说明 |；| --- | --- | --- |；| 资产类 | 库存现金 | 公司持有的现金 |\n\n# 问答线索补充\n\n以下内容由入库阶段根据章节标题、条款、列表、键值对与相邻正文提炼，供问答检索时优先命中更短、更直接的制度依据。\n\n- 一、会计要素：会计要素包括：资产、负债、所有者权益、收入、费用和利润\n- 一、会计要素：会计恒等式：资产 = 负债 + 所有者权益\n- 二、常用会计科目：表格第 2 行：科目类别=资产类\n- 二、常用会计科目：表格第 3 行：科目类别=资产类\n- 二、常用会计科目：表格第 4 行：科目类别=资产类\n- 二、常用会计科目：科目名称=应收账款",
+    "chunk_order_index": 0,
+    "full_doc_id": "b0277cd76034437997fbf5219662725a",
+    "file_path": "/app/server/storage/knowledge/财务知识库/远光软件财务基础知识手册.docx",
+    "llm_cache_list": [],
+    "create_time": 1779546577,
+    "update_time": 1779546577,
+    "_id": "chunk-6fdc554482754c7c662adc7804d3cf0b"
+  },
+  "chunk-f894acfbb6c681d00f75cf9c486d491b": {
+    "tokens": 95,
+    "content": "所有者权益\n- 二、常用会计科目：表格第 2 行：科目类别=资产类\n- 二、常用会计科目：表格第 3 行：科目类别=资产类\n- 二、常用会计科目：表格第 4 行：科目类别=资产类\n- 二、常用会计科目：科目名称=应收账款",
+    "chunk_order_index": 1,
+    "full_doc_id": "b0277cd76034437997fbf5219662725a",
+    "file_path": "/app/server/storage/knowledge/财务知识库/远光软件财务基础知识手册.docx",
+    "llm_cache_list": [],
+    "create_time": 1779546577,
+    "update_time": 1779546577,
+    "_id": "chunk-f894acfbb6c681d00f75cf9c486d491b"
  }
 }
--- a/server/tests/test_knowledge_document_extractors.py
+++ b/server/tests/test_knowledge_document_extractors.py
@@ -5,6 +5,38 @@ from zipfile import ZipFile
 from app.services.knowledge_document_extractors import _extract_document_text_from_path


+def test_extract_docx_document_text_preserves_tables_as_markdown(tmp_path) -> None:
+    file_path = tmp_path / "financial-basic.docx"
+    _write_minimal_docx_with_table(
+        file_path,
+        paragraphs=[
+            "远光软件股份有限公司",
+            "财务基础知识手册",
+            "二、常用会计科目",
+        ],
+        table=[
+            ["科目类别", "科目名称", "说明"],
+            ["资产类", "库存现金", "公司持有的现金"],
+            ["负债类", "应付账款", "因购买商品或接受劳务应付的款项"],
+            ["损益类", "销售费用", "为销售产品发生的费用"],
+        ],
+    )
+
+    text = _extract_document_text_from_path(
+        file_path=file_path,
+        original_name="远光软件财务基础知识手册.docx",
+        mime_type="application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+    )
+
+    assert "二、常用会计科目" in text
+    assert "| 科目类别 | 科目名称 | 说明 |" in text
+    assert "| 资产类 | 库存现金 | 公司持有的现金 |" in text
+    assert "| 负债类 | 应付账款 | 因购买商品或接受劳务应付的款项 |" in text
+    assert "| 损益类 | 销售费用 | 为销售产品发生的费用 |" in text
+    assert "表格第 2 行：科目类别=资产类；科目名称=库存现金；说明=公司持有的现金" in text
+    assert "科目类别\n科目名称\n说明" not in text
+
+
 def test_extract_xlsx_document_text_builds_markdown_with_row_clues(tmp_path) -> None:
    file_path = tmp_path / "company-expense-rules.xlsx"
    _write_minimal_xlsx(
@@ -58,6 +90,39 @@ def test_extract_pptx_document_text_builds_markdown_slides(tmp_path) -> None:
    assert "- 发票、审批、预算三项要素必须齐全" in text


+def _write_minimal_docx_with_table(
+    file_path,
+    *,
+    paragraphs: list[str],
+    table: list[list[str]],
+) -> None:
+    paragraph_xml = "\n".join(f"<w:p>{_docx_text_run(text)}</w:p>" for text in paragraphs)
+    table_xml = (
+        "<w:tbl>"
+        + "".join(
+            "<w:tr>"
+            + "".join(f"<w:tc><w:p>{_docx_text_run(cell)}</w:p></w:tc>" for cell in row)
+            + "</w:tr>"
+            for row in table
+        )
+        + "</w:tbl>"
+    )
+    document_xml = f"""<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
+<w:document xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main">
+  <w:body>
+    {paragraph_xml}
+    {table_xml}
+  </w:body>
+</w:document>
+"""
+    with ZipFile(file_path, "w") as archive:
+        archive.writestr("word/document.xml", document_xml)
+
+
+def _docx_text_run(text: str) -> str:
+    return f"<w:r><w:t>{text}</w:t></w:r>"
+
+
 def _write_minimal_xlsx(file_path, *, sheet_name: str, rows: list[list[str]]) -> None:
    workbook_xml = f"""<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
 <workbook xmlns="http://schemas.openxmlformats.org/spreadsheetml/2006/main"
--- a/server/tests/test_knowledge_rag_service.py
+++ b/server/tests/test_knowledge_rag_service.py
@@ -58,6 +58,34 @@ def test_build_hits_boosts_query_term_matches() -> None:
    assert [item["candidate_id"] for item in hits] == ["ent-1", "travel-1"]


+def test_build_hits_keeps_long_query_anchor_terms_for_accounting_table() -> None:
+    hits = KnowledgeRagService._build_hits_from_query_data(
+        query="远光软件财务基础知识手册里的常用会计科目是什么？",
+        chunks=[
+            {
+                "chunk_id": "generic-1",
+                "file_path": "/tmp/doc-1__远光软件财务制度培训手册.docx",
+                "content": "远光软件股份有限公司财务培训内容，介绍费用报销和财务制度。",
+            },
+            {
+                "chunk_id": "accounts-1",
+                "file_path": "/tmp/doc-2__远光软件财务基础知识手册.docx",
+                "content": (
+                    "二、常用会计科目\n\n"
+                    "| 科目类别 | 科目名称 | 说明 |\n"
+                    "| --- | --- | --- |\n"
+                    "| 资产类 | 库存现金 | 公司持有的现金 |\n"
+                    "| 损益类 | 销售费用 | 为销售产品发生的费用 |"
+                ),
+            },
+        ],
+        entities=[],
+        limit=2,
+    )
+
+    assert [item["candidate_id"] for item in hits] == ["accounts-1", "generic-1"]
+
+
 def test_build_hits_prioritizes_answer_clue_appendix_for_rule_queries() -> None:
    hits = KnowledgeRagService._build_hits_from_query_data(
        query="报销时限是多少？",
--- a/server/tests/test_ontology_service.py
+++ b/server/tests/test_ontology_service.py
@@ -589,6 +589,66 @@ def test_semantic_ontology_service_covers_common_expense_scene_keywords(
        )


+def test_semantic_ontology_service_connects_expense_application_to_ontology() -> None:
+    session_factory = build_session_factory()
+    with session_factory() as db:
+        result = SemanticOntologyService(db).parse(
+            OntologyParseRequest(
+                query="申请2026-06-01 ~ 2026-06-03去北京做客户现场验收，差旅预算18000元",
+                user_id="pytest",
+                context_json={
+                    "document_type": "expense_application",
+                    "application_stage": "pre_approval",
+                    "entry_source": "documents_application",
+                },
+            )
+        )
+
+        assert result.scenario == "expense"
+        assert result.intent == "draft"
+        assert any(
+            item.type == "document_type" and item.normalized_value == "expense_application"
+            for item in result.entities
+        )
+        assert any(
+            item.type == "workflow_stage" and item.normalized_value == "pre_approval"
+            for item in result.entities
+        )
+        assert any(
+            item.field == "document_type" and item.value == "expense_application"
+            for item in result.constraints
+        )
+        assert any(
+            item.type == "expense_type" and item.normalized_value == "travel"
+            for item in result.entities
+        )
+
+
+def test_semantic_ontology_service_requires_attachment_for_meeting_application() -> None:
+    session_factory = build_session_factory()
+    with session_factory() as db:
+        result = SemanticOntologyService(db).parse(
+            OntologyParseRequest(
+                query="发起会务申请，2026-06-01 ~ 2026-06-02上海产品发布会，预算32000元",
+                user_id="pytest",
+                context_json={
+                    "document_type": "expense_application",
+                    "application_stage": "pre_approval",
+                    "entry_source": "documents_application",
+                    "attachment_count": 0,
+                },
+            )
+        )
+
+        assert result.scenario == "expense"
+        assert result.intent == "draft"
+        assert any(
+            item.type == "expense_type" and item.normalized_value == "meeting"
+            for item in result.entities
+        )
+        assert "attachments" in result.missing_slots
+
+
 def test_semantic_ontology_service_uses_model_parse_when_available(monkeypatch) -> None:
    session_factory = build_session_factory()
    with session_factory() as db:
--- a/server/tests/test_risk_rule_generation.py
+++ b/server/tests/test_risk_rule_generation.py
@@ -1,18 +1,32 @@
 from __future__ import annotations

 import json
+from datetime import UTC, datetime
+from decimal import Decimal

 from sqlalchemy import create_engine
 from sqlalchemy.orm import Session, sessionmaker
 from sqlalchemy.pool import StaticPool

-from app.core.agent_enums import AgentAssetDomain, AgentAssetStatus
+from app.core.agent_enums import AgentAssetDomain, AgentAssetStatus, AgentReviewStatus
 from app.db.base import Base
 from app.models.agent_asset import AgentAsset
-from app.schemas.agent_asset import AgentAssetRiskRuleGenerateRequest
+from app.models.financial_record import ExpenseClaim
+from app.schemas.agent_asset import (
+    AgentAssetReviewCreate,
+    AgentAssetRiskRuleGenerateRequest,
+    AgentAssetRiskRuleReportRequest,
+    AgentAssetRiskRuleSampleTestRequest,
+    AgentAssetRiskRuleScenarioTestRequest,
+    AgentAssetRiskRuleSimulationRequest,
+)
 from app.services.agent_asset_rule_library import AgentAssetRuleLibraryManager
 from app.services.agent_asset_spreadsheet import RISK_RULES_LIBRARY
-from app.services.risk_rule_flow_diagram import RiskRuleFlowDiagramRenderer, RiskRuleFlowDiagramSpec
+from app.services.agent_assets import AgentAssetService
+from app.services.risk_rule_flow_diagram import (
+    RiskRuleFlowDiagramRenderer,
+    RiskRuleFlowDiagramSpec,
+)
 from app.services.risk_rule_generation import RiskRuleGenerationService


@@ -43,6 +57,7 @@ def test_generate_risk_rule_asset_creates_draft_json_rule(tmp_path) -> None:
        asset_id = service.generate_rule_asset(
            AgentAssetRiskRuleGenerateRequest(
                business_domain=AgentAssetDomain.EXPENSE,
+                expense_category="travel",
                risk_level="high",
                natural_language="住宿城市必须出现在本次差旅行程城市中，否则提示高风险。",
            ),
@@ -54,12 +69,18 @@ def test_generate_risk_rule_asset_creates_draft_json_rule(tmp_path) -> None:
        assert asset.status == AgentAssetStatus.DRAFT.value
        assert asset.config_json["detail_mode"] == "json_risk"
        assert asset.config_json["evaluator"] == "template_rule"
+        assert asset.config_json["expense_category"] == "travel"
+        assert asset.config_json["risk_category"] == "差旅费"
+        assert asset.scenario_json == ["差旅费"]
        assert asset.current_version == "v0.1.0"

        file_name = asset.config_json["rule_document"]["file_name"]
        rule_path = tmp_path / "rules" / RISK_RULES_LIBRARY / file_name
        payload = json.loads(rule_path.read_text(encoding="utf-8"))
        assert payload["rule_code"] == asset.code
+        assert payload["applies_to"]["expense_categories"] == ["travel"]
+        assert payload["risk_category"] == "差旅费"
+        assert payload["metadata"]["expense_category"] == "travel"
        assert payload["outcomes"]["fail"]["severity"] == "high"
        assert payload["template_key"] == "field_compare_v1"
        assert payload["metadata"]["natural_language"].startswith("住宿城市")
@@ -104,3 +125,206 @@ def test_risk_rule_flow_diagram_uses_risk_level_palette() -> None:
    assert "#dc2626" in high_svg
    assert high_svg.count("#dc2626") == 1
    assert "#10a37f" not in high_svg
+
+
+def test_risk_rule_requires_test_report_before_review_and_publish(tmp_path) -> None:
+    with build_session() as db:
+        manager = AgentAssetRuleLibraryManager(rule_root=tmp_path / "rules")
+        generator = RiskRuleGenerationService(
+            db,
+            rule_library_manager=manager,
+            runtime_chat_service=NullRuntimeChatService(),
+        )
+        asset_id = generator.generate_rule_asset(
+            AgentAssetRiskRuleGenerateRequest(
+                business_domain=AgentAssetDomain.EXPENSE,
+                risk_level="high",
+                natural_language="酒店发票城市必须与行程城市一致，不一致时标记高风险。",
+            ),
+            actor="pytest",
+        )
+        service = AgentAssetService(db)
+        service.rule_library_manager = manager
+
+        asset = db.get(AgentAsset, asset_id)
+        assert asset is not None
+        try:
+            service.create_review(
+                asset_id,
+                AgentAssetReviewCreate(
+                    version=asset.working_version or "v0.1.0",
+                    reviewer="manager",
+                    review_status=AgentReviewStatus.PENDING,
+                    review_note="送审",
+                ),
+                actor="pytest",
+            )
+        except PermissionError as exc:
+            assert "测试通过" in str(exc)
+        else:
+            raise AssertionError("未测试通过的风险规则不应允许提交审核")
+
+        simulation = service.simulate_risk_rule_message(
+            asset_id,
+            AgentAssetRiskRuleSimulationRequest(
+                message="我想仿真一张酒店报销单，酒店发票城市上海，申报目的地北京，金额580元。",
+            ),
+        )
+        assert simulation.execution_mode == "risk_rule_simulation"
+        assert simulation.ready is True
+        assert simulation.hit is True
+        assert simulation.severity == "high"
+        assert "不创建业务单据" in simulation.summary
+        assert service.get_latest_risk_rule_test_summary(asset_id).sample is None
+
+        blocked_simulation = service.simulate_risk_rule_message(
+            asset_id,
+            AgentAssetRiskRuleSimulationRequest(
+                message="请识别上传单据是否命中风险规则。",
+                attachments=[{"name": "hotel-invoice.pdf", "content_type": "application/pdf"}],
+            ),
+        )
+        assert blocked_simulation.ready is False
+        assert blocked_simulation.stage == "needs_recognition"
+        assert blocked_simulation.hit is False
+        assert "尚未完成识别" in blocked_simulation.summary
+
+        db.add(
+            ExpenseClaim(
+                claim_no="TEST-CLAIM-001",
+                employee_name="张三",
+                department_name="财务部",
+                expense_type="住宿费",
+                reason="北京出差住宿",
+                location="北京",
+                amount=Decimal("300.00"),
+                currency="CNY",
+                invoice_count=0,
+                occurred_at=datetime.now(UTC),
+                created_at=datetime.now(UTC),
+                status="draft",
+            )
+        )
+        db.commit()
+
+        sample = service.run_risk_rule_sample_test(
+            asset_id,
+            AgentAssetRiskRuleSampleTestRequest(),
+            actor="pytest",
+        )
+        assert sample.passed is True
+
+        scenario = service.run_risk_rule_scenario_test(
+            asset_id,
+            AgentAssetRiskRuleScenarioTestRequest(intent="用最近30天的住宿报销单试运行"),
+            actor="pytest",
+        )
+        assert scenario.passed is True
+        assert scenario.result_json["total_count"] == 1
+
+        report = service.confirm_risk_rule_test_report(
+            asset_id,
+            AgentAssetRiskRuleReportRequest(confirm_passed=True),
+            actor="pytest",
+        )
+        assert report.passed is True
+
+        review = service.create_review(
+            asset_id,
+            AgentAssetReviewCreate(
+                version=asset.working_version or "v0.1.0",
+                reviewer="manager",
+                review_status=AgentReviewStatus.PENDING,
+                review_note="送审",
+            ),
+            actor="pytest",
+        )
+        assert review.review_status == AgentReviewStatus.PENDING.value
+        published = service.publish_risk_rule(asset_id, actor="manager")
+        assert published.status == AgentAssetStatus.ACTIVE.value
+        assert published.published_version == asset.working_version
+
+        disabled = service.set_risk_rule_enabled(
+            asset_id,
+            enabled=False,
+            actor="manager",
+        )
+        assert disabled.config_json["enabled"] is False
+        rule_document = disabled.config_json["rule_document"]
+        manifest = manager.read_rule_library_json(
+            library=RISK_RULES_LIBRARY,
+            file_name=rule_document["file_name"],
+        )
+        assert manifest["enabled"] is False
+
+        attachment_required_id = generator.generate_rule_asset(
+            AgentAssetRiskRuleGenerateRequest(
+                business_domain=AgentAssetDomain.EXPENSE,
+                risk_level="medium",
+                natural_language="发票号码不能为空，缺失时进入中风险复核。",
+                requires_attachment=True,
+            ),
+            actor="pytest",
+        )
+        attachment_required_asset = db.get(AgentAsset, attachment_required_id)
+        assert attachment_required_asset is not None
+        assert attachment_required_asset.config_json["requires_attachment"] is True
+        attachment_rule_document = attachment_required_asset.config_json["rule_document"]
+        attachment_manifest = manager.read_rule_library_json(
+            library=RISK_RULES_LIBRARY,
+            file_name=attachment_rule_document["file_name"],
+        )
+        assert attachment_manifest["requires_attachment"] is True
+        no_attachment_simulation = service.simulate_risk_rule_message(
+            attachment_required_id,
+            AgentAssetRiskRuleSimulationRequest(message="请测试这条规则。"),
+        )
+        assert no_attachment_simulation.ready is False
+        assert no_attachment_simulation.stage == "needs_attachment"
+
+        attachment_only_simulation = service.simulate_risk_rule_message(
+            attachment_required_id,
+            AgentAssetRiskRuleSimulationRequest(
+                message="请识别上传单据是否命中风险规则。",
+                attachments=[
+                    {
+                        "name": "invoice.pdf",
+                        "content_type": "application/pdf",
+                        "document_fields": [
+                            {"key": "invoice_no", "label": "发票号码", "value": "INV-001"}
+                        ],
+                    }
+                ],
+            ),
+        )
+        assert attachment_only_simulation.ready is False
+        assert attachment_only_simulation.stage == "needs_test_intent"
+
+
+def test_delete_unpublished_risk_rule_removes_asset_and_json_file(tmp_path) -> None:
+    with build_session() as db:
+        manager = AgentAssetRuleLibraryManager(rule_root=tmp_path / "rules")
+        asset_id = RiskRuleGenerationService(
+            db,
+            rule_library_manager=manager,
+            runtime_chat_service=NullRuntimeChatService(),
+        ).generate_rule_asset(
+            AgentAssetRiskRuleGenerateRequest(
+                business_domain=AgentAssetDomain.EXPENSE,
+                risk_level="medium",
+                natural_language="报销事由不能为空，缺失时进入中风险复核。",
+            ),
+            actor="pytest",
+        )
+        asset = db.get(AgentAsset, asset_id)
+        assert asset is not None
+        file_name = asset.config_json["rule_document"]["file_name"]
+        rule_path = tmp_path / "rules" / RISK_RULES_LIBRARY / file_name
+        assert rule_path.exists()
+
+        service = AgentAssetService(db)
+        service.rule_library_manager = manager
+        service.delete_unpublished_asset(asset_id, actor="pytest")
+
+        assert db.get(AgentAsset, asset_id) is None
+        assert not rule_path.exists()
--- a/server/tests/test_user_agent_service.py
+++ b/server/tests/test_user_agent_service.py
@@ -131,6 +131,8 @@ def test_user_agent_knowledge_prompt_enforces_knowledge_boundary() -> None:
        assert "不能用常识、外部知识或主观推断补齐缺失条件" in messages[0]["content"]
        assert "不能只依赖排在最前面的片段" in messages[0]["content"]
        assert "不能把第一列的数值直接套给后面的列名" in messages[0]["content"]
+        assert "最终答复必须像助手在认真回答问题" in messages[0]["content"]
+        assert "禁止使用“已命中”“答案整理阶段”“稍后重试”" in messages[0]["content"]
        assert "knowledge_evidence_blocks" in messages[0]["content"]
        assert '"knowledge_answer_evidence": []' in messages[1]["content"]

@@ -162,8 +164,9 @@ def test_user_agent_knowledge_fallback_is_honest_and_personalized() -> None:
        )

        assert answer.startswith("张三，您好。")
-        assert "答案整理阶段本轮没有及时返回" in answer
-        assert "先给你当前最直接的依据" in answer
+        assert "我先根据当前制度依据给出可以确认的部分" in answer
+        assert "已命中" not in answer
+        assert "答案整理阶段本轮没有及时返回" not in answer
        assert "《差旅费制度》" in answer


@@ -241,6 +244,40 @@ def test_user_agent_prefers_relevant_raw_hit_over_generic_appendix() -> None:
    assert "组织人事部" in selected[0]["content"]


+def test_user_agent_model_hit_selection_keeps_later_relevant_hits() -> None:
+    selected = UserAgentService._select_knowledge_model_hits(
+        {
+            "hits": [
+                {"content": "一般说明一"},
+                {"content": "一般说明二"},
+                {"content": "一般说明三"},
+                {"content": "一般说明四"},
+                {"content": "一般说明五"},
+                {"content": "一般说明六"},
+                {"content": "一般说明七"},
+                {
+                    "content": (
+                        "# 问答线索补充\n\n"
+                        "- 第二章 报销时限：差旅费应在行程结束三个月内提交；逾期不予报销出差补贴。"
+                    )
+                },
+            ]
+        },
+        question="差旅费报销时限是多少？",
+    )
+
+    assert "三个月内提交" in selected[0]["content"]
+
+
+def test_user_agent_knowledge_terms_keep_accounting_subject_in_long_query() -> None:
+    terms = UserAgentService._extract_knowledge_query_terms(
+        "远光软件财务基础知识手册里的常用会计科目是什么？"
+    )
+
+    assert "常用会计科目" in terms
+    assert "会计科目" in terms
+
+
 def test_user_agent_uses_fast_knowledge_answer_without_model(monkeypatch) -> None:
    session_factory = build_session_factory()
    with session_factory() as db:
@@ -286,12 +323,170 @@ def test_user_agent_uses_fast_knowledge_answer_without_model(monkeypatch) -> Non
        )

        assert response.answer.startswith("张三，您好。")
-        assert "当前能直接确认的是" in response.answer
+        assert "**结论**" in response.answer
        assert "30 日内提交报销申请" in response.answer
        assert "## 依据" not in response.answer
        assert "答案整理阶段本轮没有及时返回" not in response.answer


+def test_user_agent_fast_knowledge_answer_focuses_inline_section_items() -> None:
+    session_factory = build_session_factory()
+    with session_factory() as db:
+        ontology = SemanticOntologyService(db).parse(
+            OntologyParseRequest(
+                query="主要税种介绍",
+                user_id="pytest",
+                context_json={"session_type": "knowledge"},
+            )
+        )
+        service = UserAgentService(db)
+
+        answer = service._build_fast_knowledge_answer(
+            UserAgentRequest(
+                run_id=ontology.run_id,
+                user_id="pytest",
+                message="主要税种介绍",
+                ontology=ontology,
+                context_json={
+                    "session_type": "knowledge",
+                    "user_input_text": "主要税种介绍",
+                },
+                tool_payload={
+                    "result_type": "knowledge_search",
+                    "hits": [
+                        {
+                            "title": "财务基础知识",
+                            "content": (
+                                "资产类 银行存款 企业存放在银行的款项 负债类 应付账款 "
+                                "因购买商品或接受劳务应付的款项 负债类 应交税费 应缴纳的各种税费 "
+                                "第二部分 税务基础知识 三、主要税种介绍 "
+                                "（一）增值税：公司为一般纳税人，软件服务适用6%税率，软件产品销售适用13%税率。 "
+                                "（二）企业所得税：税率为25%，高新技术企业享受15%优惠税率。 "
+                                "（三）个人所得税：员工工资薪金由公司代扣代缴。 "
+                                "（四）印花税：购销合同、账簿等按规定缴纳。"
+                            ),
+                        }
+                    ],
+                },
+            ),
+            citations=[],
+        )
+
+        assert answer is not None
+        assert "主要税种介绍包括：增值税、企业所得税、个人所得税、印花税" in answer
+        assert "软件服务适用6%税率" in answer
+        assert "软件产品销售适用13%税率" in answer
+        assert "高新技术企业享受15%优惠税率" in answer
+        assert "员工工资薪金由公司代扣代缴" in answer
+        assert "购销合同、账簿等按规定缴纳" in answer
+        assert "应付账款" not in answer
+        assert "银行存款" not in answer
+
+
+def test_user_agent_fast_knowledge_answer_summarizes_financial_statements() -> None:
+    session_factory = build_session_factory()
+    with session_factory() as db:
+        ontology = SemanticOntologyService(db).parse(
+            OntologyParseRequest(
+                query="三大财务报表 是什么？",
+                user_id="pytest",
+                context_json={"session_type": "knowledge"},
+            )
+        )
+        service = UserAgentService(db)
+
+        answer = service._build_fast_knowledge_answer(
+            UserAgentRequest(
+                run_id=ontology.run_id,
+                user_id="pytest",
+                message="三大财务报表 是什么？",
+                ontology=ontology,
+                context_json={
+                    "session_type": "knowledge",
+                    "user_input_text": "三大财务报表 是什么？",
+                },
+                tool_payload={
+                    "result_type": "knowledge_search",
+                    "hits": [
+                        {
+                            "title": "财务基础知识",
+                            "content": (
+                                "第三部分 财务报表解读 四、三大财务报表 "
+                                "（一）资产负债表：反映企业在某一特定日期的财务状况。 "
+                                "（二）利润表：反映企业在一定期间的经营成果。 "
+                                "（三）现金流量表：反映企业在一定期间现金和现金等价物的流入和流出。"
+                            ),
+                        }
+                    ],
+                },
+            ),
+            citations=[],
+        )
+
+        assert answer is not None
+        assert "三大财务报表包括：资产负债表、利润表、现金流量表" in answer
+        assert "资产负债表：反映企业在某一特定日期的财务状况" in answer
+        assert "利润表：反映企业在一定期间的经营成果" in answer
+        assert "现金流量表：反映企业在一定期间现金和现金等价物的流入和流出" in answer
+        assert "第三部分 财务报表解读" not in answer
+
+
+def test_user_agent_fast_knowledge_answer_expands_broad_accounting_table() -> None:
+    session_factory = build_session_factory()
+    with session_factory() as db:
+        ontology = SemanticOntologyService(db).parse(
+            OntologyParseRequest(
+                query="常用会计科目是什么？",
+                user_id="pytest",
+                context_json={"session_type": "knowledge"},
+            )
+        )
+        service = UserAgentService(db)
+
+        answer = service._build_fast_knowledge_answer(
+            UserAgentRequest(
+                run_id=ontology.run_id,
+                user_id="pytest",
+                message="常用会计科目是什么？",
+                ontology=ontology,
+                context_json={
+                    "session_type": "knowledge",
+                    "user_input_text": "常用会计科目是什么？",
+                },
+                tool_payload={
+                    "result_type": "knowledge_search",
+                    "hits": [
+                        {
+                            "title": "财务基础知识",
+                            "content": (
+                                "二、常用会计科目\n\n"
+                                "| 科目类别 | 科目名称 | 说明 |\n"
+                                "| --- | --- | --- |\n"
+                                "| 资产类 | 库存现金 | 公司持有的现金 |\n"
+                                "| 资产类 | 银行存款 | 存放在银行的资金 |\n"
+                                "| 资产类 | 应收账款 | 因销售商品或提供劳务应收的款项 |\n"
+                                "| 资产类 | 固定资产 | 使用年限超过一年的有形资产 |\n"
+                                "| 负债类 | 应付账款 | 因购买商品或接受劳务应付的款项 |\n"
+                                "| 负债类 | 应交税费 | 应缴纳的各种税费 |\n"
+                                "| 负债类 | 应付职工薪酬 | 应付给职工的工资、福利等 |\n"
+                                "| 损益类 | 主营业务收入 | 主要经营业务产生的收入 |\n"
+                                "| 损益类 | 管理费用 | 为管理生产经营发生的费用 |\n"
+                                "| 损益类 | 销售费用 | 为销售产品发生的费用 |\n"
+                            ),
+                        }
+                    ],
+                },
+            ),
+            citations=[],
+        )
+
+        assert answer is not None
+        assert "| 科目类别 | 科目名称 | 说明 |" in answer
+        assert "| 资产类 | 库存现金 | 公司持有的现金 |" in answer
+        assert "| 负债类 | 应付职工薪酬 | 应付给职工的工资、福利等 |" in answer
+        assert "| 损益类 | 销售费用 | 为销售产品发生的费用 |" in answer
+
+
 def test_user_agent_fast_knowledge_answer_renders_relevant_table_preview() -> None:
    session_factory = build_session_factory()
    with session_factory() as db:
@@ -337,9 +532,65 @@ def test_user_agent_fast_knowledge_answer_renders_relevant_table_preview() -> No
        assert answer is not None
        assert "| 项目 | 港澳台 | 其他地区 | 国外 |" in answer
        assert "| 餐补 | 75 | 55 | 140 |" in answer
+        assert "餐补的标准为" in answer
        assert "## 依据" not in answer


+def test_user_agent_fast_knowledge_answer_uses_user_grade_for_table_row() -> None:
+    session_factory = build_session_factory()
+    with session_factory() as db:
+        ontology = SemanticOntologyService(db).parse(
+            OntologyParseRequest(
+                query="我的住宿费标准是多少？",
+                user_id="pytest",
+                context_json={"session_type": "knowledge"},
+            )
+        )
+        service = UserAgentService(db)
+
+        answer = service._build_fast_knowledge_answer(
+            UserAgentRequest(
+                run_id=ontology.run_id,
+                user_id="pytest",
+                message="我的住宿费标准是多少？",
+                ontology=ontology,
+                context_json={
+                    "name": "张三",
+                    "grade": "P5",
+                    "position": "实施经理",
+                    "session_type": "knowledge",
+                    "user_input_text": "我的住宿费标准是多少？",
+                },
+                tool_payload={
+                    "result_type": "knowledge_search",
+                    "hits": [
+                        {
+                            "title": "费用报销制度",
+                            "content": (
+                                "# 结构化表格补充\n\n"
+                                "## 国内住宿限额标准\n\n"
+                                "| 职级 | 直辖市/特区/港澳台 | 省会城市 | 其他地区 |\n"
+                                "| --- | --- | --- | --- |\n"
+                                "| 公司领导（P8及以上） | 800 | 500 | 400 |\n"
+                                "| 高层经理（P7） | 700 | 450 | 400 |\n"
+                                "| 中层经理、基层经理（P4～P6、外聘专家） | 600 | 400 | 350 |\n"
+                                "| 其他员工 | 500 | 350 | 300 |\n"
+                            ),
+                        }
+                    ],
+                },
+            ),
+            citations=[],
+        )
+
+        assert answer is not None
+        assert answer.startswith("张三，您好。")
+        assert "中层经理、基层经理（P4～P6、外聘专家）的标准为" in answer
+        assert "| 中层经理、基层经理（P4～P6、外聘专家） | 600 | 400 | 350 |" in answer
+        assert "| 公司领导（P8及以上） | 800 | 500 | 400 |" not in answer
+        assert "| 高层经理（P7） | 700 | 450 | 400 |" not in answer
+
+
 def test_user_agent_fast_knowledge_answer_notes_missing_location_grounding() -> None:
    session_factory = build_session_factory()
    with session_factory() as db:
@@ -384,6 +635,7 @@ def test_user_agent_fast_knowledge_answer_notes_missing_location_grounding() ->

        assert answer is not None
        assert "没有直接写出“北京”对应的地区档位或映射关系" in answer
+        assert "**说明**" in answer
        assert "## 依据" not in answer


@@ -429,7 +681,7 @@ def test_user_agent_fast_knowledge_answer_expands_lead_in_list_items() -> None:
        )

        assert answer is not None
-        assert "当前能直接确认的是" in answer
+        assert "**结论**" in answer
        assert "登机牌、高速道路通行记录" in answer
        assert "支付记录" in answer
        assert "出差审批邮件、短信、微信等" in answer
--- a/web/src/assets/styles/app.css
+++ b/web/src/assets/styles/app.css
@@ -94,6 +94,7 @@
  grid-template-rows: auto minmax(0, 1fr);
  overflow: hidden;
 }
+.main.documents-main,
 .main.requests-main,
 .main.approval-main,
 .main.archive-main,
@@ -114,6 +115,7 @@
 }
 .workarea { min-height: 0; overflow: auto; padding: 24px; }
 .workarea.requests-workarea,
+.workarea.documents-workarea,
 .workarea.approval-workarea,
 .workarea.archive-workarea,
 .workarea.policies-workarea,
--- a/web/src/assets/styles/components/expense-application-dialog.css
+++ b/web/src/assets/styles/components/expense-application-dialog.css
@@ -0,0 +1,330 @@
+.expense-application-mask {
+  position: fixed;
+  inset: 0;
+  z-index: 120;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  padding: 28px;
+  background: rgba(15, 23, 42, 0.26);
+  backdrop-filter: blur(10px);
+}
+
+.expense-application-dialog {
+  width: min(1040px, calc(100vw - 56px));
+  max-height: calc(100vh - 56px);
+  display: flex;
+  flex-direction: column;
+  overflow: hidden;
+  border: 1px solid #dbe5ee;
+  border-radius: 8px;
+  background: #fff;
+  box-shadow: 0 24px 60px rgba(15, 23, 42, 0.18);
+}
+
+.application-dialog-header,
+.application-dialog-footer {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  padding: 18px 22px;
+  border-bottom: 1px solid #e8eef4;
+}
+
+.application-dialog-footer {
+  justify-content: flex-end;
+  gap: 12px;
+  border-top: 1px solid #e8eef4;
+  border-bottom: 0;
+}
+
+.application-dialog-eyebrow {
+  display: block;
+  margin-bottom: 4px;
+  color: #059669;
+  font-size: 12px;
+  font-weight: 800;
+}
+
+.application-dialog-header h2 {
+  margin: 0;
+  color: #102033;
+  font-size: 20px;
+  font-weight: 850;
+}
+
+.dialog-icon-btn {
+  width: 36px;
+  height: 36px;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  border: 1px solid #dbe5ee;
+  border-radius: 8px;
+  background: #fff;
+  color: #4b5f73;
+  cursor: pointer;
+}
+
+.application-dialog-body {
+  display: grid;
+  grid-template-columns: minmax(320px, 0.92fr) minmax(420px, 1.08fr);
+  gap: 18px;
+  min-height: 0;
+  padding: 20px 22px;
+  overflow: auto;
+  background: #f7fafc;
+}
+
+.application-input-panel,
+.application-ontology-panel {
+  border: 1px solid #e1e9f0;
+  border-radius: 8px;
+  background: #fff;
+  padding: 18px;
+}
+
+.application-field-label {
+  display: block;
+  margin-bottom: 10px;
+  color: #425466;
+  font-size: 13px;
+  font-weight: 800;
+}
+
+.application-input-panel textarea {
+  width: 100%;
+  min-height: 132px;
+  resize: vertical;
+  border: 1px solid #cfd9e3;
+  border-radius: 8px;
+  padding: 12px;
+  color: #182536;
+  font: inherit;
+  line-height: 1.55;
+  outline: none;
+}
+
+.application-input-panel textarea:focus {
+  border-color: #10b981;
+  box-shadow: 0 0 0 3px rgba(16, 185, 129, 0.12);
+}
+
+.application-example-row {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 8px;
+  margin: 14px 0;
+}
+
+.application-example-chip {
+  border: 1px solid #dbe5ee;
+  border-radius: 8px;
+  background: #f8fbfd;
+  color: #395066;
+  padding: 7px 10px;
+  font-size: 12px;
+  cursor: pointer;
+}
+
+.primary-parse-btn,
+.confirm-btn,
+.secondary-btn {
+  min-height: 38px;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 8px;
+  border-radius: 8px;
+  padding: 0 16px;
+  font-weight: 850;
+  cursor: pointer;
+}
+
+.primary-parse-btn,
+.confirm-btn {
+  border: 0;
+  background: #059669;
+  color: #fff;
+}
+
+.primary-parse-btn {
+  width: 100%;
+}
+
+.secondary-btn {
+  border: 1px solid #dbe5ee;
+  background: #fff;
+  color: #395066;
+}
+
+.primary-parse-btn:disabled,
+.confirm-btn:disabled {
+  background: #b7c5d1;
+  cursor: not-allowed;
+}
+
+.application-error {
+  margin: 12px 0 0;
+  color: #dc2626;
+  font-size: 13px;
+}
+
+.ontology-panel-head {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  margin-bottom: 14px;
+  color: #425466;
+  font-size: 13px;
+  font-weight: 850;
+}
+
+.ontology-panel-head strong {
+  color: #059669;
+}
+
+.ontology-empty-state {
+  min-height: 300px;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  justify-content: center;
+  gap: 12px;
+  color: #6b7c8f;
+  text-align: center;
+}
+
+.ontology-empty-state i {
+  color: #94a3b8;
+  font-size: 34px;
+}
+
+.ontology-empty-state p {
+  max-width: 360px;
+  margin: 0;
+  line-height: 1.6;
+}
+
+.ontology-chip-row {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 8px;
+  margin-bottom: 14px;
+}
+
+.ontology-chip {
+  border-radius: 8px;
+  background: #eefbf5;
+  color: #047857;
+  padding: 6px 10px;
+  font-size: 12px;
+  font-weight: 800;
+}
+
+.application-field-grid {
+  display: grid;
+  grid-template-columns: repeat(2, minmax(0, 1fr));
+  gap: 10px;
+}
+
+.application-field-card {
+  min-height: 76px;
+  border: 1px solid #e4ebf2;
+  border-radius: 8px;
+  padding: 12px;
+  background: #fbfdff;
+}
+
+.application-field-card span,
+.application-reason-block span,
+.application-missing-block span {
+  display: block;
+  margin-bottom: 7px;
+  color: #6b7c8f;
+  font-size: 12px;
+  font-weight: 800;
+}
+
+.application-field-card strong {
+  color: #182536;
+  font-size: 15px;
+}
+
+.application-policy-strip,
+.application-reason-block,
+.application-missing-block {
+  margin-top: 12px;
+  border: 1px solid #dbe5ee;
+  border-radius: 8px;
+  padding: 12px;
+  background: #fff;
+}
+
+.application-policy-strip {
+  display: flex;
+  gap: 12px;
+  align-items: flex-start;
+}
+
+.application-policy-strip i {
+  margin-top: 2px;
+  color: #059669;
+  font-size: 20px;
+}
+
+.application-policy-strip.required {
+  border-color: #fed7aa;
+  background: #fff7ed;
+}
+
+.application-policy-strip.required i {
+  color: #ea580c;
+}
+
+.application-policy-strip strong {
+  color: #182536;
+}
+
+.application-policy-strip p,
+.application-reason-block p {
+  margin: 5px 0 0;
+  color: #4b5f73;
+  line-height: 1.55;
+}
+
+.application-missing-block div {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 8px;
+}
+
+.application-missing-block em {
+  border-radius: 8px;
+  background: #fff7ed;
+  color: #c2410c;
+  padding: 5px 8px;
+  font-style: normal;
+  font-size: 12px;
+  font-weight: 800;
+}
+
+@media (max-width: 860px) {
+  .expense-application-mask {
+    align-items: stretch;
+    padding: 14px;
+  }
+
+  .expense-application-dialog {
+    width: 100%;
+    max-height: calc(100vh - 28px);
+  }
+
+  .application-dialog-body {
+    grid-template-columns: 1fr;
+  }
+
+  .application-field-grid {
+    grid-template-columns: 1fr;
+  }
+}
--- a/web/src/assets/styles/components/risk-rule-test-dialog.css
+++ b/web/src/assets/styles/components/risk-rule-test-dialog.css
@@ -0,0 +1,741 @@
+.risk-sim-backdrop {
+  position: fixed;
+  inset: 0;
+  z-index: 1200;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  padding: 18px;
+  background: rgba(15, 23, 42, 0.44);
+  backdrop-filter: blur(12px);
+}
+
+.risk-sim-modal {
+  width: min(1180px, 100%);
+  height: min(820px, calc(100vh - 36px));
+  display: grid;
+  grid-template-rows: auto auto minmax(0, 1fr) auto;
+  overflow: hidden;
+  border: 1px solid rgba(203, 213, 225, 0.88);
+  border-radius: 18px;
+  background: #f8fafc;
+  box-shadow: 0 26px 70px rgba(15, 23, 42, 0.26);
+}
+
+.risk-sim-head {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 18px;
+  padding: 18px 20px 14px;
+  border-bottom: 1px solid #e2e8f0;
+  background: linear-gradient(180deg, #ffffff 0%, #f8fafc 100%);
+}
+
+.risk-sim-title span,
+.risk-sim-context-panel span,
+.risk-sim-result-head span,
+.risk-sim-evidence span,
+.risk-sim-file-strip > span {
+  color: #64748b;
+  font-size: 12px;
+  font-weight: 850;
+}
+
+.risk-sim-title h3 {
+  margin: 4px 0 5px;
+  color: #0f172a;
+  font-size: 19px;
+  font-weight: 900;
+  line-height: 1.25;
+}
+
+.risk-sim-title p {
+  margin: 0;
+  color: #475569;
+  font-size: 13px;
+  line-height: 1.55;
+}
+
+.risk-sim-icon-btn,
+.risk-sim-tool-btn,
+.risk-sim-send-btn {
+  display: grid;
+  place-items: center;
+  border: 1px solid #dbe5ef;
+  background: #fff;
+  color: #475569;
+  cursor: pointer;
+}
+
+.risk-sim-icon-btn {
+  width: 38px;
+  height: 38px;
+  border-radius: 12px;
+}
+
+.risk-sim-meta {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  flex-wrap: wrap;
+  padding: 10px 20px;
+  border-bottom: 1px solid #e5edf5;
+  background: #fff;
+}
+
+.risk-sim-meta span {
+  min-height: 28px;
+  display: inline-flex;
+  align-items: center;
+  padding: 0 10px;
+  border: 1px solid #e2e8f0;
+  border-radius: 999px;
+  background: #f8fafc;
+  color: #475569;
+  font-size: 12px;
+  font-weight: 800;
+}
+
+.risk-sim-meta .tone-low {
+  border-color: #bfdbfe;
+  background: #eff6ff;
+  color: #1d4ed8;
+}
+
+.risk-sim-meta .tone-medium {
+  border-color: #fed7aa;
+  background: #fff7ed;
+  color: #c2410c;
+}
+
+.risk-sim-meta .tone-high {
+  border-color: #fecaca;
+  background: #fef2f2;
+  color: #b91c1c;
+}
+
+.risk-sim-main {
+  min-height: 0;
+  display: grid;
+  grid-template-columns: minmax(0, 1fr) 300px;
+  gap: 14px;
+  padding: 14px;
+}
+
+.risk-sim-dialog-panel,
+.risk-sim-context-panel {
+  min-height: 0;
+  border: 1px solid #dfe8f1;
+  border-radius: 14px;
+  background: #fff;
+}
+
+.risk-sim-dialog-panel {
+  display: grid;
+  grid-template-rows: minmax(0, 1fr) auto auto;
+  overflow: hidden;
+}
+
+.risk-sim-message-list {
+  min-height: 0;
+  display: grid;
+  align-content: start;
+  gap: 14px;
+  overflow-y: auto;
+  padding: 18px;
+  background: #f8fafc;
+}
+
+.risk-sim-message-row {
+  display: grid;
+  grid-template-columns: 34px minmax(0, 1fr);
+  gap: 10px;
+  align-items: start;
+}
+
+.risk-sim-message-row.user {
+  grid-template-columns: minmax(0, 1fr) 34px;
+}
+
+.risk-sim-message-row.user .risk-sim-avatar {
+  grid-column: 2;
+  grid-row: 1;
+}
+
+.risk-sim-message-row.user .risk-sim-bubble {
+  grid-column: 1;
+  justify-self: end;
+  background: #0f766e;
+  color: #fff;
+}
+
+.risk-sim-message-row.user .risk-sim-bubble header,
+.risk-sim-message-row.user .risk-sim-bubble p {
+  color: #fff;
+}
+
+.risk-sim-avatar {
+  width: 34px;
+  height: 34px;
+  display: grid;
+  place-items: center;
+  border: 1px solid #dbe5ef;
+  border-radius: 12px;
+  background: #fff;
+  color: #0f766e;
+}
+
+.risk-sim-message-row.user .risk-sim-avatar {
+  color: #2563eb;
+}
+
+.risk-sim-bubble {
+  max-width: min(100%, 760px);
+  padding: 11px 13px;
+  border: 1px solid #e2e8f0;
+  border-radius: 14px;
+  background: #fff;
+  box-shadow: 0 8px 18px rgba(15, 23, 42, 0.05);
+}
+
+.risk-sim-bubble header {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  margin-bottom: 6px;
+  color: #64748b;
+  font-size: 11px;
+}
+
+.risk-sim-bubble header strong {
+  color: #0f172a;
+  font-size: 12px;
+  font-weight: 850;
+}
+
+.risk-sim-bubble p {
+  margin: 0;
+  color: #1e293b;
+  font-size: 13px;
+  line-height: 1.65;
+  white-space: pre-wrap;
+}
+
+.risk-sim-thinking {
+  display: inline-flex;
+  align-items: center;
+  gap: 8px;
+}
+
+.risk-sim-message-files,
+.risk-sim-file-strip div {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  flex-wrap: wrap;
+}
+
+.risk-sim-message-files {
+  margin-top: 9px;
+}
+
+.risk-sim-message-files span,
+.risk-sim-file-chip {
+  min-width: 0;
+  max-width: 260px;
+  display: inline-flex;
+  align-items: center;
+  gap: 5px;
+  min-height: 28px;
+  padding: 0 9px;
+  border: 1px solid rgba(191, 219, 254, 0.8);
+  border-radius: 999px;
+  background: #eff6ff;
+  color: #1d4ed8;
+  font-size: 12px;
+  font-weight: 750;
+}
+
+.risk-sim-message-files span {
+  color: #dbeafe;
+  border-color: rgba(219, 234, 254, 0.38);
+  background: rgba(255, 255, 255, 0.12);
+}
+
+.risk-sim-result-card {
+  margin-top: 12px;
+  overflow: hidden;
+  border: 1px solid #e2e8f0;
+  border-radius: 12px;
+  background: #fff;
+}
+
+.risk-sim-result-card.high {
+  border-color: #fecaca;
+}
+
+.risk-sim-result-card.medium {
+  border-color: #fed7aa;
+}
+
+.risk-sim-result-card.low {
+  border-color: #bfdbfe;
+}
+
+.risk-sim-result-head {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+  padding: 11px 12px;
+  border-bottom: 1px solid #edf2f7;
+}
+
+.risk-sim-result-head strong {
+  display: block;
+  margin-top: 3px;
+  color: #0f172a;
+  font-size: 14px;
+  font-weight: 900;
+}
+
+.risk-sim-result-head b {
+  padding: 5px 9px;
+  border-radius: 999px;
+  background: #f1f5f9;
+  color: #475569;
+  font-size: 12px;
+}
+
+.risk-sim-result-card.high .risk-sim-result-head b {
+  background: #fef2f2;
+  color: #b91c1c;
+}
+
+.risk-sim-result-card.medium .risk-sim-result-head b {
+  background: #fff7ed;
+  color: #c2410c;
+}
+
+.risk-sim-result-card.low .risk-sim-result-head b {
+  background: #eff6ff;
+  color: #1d4ed8;
+}
+
+.risk-sim-result-message {
+  padding: 10px 12px 0;
+}
+
+.risk-sim-blocking-message {
+  padding: 10px 12px 0;
+  color: #92400e;
+}
+
+.risk-sim-field-grid {
+  display: grid;
+  grid-template-columns: repeat(2, minmax(0, 1fr));
+  gap: 8px;
+  padding: 12px;
+}
+
+.risk-sim-field-grid div {
+  min-width: 0;
+  padding: 9px 10px;
+  border: 1px solid #edf2f7;
+  border-radius: 10px;
+  background: #f8fafc;
+}
+
+.risk-sim-field-grid span {
+  display: block;
+  color: #64748b;
+  font-size: 11px;
+  font-weight: 800;
+}
+
+.risk-sim-field-grid strong {
+  display: block;
+  margin-top: 4px;
+  overflow: hidden;
+  color: #0f172a;
+  font-size: 12px;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
+.risk-sim-evidence {
+  display: grid;
+  gap: 6px;
+  padding: 0 12px 12px;
+}
+
+.risk-sim-missing-fields {
+  display: flex;
+  gap: 7px;
+  flex-wrap: wrap;
+  padding: 0 12px 12px;
+}
+
+.risk-sim-missing-fields span {
+  flex: 0 0 100%;
+  color: #92400e;
+  font-size: 12px;
+  font-weight: 850;
+}
+
+.risk-sim-missing-fields b {
+  min-height: 26px;
+  display: inline-flex;
+  align-items: center;
+  padding: 0 8px;
+  border: 1px solid #fed7aa;
+  border-radius: 999px;
+  background: #fff7ed;
+  color: #c2410c;
+  font-size: 11px;
+  font-weight: 850;
+}
+
+.risk-sim-evidence ul {
+  display: grid;
+  gap: 5px;
+  margin: 0;
+  padding-left: 18px;
+  color: #334155;
+  font-size: 12px;
+  line-height: 1.55;
+}
+
+.risk-sim-file-strip {
+  display: grid;
+  gap: 8px;
+  padding: 10px 14px;
+  border-top: 1px solid #e2e8f0;
+  background: #fff;
+}
+
+.risk-sim-file-chip {
+  cursor: pointer;
+}
+
+.risk-sim-file-chip span {
+  min-width: 0;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
+.risk-sim-file-chip em {
+  color: inherit;
+  font-size: 11px;
+  font-style: normal;
+  opacity: 0.78;
+  white-space: nowrap;
+}
+
+.risk-sim-file-chip.pending {
+  border-color: #dbe5ef;
+  background: #f8fafc;
+  color: #475569;
+}
+
+.risk-sim-file-chip.recognizing {
+  border-color: #dbeafe;
+  background: #eff6ff;
+  color: #2563eb;
+}
+
+.risk-sim-file-chip.recognized {
+  border-color: #bbf7d0;
+  background: #ecfdf5;
+  color: #047857;
+}
+
+.risk-sim-file-chip.failed {
+  border-color: #fed7aa;
+  background: #fff7ed;
+  color: #c2410c;
+}
+
+.risk-sim-composer {
+  display: grid;
+  grid-template-columns: 42px minmax(0, 1fr) 42px;
+  gap: 10px;
+  align-items: center;
+  padding: 12px 14px 14px;
+  border-top: 1px solid #e2e8f0;
+  background: #fff;
+}
+
+.risk-sim-composer.text-only {
+  grid-template-columns: minmax(0, 1fr) 42px;
+}
+
+.risk-sim-file-input {
+  display: none;
+}
+
+.risk-sim-tool-btn,
+.risk-sim-send-btn {
+  width: 42px;
+  height: 42px;
+  border-radius: 999px;
+}
+
+.risk-sim-send-btn {
+  border-color: #0f766e;
+  background: #0f766e;
+  color: #fff;
+}
+
+.risk-sim-tool-btn:disabled,
+.risk-sim-send-btn:disabled,
+.risk-sim-primary-btn:disabled,
+.risk-sim-secondary-btn:disabled {
+  opacity: 0.48;
+  cursor: not-allowed;
+}
+
+.risk-sim-composer-shell {
+  min-height: 42px;
+  display: flex;
+  align-items: center;
+  border: 1px solid #cbd5e1;
+  border-radius: 999px;
+  background: #fff;
+}
+
+.risk-sim-composer-shell:focus-within {
+  border-color: rgba(15, 118, 110, 0.58);
+  box-shadow: 0 0 0 3px rgba(15, 118, 110, 0.1);
+}
+
+.risk-sim-composer-shell textarea {
+  width: 100%;
+  min-height: 34px;
+  max-height: 110px;
+  resize: none;
+  border: 0;
+  padding: 8px 14px;
+  background: transparent;
+  color: #0f172a;
+  font-size: 13px;
+  line-height: 1.45;
+}
+
+.risk-sim-composer-shell textarea:focus {
+  outline: none;
+}
+
+.risk-sim-context-panel {
+  display: grid;
+  align-content: start;
+  gap: 12px;
+  padding: 14px;
+}
+
+.risk-sim-context-panel section {
+  display: grid;
+  gap: 7px;
+  padding: 12px;
+  border: 1px solid #edf2f7;
+  border-radius: 12px;
+  background: #f8fafc;
+}
+
+.risk-sim-step-list,
+.risk-sim-recognition-list {
+  display: grid;
+  gap: 9px;
+}
+
+.risk-sim-step {
+  display: grid;
+  grid-template-columns: 28px minmax(0, 1fr);
+  gap: 9px;
+  align-items: flex-start;
+  padding: 9px;
+  border: 1px solid #e2e8f0;
+  border-radius: 10px;
+  background: #fff;
+}
+
+.risk-sim-step i {
+  width: 28px;
+  height: 28px;
+  display: grid;
+  place-items: center;
+  border-radius: 10px;
+  background: #f1f5f9;
+  color: #64748b;
+}
+
+.risk-sim-step.done i {
+  background: #ecfdf5;
+  color: #047857;
+}
+
+.risk-sim-step.running i {
+  background: #eff6ff;
+  color: #2563eb;
+}
+
+.risk-sim-step.warning i {
+  background: #fff7ed;
+  color: #c2410c;
+}
+
+.risk-sim-step strong {
+  color: #0f172a;
+  font-size: 12px;
+  font-weight: 900;
+}
+
+.risk-sim-step p,
+.risk-sim-recognition-list p,
+.risk-sim-recognition-list small {
+  margin: 2px 0 0;
+  color: #64748b;
+  font-size: 11px;
+  line-height: 1.45;
+}
+
+.risk-sim-recognition-list article {
+  min-width: 0;
+  padding: 9px;
+  border: 1px solid #e2e8f0;
+  border-radius: 10px;
+  background: #fff;
+}
+
+.risk-sim-recognition-list article strong {
+  display: block;
+  color: #0f172a;
+  font-size: 12px;
+  font-weight: 900;
+}
+
+.risk-sim-recognition-list small {
+  display: block;
+}
+
+.risk-sim-error-text {
+  color: #c2410c;
+}
+
+.risk-sim-context-panel strong {
+  color: #0f172a;
+  font-size: 14px;
+  font-weight: 900;
+}
+
+.risk-sim-context-panel p,
+.risk-sim-field-list em {
+  margin: 0;
+  color: #475569;
+  font-size: 12px;
+  line-height: 1.55;
+}
+
+.risk-sim-field-list {
+  display: flex;
+  gap: 7px;
+  flex-wrap: wrap;
+}
+
+.risk-sim-field-list b {
+  max-width: 100%;
+  min-height: 26px;
+  display: inline-flex;
+  align-items: center;
+  padding: 0 8px;
+  overflow: hidden;
+  border: 1px solid #dbe5ef;
+  border-radius: 999px;
+  background: #fff;
+  color: #334155;
+  font-size: 11px;
+  font-weight: 800;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
+.risk-sim-foot {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+  padding: 12px 14px;
+  border-top: 1px solid #dfe8f1;
+  background: #fff;
+}
+
+.risk-sim-foot > span {
+  color: #64748b;
+  font-size: 12px;
+  font-weight: 750;
+}
+
+.risk-sim-foot div {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+}
+
+.risk-sim-primary-btn,
+.risk-sim-secondary-btn {
+  min-height: 36px;
+  display: inline-flex;
+  align-items: center;
+  gap: 7px;
+  padding: 0 13px;
+  border-radius: 10px;
+  font-size: 13px;
+  font-weight: 850;
+}
+
+.risk-sim-primary-btn {
+  border: 1px solid #0f766e;
+  background: #0f766e;
+  color: #fff;
+}
+
+.risk-sim-secondary-btn {
+  border: 1px solid #dbe5ef;
+  background: #f8fafc;
+  color: #334155;
+}
+
+.risk-sim-dialog-enter-active,
+.risk-sim-dialog-leave-active {
+  transition: opacity 0.18s ease;
+}
+
+.risk-sim-dialog-enter-from,
+.risk-sim-dialog-leave-to {
+  opacity: 0;
+}
+
+@media (max-width: 920px) {
+  .risk-sim-main {
+    grid-template-columns: 1fr;
+  }
+
+  .risk-sim-context-panel {
+    display: none;
+  }
+
+  .risk-sim-field-grid {
+    grid-template-columns: 1fr;
+  }
+
+  .risk-sim-foot {
+    align-items: stretch;
+    flex-direction: column;
+  }
+
+  .risk-sim-foot div {
+    justify-content: flex-end;
+  }
+}
--- a/web/src/assets/styles/views/audit-view-part2.css
+++ b/web/src/assets/styles/views/audit-view-part2.css
@@ -155,6 +155,53 @@
  box-shadow: 0 0 0 3px rgba(16, 185, 129, 0.12);
 }

+.review-submit-test-state,
+.risk-rule-action-confirm,
+.risk-rule-action-note {
+  display: grid;
+  gap: 6px;
+  padding: 10px 12px;
+  border: 1px solid #e2e8f0;
+  border-radius: 8px;
+  background: #f8fafc;
+}
+
+.review-submit-test-state span,
+.risk-rule-action-confirm span,
+.risk-rule-action-note span {
+  color: #64748b;
+  font-size: 12px;
+  font-weight: 800;
+}
+
+.review-submit-test-state strong,
+.risk-rule-action-confirm strong {
+  color: #b45309;
+  font-size: 13px;
+  font-weight: 850;
+}
+
+.review-submit-test-state strong.passed {
+  color: #047857;
+}
+
+.review-submit-test-state p {
+  margin: 0;
+  color: #64748b;
+  font-size: 12px;
+  line-height: 1.5;
+}
+
+.risk-rule-action-note textarea {
+  width: 100%;
+  padding: 10px 12px;
+  border: 1px solid #cbd5e1;
+  border-radius: 8px;
+  resize: vertical;
+  color: #0f172a;
+  font-size: 13px;
+}
+
 .review-submit-hint {
  margin: 0;
  padding: 10px 12px;
@@ -847,6 +894,41 @@
  grid-column: 1 / -1;
 }

+.risk-rule-create-toggle {
+  display: flex !important;
+  align-items: flex-start;
+  gap: 10px;
+  padding: 12px;
+  border: 1px solid #dbe5ef;
+  border-radius: 10px;
+  background: #f8fafc;
+}
+
+.risk-rule-create-toggle input {
+  width: 16px;
+  height: 16px;
+  margin-top: 3px;
+  accent-color: #0f766e;
+}
+
+.risk-rule-create-toggle span {
+  display: grid;
+  gap: 3px;
+  min-width: 0;
+}
+
+.risk-rule-create-toggle strong {
+  color: #0f172a;
+  font-size: 13px;
+  font-weight: 850;
+}
+
+.risk-rule-create-toggle small {
+  color: #64748b;
+  font-size: 12px;
+  line-height: 1.5;
+}
+
 .risk-rule-create-form label span {
  color: #475569;
  font-size: 12px;
@@ -949,6 +1031,20 @@
  color: #059669;
 }

+.minor-action.enable-action {
+  border-color: rgba(100, 116, 139, 0.26);
+  color: #64748b;
+}
+
+.minor-action.enable-action.is-on {
+  border-color: rgba(5, 150, 105, 0.26);
+  color: #059669;
+}
+
+.minor-action.enable-action i {
+  font-size: 18px;
+}
+
 .minor-action.danger-action {
  border-color: rgba(220, 38, 38, 0.2);
  color: #dc2626;
@@ -1009,6 +1105,16 @@
  color: #fff;
 }

+.mini-btn.danger {
+  border-color: rgba(220, 38, 38, 0.24);
+  color: #dc2626;
+}
+
+.mini-btn.warning {
+  border-color: rgba(245, 158, 11, 0.28);
+  color: #b45309;
+}
+
@media (max-width: 1320px) {
  .detail-hero {
    grid-template-columns: 1fr;
@@ -1325,16 +1431,18 @@
  padding-bottom: 2px;
 }

+.json-risk-summary-card,
+.json-risk-flow-card,
+.json-risk-description-card {
+  background: #ffffff;
+  border: 1px solid #e2e8f0;
+}
+
 .json-risk-flow-card {
  min-width: 0;
  overflow: hidden;
 }

-.json-risk-description-card {
-  border-color: #fecdd3;
-  background: linear-gradient(180deg, #fffafb 0%, #ffffff 100%);
-}
-
 .json-risk-description-text {
  margin: 0;
  padding: 0 4px 8px;
@@ -1348,36 +1456,109 @@
 .json-risk-description-source {
  margin: 0;
  padding: 8px 12px 4px;
-  border-top: 1px solid #ffe4e6;
+  border-top: 1px solid #e2e8f0;
  color: #94a3b8;
  font-size: 12px;
  line-height: 1.5;
 }

-.json-risk-summary-grid {
+.json-risk-meta-grid {
  display: grid;
  grid-template-columns: repeat(2, minmax(0, 1fr));
-  gap: 10px;
+  column-gap: 36px;
+  row-gap: 0;
 }

-.json-risk-summary-grid span {
-  min-height: 34px;
+.json-risk-meta-item {
  display: flex;
-  flex-direction: column;
-  gap: 4px;
-  padding: 10px 12px;
-  border-radius: 10px;
-  background: #f8fafc;
-  color: #475569;
-  font-size: 12px;
+  align-items: center;
+  padding: 10px 0;
+  border-bottom: 1px solid #f1f5f9;
+  font-size: 13px;
+  min-height: 40px;
 }

-.json-risk-summary-grid strong {
+.json-risk-meta-item.full-width {
+  grid-column: span 2;
+}
+
+.json-risk-meta-label {
+  width: 84px;
+  flex-shrink: 0;
+  color: #64748b;
+  font-weight: 500;
+}
+
+.json-risk-meta-value {
  color: #0f172a;
+  font-weight: 600;
+  display: flex;
+  align-items: center;
+  flex-wrap: wrap;
+  gap: 6px;
+  min-width: 0;
+}
+
+.meta-value-hint {
+  color: #64748b;
+  font-size: 11.5px;
+  font-weight: 400;
+  margin-left: 2px;
+}
+
+.json-risk-meta-badge {
+  display: inline-flex;
+  align-items: center;
+  padding: 2px 8px;
+  border-radius: 999px;
  font-size: 11px;
  font-weight: 800;
-  text-transform: uppercase;
-  letter-spacing: 0.04em;
+  line-height: 1.2;
+}
+
+.json-risk-meta-badge.high {
+  background: #fef2f2;
+  color: #dc2626;
+}
+
+.json-risk-meta-badge.medium {
+  background: #fff7ed;
+  color: #ea580c;
+}
+
+.json-risk-meta-badge.low {
+  background: #ecfdf5;
+  color: #059669;
+}
+
+.json-risk-meta-badge.test-passed {
+  background: #ecfdf5;
+  color: #047857;
+}
+
+.json-risk-meta-badge.test-pending {
+  background: #fff7ed;
+  color: #b45309;
+}
+
+.meta-status-indicator {
+  display: inline-flex;
+  align-items: center;
+  gap: 6px;
+  font-size: 12.5px;
+  font-weight: 600;
+}
+
+.meta-status-indicator .indicator-dot {
+  width: 6px;
+  height: 6px;
+  border-radius: 50%;
+  background: #94a3b8;
+}
+
+.meta-status-indicator.is-active .indicator-dot {
+  background: #10b981;
+  box-shadow: 0 0 4px rgba(16, 185, 129, 0.4);
 }

 .json-risk-editor-toolbar {
@@ -1403,9 +1584,12 @@

@media (max-width: 760px) {
  .risk-rule-create-form,
-  .json-risk-summary-grid {
+  .json-risk-meta-grid {
    grid-template-columns: 1fr;
  }
+  .json-risk-meta-item.full-width {
+    grid-column: span 1;
+  }
 }

@media (max-width: 860px) {
@@ -1417,4 +1601,5 @@
  .json-risk-editor-actions {
    justify-content: flex-start;
  }
+
 }
--- a/web/src/assets/styles/views/audit-view.css
+++ b/web/src/assets/styles/views/audit-view.css
@@ -425,7 +425,7 @@

 table {
  width: 100%;
-  min-width: 1120px;
+  min-width: 1260px;
  border-collapse: collapse;
 }

--- a/web/src/assets/styles/views/documents-center-view.css
+++ b/web/src/assets/styles/views/documents-center-view.css
@@ -0,0 +1,720 @@
+.documents-page {
+  height: 100%;
+  min-height: 0;
+  display: grid;
+  grid-template-rows: minmax(0, 1fr);
+  animation: fadeUp 220ms var(--ease) both;
+  overflow: hidden;
+}
+
+.documents-list {
+  min-height: 0;
+  display: grid;
+  grid-template-rows: auto auto minmax(0, 1fr) auto;
+  padding: 16px 18px;
+  overflow: hidden;
+}
+
+.status-tabs {
+  display: flex;
+  gap: 28px;
+  margin-top: 14px;
+  border-bottom: 1px solid #dbe4ee;
+}
+
+.status-tabs button {
+  position: relative;
+  min-height: 36px;
+  display: inline-flex;
+  align-items: center;
+  gap: 7px;
+  border: 0;
+  background: transparent;
+  color: #64748b;
+  font-size: 14px;
+  font-weight: 750;
+}
+
+.status-tabs button.active {
+  color: #059669;
+}
+
+.status-tabs button.active::after {
+  content: "";
+  position: absolute;
+  left: 0;
+  right: 0;
+  bottom: -1px;
+  height: 3px;
+  border-radius: 999px 999px 0 0;
+  background: #10b981;
+}
+
+.scope-tab-badge {
+  min-width: 18px;
+  height: 18px;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  padding: 0 5px;
+  border-radius: 999px;
+  background: #ef4444;
+  color: #fff;
+  font-size: 11px;
+  font-weight: 850;
+  line-height: 1;
+  box-shadow: 0 6px 14px rgba(239, 68, 68, 0.22);
+}
+
+.document-toolbar {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 16px;
+  margin-top: 14px;
+}
+
+.filter-set,
+.document-actions {
+  display: flex;
+  align-items: center;
+  gap: 12px;
+  flex-wrap: wrap;
+}
+
+.list-search {
+  position: relative;
+  width: 280px;
+}
+
+.list-search .mdi {
+  position: absolute;
+  left: 12px;
+  top: 50%;
+  transform: translateY(-50%);
+  color: #64748b;
+  font-size: 15px;
+}
+
+.list-search input {
+  width: 100%;
+  height: 38px;
+  padding: 0 12px 0 36px;
+  border: 1px solid #d7e0ea;
+  border-radius: 8px;
+  background: #fff;
+  color: #0f172a;
+  font-size: 13px;
+  transition: border-color 160ms ease, box-shadow 160ms ease;
+}
+
+.list-search input::placeholder {
+  color: #8da0b4;
+}
+
+.list-search input:focus {
+  border-color: #10b981;
+  box-shadow: 0 0 0 3px rgba(16, 185, 129, 0.14);
+  outline: none;
+}
+
+.filter-btn,
+.page-size {
+  min-height: 38px;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 9px;
+  padding: 0 14px;
+  border: 1px solid #d7e0ea;
+  border-radius: 8px;
+  background: #fff;
+  color: #334155;
+  font-size: 14px;
+  font-weight: 750;
+  white-space: nowrap;
+}
+
+.filter-btn {
+  min-width: 120px;
+  justify-content: space-between;
+}
+
+.filter-btn:hover,
+.page-size:hover {
+  border-color: rgba(16, 185, 129, .32);
+  color: #0f9f78;
+}
+
+.document-filter,
+.date-range-filter {
+  position: relative;
+}
+
+.document-filter-menu,
+.date-range-popover,
+.page-size-dropdown {
+  position: absolute;
+  z-index: 40;
+  border: 1px solid #d7e0ea;
+  border-radius: 10px;
+  background: #fff;
+  box-shadow: 0 16px 32px rgba(15, 23, 42, 0.12);
+  overflow: hidden;
+}
+
+.document-filter-menu {
+  top: calc(100% + 8px);
+  left: 0;
+  min-width: 150px;
+  max-height: 280px;
+  padding: 6px;
+  overflow-y: auto;
+}
+
+.document-filter-menu button,
+.page-size-dropdown button {
+  display: block;
+  width: 100%;
+  min-height: 36px;
+  padding: 0 12px;
+  border: 0;
+  border-radius: 8px;
+  background: transparent;
+  color: #334155;
+  font-size: 13px;
+  font-weight: 650;
+  text-align: left;
+  white-space: nowrap;
+}
+
+.document-filter-menu button:hover,
+.document-filter-menu button.active {
+  background: rgba(16, 185, 129, 0.1);
+  color: #047857;
+}
+
+.date-range-trigger {
+  min-width: 150px;
+}
+
+.date-range-label {
+  max-width: 104px;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
+.date-range-popover {
+  top: calc(100% + 8px);
+  left: 0;
+  width: 320px;
+  display: grid;
+  gap: 14px;
+  padding: 16px;
+}
+
+.date-range-popover header,
+.date-range-popover footer {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+}
+
+.date-range-popover header strong {
+  color: #0f172a;
+  font-size: 15px;
+}
+
+.date-range-popover header button {
+  width: 30px;
+  height: 30px;
+  display: grid;
+  place-items: center;
+  border: 0;
+  border-radius: 8px;
+  background: transparent;
+  color: #64748b;
+}
+
+.date-range-fields {
+  display: grid;
+  grid-template-columns: 1fr 1fr;
+  gap: 10px;
+}
+
+.date-range-fields label {
+  display: grid;
+  gap: 6px;
+}
+
+.date-range-fields span {
+  color: #64748b;
+  font-size: 12px;
+  font-weight: 700;
+}
+
+.date-range-fields input {
+  width: 100%;
+  height: 38px;
+  padding: 0 9px;
+  border: 1px solid #d7e0ea;
+  border-radius: 8px;
+  color: #0f172a;
+  font-size: 13px;
+}
+
+.ghost-btn,
+.apply-btn {
+  height: 36px;
+  padding: 0 14px;
+  border-radius: 8px;
+  font-size: 13px;
+  font-weight: 750;
+}
+
+.ghost-btn {
+  border: 1px solid #d7e0ea;
+  background: #fff;
+  color: #334155;
+}
+
+.apply-btn {
+  border: 0;
+  background: #10b981;
+  color: #fff;
+}
+
+.apply-btn:disabled {
+  cursor: not-allowed;
+  background: #cbd5e1;
+}
+
+.create-request-btn {
+  min-height: 40px;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 8px;
+  padding: 0 18px;
+  border: 0;
+  border-radius: 10px;
+  background: linear-gradient(135deg, #10b981, #059669);
+  color: #fff;
+  font-size: 14px;
+  font-weight: 800;
+  white-space: nowrap;
+  box-shadow: 0 10px 24px rgba(5, 150, 105, 0.2);
+  transition: transform 160ms ease, box-shadow 160ms ease, filter 160ms ease;
+}
+
+.create-request-btn.secondary {
+  border: 1px solid #d7e0ea;
+  background: #fff;
+  color: #334155;
+  box-shadow: none;
+}
+
+.create-request-btn:hover {
+  transform: translateY(-1px);
+  box-shadow: 0 14px 28px rgba(5, 150, 105, 0.24);
+  filter: saturate(1.02);
+}
+
+.create-request-btn.secondary:hover {
+  border-color: rgba(16, 185, 129, .32);
+  color: #047857;
+  box-shadow: 0 8px 18px rgba(15, 23, 42, 0.08);
+}
+
+.document-status-filter {
+  display: inline-flex;
+  align-items: center;
+  justify-content: flex-start;
+  gap: 10px;
+  min-height: 38px;
+}
+
+.status-dropdown-filter {
+  min-width: 154px;
+}
+
+.status-filter-trigger {
+  min-width: 154px;
+}
+
+.status-filter-trigger > .mdi:first-child {
+  color: #10b981;
+}
+
+.status-filter-menu {
+  min-width: 154px;
+}
+
+.table-wrap {
+  min-height: 400px;
+  margin-top: 10px;
+  overflow-x: auto;
+  overflow-y: auto;
+  border: 1px solid #edf2f7;
+  border-radius: 10px;
+  background: linear-gradient(180deg, #fcfefd 0%, #f4f8f6 100%);
+  display: flex;
+  flex-direction: column;
+}
+
+.table-wrap.is-empty {
+  align-items: center;
+  justify-content: center;
+}
+
+.table-wrap table {
+  width: 100%;
+  align-self: flex-start;
+}
+
+.table-state {
+  width: 100%;
+  min-height: 260px;
+  display: grid;
+  place-items: center;
+  gap: 10px;
+  padding: 28px 20px;
+  text-align: center;
+  color: #64748b;
+  background: linear-gradient(180deg, #fcfffd 0%, #f5f9f7 100%);
+}
+
+.table-state .mdi {
+  font-size: 28px;
+  color: #10b981;
+}
+
+.table-state strong {
+  color: #0f172a;
+  font-size: 15px;
+}
+
+.table-state p {
+  max-width: 420px;
+  margin: 0;
+  font-size: 13px;
+  line-height: 1.6;
+}
+
+.table-state.error {
+  background: linear-gradient(180deg, #fffdfd 0%, #fff6f6 100%);
+}
+
+.table-state.error .mdi {
+  color: #ef4444;
+}
+
+.retry-btn {
+  height: 36px;
+  padding: 0 14px;
+  border: 1px solid #f1c5c5;
+  border-radius: 8px;
+  background: #fff;
+  color: #b91c1c;
+  font-size: 13px;
+  font-weight: 750;
+}
+
+table {
+  width: 100%;
+  min-width: 1320px;
+  border-collapse: collapse;
+  table-layout: fixed;
+}
+
+.col-id { width: 11%; }
+.col-created { width: 10%; }
+.col-stay { width: 9%; }
+.col-doc-type { width: 9%; }
+.col-scene { width: 10%; }
+.col-title { width: 18%; }
+.col-amount { width: 9%; }
+.col-node { width: 12%; }
+.col-status { width: 8%; }
+.col-updated { width: 9%; }
+
+th,
+td {
+  padding: 13px 12px;
+  border-bottom: 1px solid #edf2f7;
+  color: #24324a;
+  font-size: 14px;
+  line-height: 1.35;
+  text-align: center;
+  vertical-align: middle;
+  white-space: nowrap;
+  overflow: hidden;
+  text-overflow: ellipsis;
+}
+
+th {
+  position: sticky;
+  top: 0;
+  z-index: 1;
+  background: #f7fafc;
+  color: #64748b;
+  font-size: 13px;
+  font-weight: 800;
+}
+
+tbody tr {
+  cursor: pointer;
+}
+
+tbody tr:hover {
+  background: linear-gradient(90deg, rgba(16, 185, 129, .08), rgba(16, 185, 129, .03));
+}
+
+tbody tr:last-child td {
+  border-bottom: 0;
+}
+
+.doc-id {
+  color: #059669;
+  font-weight: 800;
+}
+
+.doc-kind-tag,
+.type-tag,
+.status-tag {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  white-space: nowrap;
+}
+
+.doc-kind-tag {
+  min-height: 26px;
+  padding: 0 10px;
+  border-radius: 7px;
+  font-size: 12px;
+  font-weight: 800;
+}
+
+.doc-kind-tag.reimbursement {
+  background: #ecfdf5;
+  color: #059669;
+}
+
+.doc-kind-tag.application {
+  background: #eff6ff;
+  color: #2563eb;
+}
+
+.type-tag {
+  min-height: 26px;
+  padding: 0 10px;
+  border-radius: 999px;
+  font-size: 12px;
+  font-weight: 800;
+}
+
+.type-tag.travel,
+.type-tag.hotel,
+.type-tag.transport {
+  background: #ecfdf5;
+  color: #047857;
+}
+
+.type-tag.entertainment,
+.type-tag.meal {
+  background: #fff7ed;
+  color: #ea580c;
+}
+
+.type-tag.office {
+  background: #eff6ff;
+  color: #2563eb;
+}
+
+.type-tag.meeting,
+.type-tag.training {
+  background: #eef2ff;
+  color: #4f46e5;
+}
+
+.type-tag.other {
+  background: #f8fafc;
+  color: #475569;
+}
+
+.status-tag {
+  min-height: 24px;
+  padding: 0 9px;
+  border: 1px solid transparent;
+  border-radius: 6px;
+  font-size: 12px;
+  font-weight: 750;
+}
+
+.status-tag.info {
+  border-color: #bfdbfe;
+  background: #eff6ff;
+  color: #2563eb;
+}
+
+.status-tag.success,
+.status-tag.archived {
+  border-color: #bbf7d0;
+  background: #ecfdf5;
+  color: #059669;
+}
+
+.status-tag.warning,
+.status-tag.draft {
+  border-color: #fed7aa;
+  background: #fff7ed;
+  color: #f97316;
+}
+
+.status-tag.danger {
+  border-color: #fecaca;
+  background: #fef2f2;
+  color: #dc2626;
+}
+
+.status-tag.neutral {
+  border-color: #cbd5e1;
+  background: #f8fafc;
+  color: #475569;
+}
+
+.list-foot {
+  display: grid;
+  grid-template-columns: 1fr auto 1fr;
+  align-items: center;
+  gap: 16px;
+  margin-top: 12px;
+}
+
+.page-summary {
+  color: #64748b;
+  font-size: 14px;
+  font-weight: 650;
+}
+
+.pager {
+  display: inline-flex;
+  justify-content: center;
+  gap: 6px;
+  padding: 4px;
+  border: 1px solid #e2e8f0;
+  border-radius: 12px;
+  background: #f8fafc;
+}
+
+.pager button {
+  width: 32px;
+  height: 32px;
+  border: 0;
+  border-radius: 9px;
+  background: transparent;
+  color: #334155;
+  font-size: 14px;
+  font-weight: 800;
+}
+
+.pager button:hover:not(.active) {
+  background: #fff;
+  color: #059669;
+  box-shadow: 0 1px 4px rgba(15, 23, 42, .08);
+}
+
+.pager button.active {
+  background: #059669;
+  color: #fff;
+  box-shadow: 0 8px 16px rgba(5, 150, 105, .20);
+}
+
+.pager button:disabled {
+  color: #cbd5e1;
+  cursor: not-allowed;
+}
+
+.page-size-wrap {
+  position: relative;
+  justify-self: end;
+}
+
+.page-size {
+  min-width: 112px;
+  border-radius: 10px;
+  background: #fff;
+  box-shadow: 0 1px 2px rgba(15, 23, 42, .04);
+}
+
+.page-size-dropdown {
+  right: 0;
+  bottom: calc(100% + 6px);
+  display: grid;
+}
+
+.page-size-dropdown button {
+  border-radius: 0;
+  text-align: center;
+  padding: 0 20px;
+}
+
+.page-size-dropdown button:hover {
+  background: #f0fdf4;
+  color: #059669;
+}
+
+.page-size-dropdown button.active {
+  background: #059669;
+  color: #fff;
+}
+
+@media (max-width: 1200px) {
+  .document-toolbar,
+  .list-foot {
+    grid-template-columns: 1fr;
+  }
+
+  .document-toolbar {
+    align-items: stretch;
+    flex-direction: column;
+  }
+
+  .document-actions {
+    justify-content: flex-start;
+  }
+}
+
+@media (max-width: 760px) {
+  .documents-list {
+    padding: 16px;
+  }
+
+  .status-tabs {
+    gap: 18px;
+    overflow-x: auto;
+  }
+
+  .filter-set,
+  .document-actions,
+  .document-status-filter,
+  .list-search,
+  .filter-btn,
+  .page-size {
+    width: 100%;
+  }
+
+  .document-status-filter {
+    align-items: stretch;
+    flex-direction: column;
+    gap: 6px;
+  }
+
+  .list-foot {
+    display: grid;
+    justify-items: stretch;
+  }
+}
--- a/web/src/assets/styles/views/logs-view.css
+++ b/web/src/assets/styles/views/logs-view.css
@@ -237,7 +237,7 @@
  color: #475569;
  font-size: 12px;
  font-weight: 800;
-  text-align: left;
+  text-align: center;
  white-space: nowrap;
 }

@@ -260,11 +260,13 @@
  border-bottom: 1px solid #eef3f8;
  color: #0f172a;
  font-size: 13px;
-  vertical-align: top;
+  text-align: center;
+  vertical-align: middle;
 }

 .summary-cell {
  min-width: 220px;
+  text-align: left;
 }

 .summary-cell strong,
@@ -301,6 +303,7 @@
  display: grid;
  gap: 5px;
  align-content: start;
+  justify-items: center;
 }

 .status-note {
@@ -632,52 +635,60 @@
  display: inline-flex;
  align-items: center;
  justify-content: center;
-  min-width: 56px;
-  padding: 4px 9px;
-  border-radius: 999px;
+  padding: 3px 8px;
+  border-radius: 6px;
  font-size: 12px;
-  font-weight: 800;
+  font-weight: 600;
  white-space: nowrap;
+  border: 1px solid transparent;
 }

 .status-pill.success {
  background: rgba(22, 163, 74, 0.12);
+  border-color: rgba(22, 163, 74, 0.2);
  color: #166534;
 }

 .status-pill.warning {
  background: rgba(245, 158, 11, 0.14);
+  border-color: rgba(245, 158, 11, 0.25);
  color: #b45309;
 }

 .status-pill.danger {
  background: rgba(239, 68, 68, 0.14);
+  border-color: rgba(239, 68, 68, 0.25);
  color: #b91c1c;
 }

 .status-pill.muted {
  background: rgba(148, 163, 184, 0.14);
+  border-color: rgba(148, 163, 184, 0.25);
  color: #475569;
 }

 .status-pill.info,
 .level-pill.info {
  background: rgba(37, 99, 235, 0.12);
+  border-color: rgba(37, 99, 235, 0.25);
  color: #1d4ed8;
 }

 .level-pill.warning {
  background: rgba(245, 158, 11, 0.14);
+  border-color: rgba(245, 158, 11, 0.25);
  color: #b45309;
 }

 .level-pill.danger {
  background: rgba(239, 68, 68, 0.14);
+  border-color: rgba(239, 68, 68, 0.25);
  color: #b91c1c;
 }

 .level-pill.muted {
  background: rgba(148, 163, 184, 0.14);
+  border-color: rgba(148, 163, 184, 0.25);
  color: #475569;
 }

--- a/web/src/components/layout/SidebarRail.vue
+++ b/web/src/components/layout/SidebarRail.vue
@@ -84,6 +84,7 @@ const {
 const sidebarMeta = {
  overview: { label: '财务总览' },
  workbench: { label: '个人工作台' },
+  documents: { label: '单据中心' },
  requests: { label: '报销中心' },
  approval: { label: '审批中心' },
  archive: { label: '归档中心' },
--- a/web/src/components/layout/TopBar.vue
+++ b/web/src/components/layout/TopBar.vue
@@ -109,6 +109,16 @@
        </div>
      </template>

+      <template v-else-if="isDocuments">
+        <div class="kpi-chips">
+          <div v-for="kpi in documentKpis" :key="kpi.label" class="kpi-chip" :style="{ '--chip-color': kpi.color }">
+            <span class="chip-value">{{ kpi.value }}<small>单</small></span>
+            <span class="chip-label">{{ kpi.label }}</span>
+            <span class="chip-delta" :class="kpi.trend">{{ kpi.delta }} <i :class="kpi.arrow"></i></span>
+          </div>
+        </div>
+      </template>
+
      <template v-else-if="isRequests">
        <div class="kpi-chips">
          <div v-for="kpi in requestKpis" :key="kpi.label" class="kpi-chip" :style="{ '--chip-color': kpi.color }">
@@ -192,6 +202,10 @@ const props = defineProps({
    type: Object,
    default: () => null
  },
+  documentSummary: {
+    type: Object,
+    default: () => null
+  },
  workbenchSummary: {
    type: Object,
    default: () => null
@@ -226,7 +240,8 @@ const emit = defineEmits([
 const isChat = computed(() => props.activeView === 'chat')
 const isOverview = computed(() => props.activeView === 'overview')
 const isWorkbench = computed(() => props.activeView === 'workbench')
-const isRequestDetail = computed(() => props.activeView === 'requests' && props.detailMode)
+const isRequestDetail = computed(() => ['requests', 'documents'].includes(props.activeView) && props.detailMode)
+const isDocuments = computed(() => props.activeView === 'documents' && !props.detailMode)
 const isRequests = computed(() => props.activeView === 'requests')
 const isLogs = computed(() => props.activeView === 'logs' && !props.logDetailMode)
 const isApproval = computed(() => props.activeView === 'approval')
@@ -291,6 +306,21 @@ const requestKpis = computed(() => {
  ]
 })

+const documentKpis = computed(() => {
+  const summary = props.documentSummary ?? {}
+  const total = Number(summary.total ?? 0)
+  const toSubmit = Number(summary.toSubmit ?? 0)
+  const toProcess = Number(summary.toProcess ?? 0)
+  const archived = Number(summary.archived ?? 0)
+
+  return [
+    { label: '全部单据', value: total, delta: '当前', trend: 'up', arrow: 'mdi mdi-minus', color: '#10b981' },
+    { label: '待提交', value: toSubmit, delta: '草稿待办', trend: toSubmit > 0 ? 'down' : 'up', arrow: toSubmit > 0 ? 'mdi mdi-arrow-down' : 'mdi mdi-minus', color: '#f59e0b' },
+    { label: '待我处理', value: toProcess, delta: '审批待办', trend: toProcess > 0 ? 'down' : 'up', arrow: toProcess > 0 ? 'mdi mdi-arrow-down' : 'mdi mdi-minus', color: '#3b82f6' },
+    { label: '已归档', value: archived, delta: '归档入账', trend: 'up', arrow: 'mdi mdi-arrow-up', color: '#10b981' }
+  ]
+})
+
 const logsKpis = computed(() => {
  const summary = props.logsSummary ?? {}
  const total = Number(summary.total ?? 0)
--- a/web/src/components/shared/ExpenseApplicationDialog.vue
+++ b/web/src/components/shared/ExpenseApplicationDialog.vue
@@ -0,0 +1,185 @@
+<template>
+  <div class="expense-application-mask" @click.self="emit('close')">
+    <section class="expense-application-dialog" role="dialog" aria-modal="true" aria-labelledby="expense-application-title">
+      <header class="application-dialog-header">
+        <div>
+          <span class="application-dialog-eyebrow">费用申请</span>
+          <h2 id="expense-application-title">发起申请</h2>
+        </div>
+        <button class="dialog-icon-btn" type="button" aria-label="关闭" @click="emit('close')">
+          <i class="mdi mdi-close"></i>
+        </button>
+      </header>
+
+      <div class="application-dialog-body">
+        <section class="application-input-panel">
+          <label class="application-field-label" for="application-intent-input">申请意图</label>
+          <textarea
+            id="application-intent-input"
+            v-model="draft"
+            rows="5"
+            placeholder="例如：申请下周去北京做客户现场验收，差旅预算18000元"
+          ></textarea>
+
+          <div class="application-example-row">
+            <button
+              v-for="example in APPLICATION_EXAMPLES"
+              :key="example"
+              class="application-example-chip"
+              type="button"
+              @click="applyExample(example)"
+            >
+              {{ example }}
+            </button>
+          </div>
+
+          <button class="primary-parse-btn" type="button" :disabled="parsing || !draft.trim()" @click="parseApplication">
+            <i :class="parsing ? 'mdi mdi-loading mdi-spin' : 'mdi mdi-brain'"></i>
+            <span>{{ parsing ? '识别中' : '识别申请字段' }}</span>
+          </button>
+
+          <p v-if="error" class="application-error">{{ error }}</p>
+        </section>
+
+        <section class="application-ontology-panel">
+          <div class="ontology-panel-head">
+            <span>本体识别结果</span>
+            <strong>{{ confidenceLabel }}</strong>
+          </div>
+
+          <div v-if="!fields" class="ontology-empty-state">
+            <i class="mdi mdi-file-search-outline"></i>
+            <p>输入申请事项后，系统会调用本体解析费用场景、金额、时间、地点和附件要求。</p>
+          </div>
+
+          <template v-else>
+            <div class="ontology-chip-row">
+              <span class="ontology-chip">scenario: {{ ontology?.scenario || '-' }}</span>
+              <span class="ontology-chip">intent: {{ ontology?.intent || '-' }}</span>
+              <span class="ontology-chip">{{ fields.documentTypeLabel }}</span>
+            </div>
+
+            <div class="application-field-grid">
+              <div class="application-field-card">
+                <span>费用场景</span>
+                <strong>{{ fields.expenseTypeLabel }}</strong>
+              </div>
+              <div class="application-field-card">
+                <span>申请金额</span>
+                <strong>{{ fields.amountDisplay }}</strong>
+              </div>
+              <div class="application-field-card">
+                <span>业务时间</span>
+                <strong>{{ fields.timeRange }}</strong>
+              </div>
+              <div class="application-field-card">
+                <span>业务地点</span>
+                <strong>{{ fields.location }}</strong>
+              </div>
+              <div class="application-field-card">
+                <span>申请人</span>
+                <strong>{{ fields.applicant }}</strong>
+              </div>
+              <div class="application-field-card">
+                <span>所属部门</span>
+                <strong>{{ fields.department }}</strong>
+              </div>
+            </div>
+
+            <div class="application-policy-strip" :class="fields.attachmentPolicy.level">
+              <i class="mdi mdi-paperclip-check"></i>
+              <div>
+                <strong>附件要求：{{ fields.attachmentPolicy.label }}</strong>
+                <p>{{ fields.attachmentPolicy.description }}</p>
+              </div>
+            </div>
+
+            <div class="application-reason-block">
+              <span>申请事由</span>
+              <p>{{ fields.reason }}</p>
+            </div>
+
+            <div v-if="fields.missingSlots.length" class="application-missing-block">
+              <span>待补充字段</span>
+              <div>
+                <em v-for="slot in fields.missingSlots" :key="slot.key">{{ slot.label }}</em>
+              </div>
+            </div>
+          </template>
+        </section>
+      </div>
+
+      <footer class="application-dialog-footer">
+        <button class="secondary-btn" type="button" @click="emit('close')">取消</button>
+        <button class="confirm-btn" type="button" :disabled="!fields" @click="confirmApplication">
+          <i class="mdi mdi-check-circle-outline"></i>
+          <span>确认本体字段</span>
+        </button>
+      </footer>
+    </section>
+  </div>
+</template>
+
+<script setup>
+import { computed, ref } from 'vue'
+
+import { useSystemState } from '../../composables/useSystemState.js'
+import { fetchOntologyParse } from '../../services/ontology.js'
+import {
+  APPLICATION_EXAMPLES,
+  buildApplicationFieldsFromOntology,
+  buildExpenseApplicationOntologyContext
+} from '../../utils/expenseApplicationOntology.js'
+
+const emit = defineEmits(['close', 'confirmed'])
+
+const { currentUser } = useSystemState()
+const draft = ref(APPLICATION_EXAMPLES[0])
+const parsing = ref(false)
+const error = ref('')
+const ontology = ref(null)
+const fields = ref(null)
+
+const confidenceLabel = computed(() => {
+  if (!ontology.value) return '待识别'
+  return `${Math.round(Number(ontology.value.confidence || 0) * 100)}%`
+})
+
+function applyExample(example) {
+  draft.value = example
+}
+
+async function parseApplication() {
+  const query = draft.value.trim()
+  if (!query) return
+
+  parsing.value = true
+  error.value = ''
+  try {
+    const payload = await fetchOntologyParse({
+      query,
+      user_id: currentUser.value?.username || currentUser.value?.name || 'anonymous',
+      context_json: buildExpenseApplicationOntologyContext(currentUser.value || {})
+    })
+    ontology.value = payload
+    fields.value = buildApplicationFieldsFromOntology(payload, query, currentUser.value || {})
+  } catch (err) {
+    ontology.value = null
+    fields.value = null
+    error.value = err?.message || '申请字段识别失败，请稍后重试。'
+  } finally {
+    parsing.value = false
+  }
+}
+
+function confirmApplication() {
+  if (!fields.value) return
+  emit('confirmed', {
+    ontology: ontology.value,
+    fields: fields.value,
+    sourceText: draft.value.trim()
+  })
+}
+</script>
+
+<style scoped src="../../assets/styles/components/expense-application-dialog.css"></style>
--- a/web/src/components/shared/RiskRuleFlowDiagram.vue
+++ b/web/src/components/shared/RiskRuleFlowDiagram.vue
@@ -46,18 +46,115 @@
        />
        <div
          v-else
-          class="risk-rule-flow-svg"
-          role="img"
-          aria-label="风险规则流程说明"
-          v-html="displaySvg"
-        ></div>
+          class="risk-rule-flow-svg-viewport"
+          @mousedown="onDragStart"
+          @touchstart="onTouchStart"
+          @dblclick="resetZoom"
+        >
+          <div
+            class="risk-rule-flow-svg-canvas"
+            :style="transformStyle"
+            v-html="displaySvg"
+          ></div>
+          <div class="diagram-zoom-controls" @mousedown.stop @touchstart.stop>
+            <button class="zoom-btn" @click="zoomIn" title="放大">
+              <i class="mdi mdi-plus"></i>
+            </button>
+            <button class="zoom-btn" @click="zoomOut" title="缩小">
+              <i class="mdi mdi-minus"></i>
+            </button>
+            <button class="zoom-btn" @click="resetZoom" title="重置">
+              <i class="mdi mdi-arrow-expand-all"></i>
+            </button>
+          </div>
+        </div>
      </div>
    </div>
  </section>
 </template>

 <script setup>
-import { computed } from 'vue'
+import { computed, ref, onUnmounted } from 'vue'
+
+const scale = ref(1)
+const translateX = ref(0)
+const translateY = ref(0)
+const isDragging = ref(false)
+const dragStart = { x: 0, y: 0 }
+
+const transformStyle = computed(() => ({
+  transform: `translate(${translateX.value}px, ${translateY.value}px) scale(${scale.value})`,
+  transformOrigin: 'center center',
+  transition: isDragging.value ? 'none' : 'transform 0.15s ease-out'
+}))
+
+function onDragStart(e) {
+  if (e.button !== 0) return
+  isDragging.value = true
+  dragStart.x = e.clientX - translateX.value
+  dragStart.y = e.clientY - translateY.value
+  
+  window.addEventListener('mousemove', onDragging)
+  window.addEventListener('mouseup', onDragEnd)
+}
+
+function onDragging(e) {
+  if (!isDragging.value) return
+  translateX.value = e.clientX - dragStart.x
+  translateY.value = e.clientY - dragStart.y
+}
+
+function onDragEnd() {
+  isDragging.value = false
+  window.removeEventListener('mousemove', onDragging)
+  window.removeEventListener('mouseup', onDragEnd)
+}
+
+function onTouchStart(e) {
+  if (e.touches.length !== 1) return
+  isDragging.value = true
+  const touch = e.touches[0]
+  dragStart.x = touch.clientX - translateX.value
+  dragStart.y = touch.clientY - translateY.value
+  
+  window.addEventListener('touchmove', onTouchMove, { passive: false })
+  window.addEventListener('touchend', onTouchEnd)
+}
+
+function onTouchMove(e) {
+  if (!isDragging.value || e.touches.length !== 1) return
+  e.preventDefault()
+  const touch = e.touches[0]
+  translateX.value = touch.clientX - dragStart.x
+  translateY.value = touch.clientY - dragStart.y
+}
+
+function onTouchEnd() {
+  isDragging.value = false
+  window.removeEventListener('touchmove', onTouchMove)
+  window.removeEventListener('touchend', onTouchEnd)
+}
+
+function zoomIn() {
+  scale.value = Math.min(scale.value + 0.15, 3)
+}
+
+function zoomOut() {
+  scale.value = Math.max(scale.value - 0.15, 0.4)
+}
+
+function resetZoom() {
+  scale.value = 1
+  translateX.value = 0
+  translateY.value = 0
+}
+
+onUnmounted(() => {
+  window.removeEventListener('mousemove', onDragging)
+  window.removeEventListener('mouseup', onDragEnd)
+  window.removeEventListener('touchmove', onTouchMove)
+  window.removeEventListener('touchend', onTouchEnd)
+})

 const props = defineProps({
  svg: { type: String, default: '' },
@@ -68,8 +165,7 @@ const props = defineProps({
  severityLabel: { type: String, default: '中风险' }
 })

-const FONT =
-  "-apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Helvetica Neue, 'PingFang SC', 'Microsoft YaHei', 'Microsoft JhengHei', 'SimHei', sans-serif"
+const FONT = "Helvetica, Arial, sans-serif"
 const TEXT = '#0d0d0d'
 const MUTED = '#6e6e80'
 const NEUTRAL_LINE = '#cbd5e1'
@@ -95,6 +191,16 @@ const PALETTES = {
  }
 }

+const DRAWIO_PALETTES = {
+  neutral: { fill: '#ffffff', stroke: '#e2e8f0' },
+  blue: { fill: '#ffffff', stroke: '#e2e8f0' },
+  yellow: { fill: '#ffffff', stroke: '#e2e8f0' },
+  green: { fill: '#ffffff', stroke: '#e2e8f0' },
+  low: { fill: '#eff6ff', stroke: '#bfdbfe' },
+  medium: { fill: '#fff7ed', stroke: '#fed7aa' },
+  high: { fill: '#fef2f2', stroke: '#fecaca' }
+}
+
 function normalizeText(value, fallback = '') {
  return String(value || fallback || '').trim()
 }
@@ -156,15 +262,12 @@ function textLines(lines, x, y, anchor = 'middle', color = MUTED, fontSize = 13)
    .join('')
 }

-function node(title, body, x, y, width, height, currentPalette = null) {
-  const border = currentPalette?.border || NEUTRAL_BORDER
-  const stripe = currentPalette?.accent || NEUTRAL_LINE
-  const surface = currentPalette?.surface || '#ffffff'
-  return `<g>
-    <rect x="${x}" y="${y}" width="${width}" height="${height}" rx="7" ry="7" fill="${surface}" stroke="${border}" stroke-width="1.2"/>
-    <rect x="${x}" y="${y}" width="3.5" height="${height}" rx="1.75" ry="1.75" fill="${stripe}"/>
-    <text x="${x + width / 2}" y="${y + 24}" text-anchor="middle" fill="${TEXT}" font-family="${FONT}" font-size="13" font-weight="600">${escapeSvg(title)}</text>
-    ${textLines(wrapText(body, width <= 126 ? 10 : 11, 1), x + width / 2, y + 43, 'middle', MUTED, 11)}
+function node(title, body, x, y, width, height, type = 'blue') {
+  const palette = DRAWIO_PALETTES[type] || DRAWIO_PALETTES.blue
+  return `<g class="drawio-node">
+    <rect x="${x}" y="${y}" width="${width}" height="${height}" rx="3" ry="3" fill="${palette.fill}" stroke="${palette.stroke}" stroke-width="1.2" filter="url(#shadow)"/>
+    <text x="${x + width / 2}" y="${y + 24}" text-anchor="middle" fill="#0f172a" font-family="${FONT}" font-size="13" font-weight="600">${escapeSvg(title)}</text>
+    ${textLines(wrapText(body, width <= 126 ? 10 : 11, 1), x + width / 2, y + 43, 'middle', '#475569', 11)}
  </g>`
 }

@@ -172,18 +275,19 @@ function diamond(title, body, x, y, width, height) {
  const cx = x + width / 2
  const cy = y + height / 2
  const points = `${cx},${y} ${x + width},${cy} ${cx},${y + height} ${x},${cy}`
-  return `<g>
-    <polygon points="${points}" fill="#ffffff" stroke="${NEUTRAL_BORDER}" stroke-width="1.25"/>
-    <text x="${cx}" y="${cy - 10}" text-anchor="middle" fill="${TEXT}" font-family="${FONT}" font-size="12.5" font-weight="600">${escapeSvg(title)}</text>
-    ${textLines(wrapText(body, 8, 2), cx, cy + 11, 'middle', MUTED, 10.2)}
+  const palette = DRAWIO_PALETTES.yellow
+  return `<g class="drawio-node">
+    <polygon points="${points}" fill="${palette.fill}" stroke="${palette.stroke}" stroke-width="1.2" filter="url(#shadow)"/>
+    <text x="${cx}" y="${cy - 8}" text-anchor="middle" fill="#0f172a" font-family="${FONT}" font-size="12.5" font-weight="600">${escapeSvg(title)}</text>
+    ${textLines(wrapText(body, 8, 2), cx, cy + 12, 'middle', '#475569', 10.2)}
  </g>`
 }

 function note(body) {
-  return `<g>
-    <rect x="214" y="218" width="290" height="36" rx="7" ry="7" fill="#ffffff" stroke="${NEUTRAL_BORDER}" stroke-width="1" stroke-dasharray="4,3"/>
-    <text x="226" y="240" fill="${MUTED}" font-family="${FONT}" font-size="10" font-weight="500">BASIS</text>
-    ${textLines(wrapText(body, 22, 1), 268, 240, 'start', TEXT, 10.2)}
+  return `<g class="drawio-node note">
+    <rect x="214" y="218" width="290" height="36" rx="3" ry="3" fill="#ffffff" stroke="#cbd5e1" stroke-width="1.2" stroke-dasharray="3,3" filter="url(#shadow)"/>
+    <text x="226" y="240" fill="#64748b" font-family="${FONT}" font-size="10.5" font-weight="700">BASIS</text>
+    ${textLines(wrapText(body, 22, 1), 272, 240, 'start', '#334155', 10.5)}
  </g>`
 }

@@ -262,29 +366,47 @@ const displaySvg = computed(() => {
  }

  const flow = flowModel.value
-  const currentPalette = palette.value
+  const severity = props.severity

  return `<svg xmlns="http://www.w3.org/2000/svg" width="760" height="280" viewBox="0 0 760 280" data-risk-flow-style="review-node-only" role="img" aria-label="风险规则流程说明">
  <defs>
-    <marker id="arrow-neutral" markerWidth="10" markerHeight="7" refX="9" refY="3.5" orient="auto">
-      <polygon points="0 0, 10 3.5, 0 7" fill="${NEUTRAL_LINE}"/>
+    <pattern id="grid" width="16" height="16" patternUnits="userSpaceOnUse">
+      <path d="M 16 0 L 0 0 0 16" fill="none" stroke="#e8ecef" stroke-width="0.75"/>
+    </pattern>
+    <marker id="arrow" markerWidth="8" markerHeight="6" refX="7" refY="3" orient="auto">
+      <path d="M 0 0 L 8 3 L 0 6 Z" fill="#666666"/>
    </marker>
+    <filter id="shadow" x="-10%" y="-10%" width="120%" height="120%">
+      <feDropShadow dx="1" dy="2" stdDeviation="1.5" flood-color="#000000" flood-opacity="0.08" />
+    </filter>
  </defs>
  <rect width="760" height="280" fill="#ffffff"/>
-  <rect x="18" y="18" width="724" height="244" rx="8" ry="8" fill="none" stroke="${NEUTRAL_BORDER}" stroke-width="1" stroke-dasharray="4,3"/>
-  <text x="34" y="43" fill="${MUTED}" font-family="${FONT}" font-size="11" font-weight="500">RULE FLOW</text>
-  ${node('业务输入', flow.start, 48, 118, 124, 60)}
-  ${node('字段取数', '读取字段证据', 214, 118, 132, 60)}
+  <rect width="760" height="280" fill="url(#grid)"/>
+  <rect x="0.5" y="0.5" width="759.5" height="279.5" rx="6" fill="none" stroke="#cbd5e1" stroke-width="1"/>
+  
+  <text x="34" y="43" fill="#94a3b8" font-family="${FONT}" font-size="10.5" font-weight="700" letter-spacing="0.05em">RULE FLOW CANVAS</text>
+  
+  ${node('业务输入', flow.start, 48, 118, 124, 60, 'neutral')}
+  ${node('字段取数', '读取字段证据', 214, 118, 132, 60, 'blue')}
  ${diamond('判断依据', flow.decision, 392, 92, 112, 112)}
-  ${node('继续流转', flow.pass, 562, 74, 126, 60)}
-  ${node('进入复核', flow.fail, 562, 190, 126, 62, currentPalette)}
+  ${node('继续流转', flow.pass, 562, 74, 126, 60, 'green')}
+  ${node('进入复核', flow.fail, 562, 190, 126, 62, severity)}
  ${note(flow.basis)}
-  <line x1="172" y1="148" x2="214" y2="148" stroke="${NEUTRAL_LINE}" stroke-width="1.45" marker-end="url(#arrow-neutral)"/>
-  <line x1="346" y1="148" x2="392" y2="148" stroke="${NEUTRAL_LINE}" stroke-width="1.45" marker-end="url(#arrow-neutral)"/>
-  <path d="M 504 127 L 532 127 L 532 104 L 562 104" fill="none" stroke="${NEUTRAL_LINE}" stroke-width="1.35" marker-end="url(#arrow-neutral)"/>
-  <text x="534" y="119" text-anchor="middle" fill="${MUTED}" font-family="${FONT}" font-size="10.5" font-weight="400">否</text>
-  <path d="M 504 169 L 532 169 L 532 221 L 562 221" fill="none" stroke="${NEUTRAL_LINE}" stroke-width="1.8" marker-end="url(#arrow-neutral)"/>
-  <text x="534" y="195" text-anchor="middle" fill="${MUTED}" font-family="${FONT}" font-size="10.5" font-weight="600">是</text>
+  
+  <line x1="172" y1="148" x2="214" y2="148" stroke="#666666" stroke-width="1.5" marker-end="url(#arrow)"/>
+  <line x1="346" y1="148" x2="392" y2="148" stroke="#666666" stroke-width="1.5" marker-end="url(#arrow)"/>
+  
+  <path d="M 504 127 L 532 127 L 532 104 L 562 104" fill="none" stroke="#666666" stroke-width="1.5" marker-end="url(#arrow)"/>
+  <g>
+    <rect x="521" y="108" width="22" height="15" fill="#ffffff" stroke="#cbd5e1" stroke-width="1" rx="2"/>
+    <text x="532" y="120" text-anchor="middle" fill="#475569" font-family="${FONT}" font-size="10" font-weight="bold">否</text>
+  </g>
+  
+  <path d="M 504 169 L 532 169 L 532 221 L 562 221" fill="none" stroke="#666666" stroke-width="1.5" marker-end="url(#arrow)"/>
+  <g>
+    <rect x="521" y="187" width="22" height="15" fill="#ffffff" stroke="#cbd5e1" stroke-width="1" rx="2"/>
+    <text x="532" y="199" text-anchor="middle" fill="#475569" font-family="${FONT}" font-size="10" font-weight="bold">是</text>
+  </g>
 </svg>`
 })
 </script>
@@ -294,10 +416,9 @@ const displaySvg = computed(() => {
  width: 100%;
  min-height: 0;
  overflow: hidden;
-  border: 1px solid #e2e8f0;
-  border-radius: 8px;
-  background: #ffffff;
-  padding: 14px;
+  border: none;
+  background: transparent;
+  padding: 0;
  cursor: default;
 }

@@ -442,30 +563,84 @@ const displaySvg = computed(() => {
  justify-content: flex-start;
 }

-.risk-rule-flow-image,
-.risk-rule-flow-svg {
+.risk-rule-flow-svg-viewport {
+  position: relative;
+  width: 100%;
+  height: 280px;
+  overflow: hidden;
+  border-radius: 6px;
+  background: #ffffff;
+  border: 1px solid #e2e8f0;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  user-select: none;
+  cursor: grab;
+}
+
+.risk-rule-flow-svg-viewport:active {
+  cursor: grabbing;
+}
+
+.risk-rule-flow-svg-canvas {
+  width: 760px;
+  height: 280px;
+  flex-shrink: 0;
+}
+
+.risk-rule-flow-svg-canvas :deep(svg) {
+  width: 100%;
+  height: 100%;
+  display: block;
+}
+
+.diagram-zoom-controls {
+  position: absolute;
+  bottom: 12px;
+  right: 12px;
+  display: flex;
+  flex-direction: column;
+  gap: 6px;
+  background: rgba(255, 255, 255, 0.85);
+  backdrop-filter: blur(8px);
+  border: 1px solid #e2e8f0;
+  border-radius: 8px;
+  padding: 4px;
+  box-shadow: 0 4px 12px rgba(0, 0, 0, 0.05);
+  z-index: 10;
+}
+
+.zoom-btn {
+  width: 28px;
+  height: 28px;
+  display: grid;
+  place-items: center;
+  border: none;
+  border-radius: 6px;
+  background: transparent;
+  color: #475569;
+  cursor: pointer;
+  transition: all 0.2s ease;
+}
+
+.zoom-btn:hover {
+  background: #f1f5f9;
+  color: #0f172a;
+}
+
+.zoom-btn i {
+  font-size: 16px;
+}
+
+.risk-rule-flow-image {
  width: min(760px, 100%);
  display: block;
  pointer-events: none;
  user-select: none;
-}
-
-.risk-rule-flow-image {
  height: auto;
  object-fit: contain;
 }

-.risk-rule-flow-svg :deep(svg) {
-  width: 100%;
-  height: auto;
-  display: block;
-}
-
-.risk-rule-flow-svg :deep(*) {
-  pointer-events: none !important;
-  user-select: none;
-}
-
@media (max-width: 980px) {
  .risk-rule-flow-content {
    grid-template-columns: 1fr;
--- a/web/src/components/shared/RiskRuleTestDialog.vue
+++ b/web/src/components/shared/RiskRuleTestDialog.vue
@@ -0,0 +1,792 @@
+<template>
+  <Transition name="risk-sim-dialog">
+    <div v-if="open" class="risk-sim-backdrop" @click.self="handleClose">
+      <section class="risk-sim-modal" role="dialog" aria-modal="true" aria-label="风险规则仿真测试">
+        <header class="risk-sim-head">
+          <div class="risk-sim-title">
+            <span>独立仿真测试</span>
+            <h3>{{ rule?.name || '风险规则' }}</h3>
+            <p>临时对话只做单据识别和风险规则执行，不创建报销单，不写入主工作台会话。</p>
+          </div>
+          <button type="button" class="risk-sim-icon-btn" aria-label="关闭" @click="handleClose">
+            <i class="mdi mdi-close"></i>
+          </button>
+        </header>
+
+        <div class="risk-sim-meta">
+          <span>版本：{{ testVersion }}</span>
+          <span :class="severityTone">{{ rule?.riskRuleSeverityLabel || '中风险' }}</span>
+          <span>{{ requiresAttachment ? '测试需附件' : '文字测试' }}</span>
+          <span>{{ latestSummary?.test_passed ? '测试结论已保存' : '测试结论未保存' }}</span>
+          <span>Session：{{ sessionShortId }}</span>
+        </div>
+
+        <div class="risk-sim-main">
+          <section class="risk-sim-dialog-panel">
+            <div ref="messageListRef" class="risk-sim-message-list" aria-live="polite">
+              <article
+                v-for="message in messages"
+                :key="message.id"
+                class="risk-sim-message-row"
+                :class="message.role"
+              >
+                <span class="risk-sim-avatar" aria-hidden="true">
+                  <i :class="message.role === 'assistant' ? 'mdi mdi-shield-search-outline' : 'mdi mdi-account-outline'"></i>
+                </span>
+
+                <div class="risk-sim-bubble">
+                  <header>
+                    <strong>{{ message.role === 'assistant' ? '风险仿真助手' : '我' }}</strong>
+                    <time>{{ message.time }}</time>
+                  </header>
+
+                  <p v-if="message.text">{{ message.text }}</p>
+
+                  <div v-if="message.attachments?.length" class="risk-sim-message-files">
+                    <span v-for="file in message.attachments" :key="`${message.id}-${file.id}`">
+                      <i class="mdi mdi-file-document-outline"></i>
+                      {{ file.name }}
+                    </span>
+                  </div>
+
+                  <div v-if="message.result" class="risk-sim-result-card" :class="message.result.severity">
+                    <div class="risk-sim-result-head">
+                      <div>
+                        <span>{{ message.result.ready === false ? '流程状态' : '识别结果' }}</span>
+                        <strong>
+                          {{ message.result.ready === false ? '待补充后再判断' : (message.result.hit ? '命中风险' : '未命中风险') }}
+                        </strong>
+                      </div>
+                      <b>{{ message.result.severity_label }}</b>
+                    </div>
+
+                    <p v-if="message.result.blocking_reason" class="risk-sim-blocking-message">
+                      {{ message.result.blocking_reason }}
+                    </p>
+
+                    <p v-if="message.result.message" class="risk-sim-result-message">
+                      {{ message.result.message }}
+                    </p>
+
+                    <div class="risk-sim-field-grid">
+                      <div v-for="item in buildResultFields(message.result)" :key="item.key">
+                        <span>{{ item.label }}</span>
+                        <strong>{{ item.value }}</strong>
+                      </div>
+                    </div>
+
+                    <div v-if="buildEvidenceItems(message.result).length" class="risk-sim-evidence">
+                      <span>判断依据</span>
+                      <ul>
+                        <li v-for="item in buildEvidenceItems(message.result)" :key="item">{{ item }}</li>
+                      </ul>
+                    </div>
+
+                    <div v-if="message.result.missing_fields?.length" class="risk-sim-missing-fields">
+                      <span>待补充字段</span>
+                      <b v-for="field in message.result.missing_fields" :key="field.key">
+                        {{ formatFieldLabel(field) }}
+                      </b>
+                    </div>
+                  </div>
+                </div>
+              </article>
+
+              <article v-if="busyAction === 'simulate' || recognitionBusy" class="risk-sim-message-row assistant">
+                <span class="risk-sim-avatar" aria-hidden="true">
+                  <i class="mdi mdi-shield-search-outline"></i>
+                </span>
+                <div class="risk-sim-bubble">
+                  <header>
+                    <strong>风险仿真助手</strong>
+                    <time>{{ currentTime }}</time>
+                  </header>
+                  <p class="risk-sim-thinking">
+                    <i class="mdi mdi-loading mdi-spin"></i>
+                    {{ recognitionBusy ? '正在识别临时单据...' : '正在调用规则执行器识别风险...' }}
+                  </p>
+                </div>
+              </article>
+            </div>
+
+            <div v-if="requiresAttachment && uploadedFiles.length" class="risk-sim-file-strip">
+              <span>本轮临时附件</span>
+              <div>
+                <button
+                  v-for="file in uploadedFiles"
+                  :key="file.id"
+                  type="button"
+                  class="risk-sim-file-chip"
+                  :class="file.status"
+                  :title="`${file.name} · ${formatFileSize(file.size)}`"
+                  @click="removeFile(file.id)"
+                >
+                  <i class="mdi mdi-file-document-outline"></i>
+                  <span>{{ file.name }}</span>
+                  <em>{{ resolveFileStatusLabel(file) }}</em>
+                  <i class="mdi mdi-close"></i>
+                </button>
+              </div>
+            </div>
+
+            <footer class="risk-sim-composer" :class="{ 'text-only': !requiresAttachment }">
+              <input
+                v-if="requiresAttachment"
+                ref="fileInputRef"
+                class="risk-sim-file-input"
+                type="file"
+                multiple
+                @change="handleFileChange"
+              />
+              <button
+                v-if="requiresAttachment"
+                type="button"
+                class="risk-sim-tool-btn"
+                :disabled="busy"
+                aria-label="上传临时单据"
+                @click="triggerFilePick"
+              >
+                <i class="mdi mdi-paperclip"></i>
+              </button>
+              <div class="risk-sim-composer-shell">
+                <textarea
+                  ref="composerRef"
+                  v-model="draft"
+                  :disabled="busy"
+                  rows="1"
+                  :placeholder="composerPlaceholder"
+                  @keydown.enter.exact.prevent="sendMessage"
+                ></textarea>
+              </div>
+              <button
+                type="button"
+                class="risk-sim-send-btn"
+                :disabled="busy || !canSend"
+                :title="sendBlockedReason"
+                aria-label="执行风险识别"
+                @click="sendMessage"
+              >
+                <i :class="busyAction === 'simulate' ? 'mdi mdi-loading mdi-spin' : 'mdi mdi-send'"></i>
+              </button>
+            </footer>
+          </section>
+
+          <aside class="risk-sim-context-panel">
+            <section>
+              <span>仿真流程</span>
+              <div class="risk-sim-step-list">
+                <div
+                  v-for="step in simulationSteps"
+                  :key="step.key"
+                  class="risk-sim-step"
+                  :class="step.status"
+                >
+                  <i :class="step.icon"></i>
+                  <div>
+                    <strong>{{ step.label }}</strong>
+                    <p>{{ step.description }}</p>
+                  </div>
+                </div>
+              </div>
+            </section>
+
+            <section v-if="recognizedDocuments.length || recognitionError">
+              <span>单据识别</span>
+              <div class="risk-sim-recognition-list">
+                <article v-for="item in recognizedDocuments" :key="item.filename">
+                  <strong>{{ item.document_type_label || item.scene_label || '已识别单据' }}</strong>
+                  <p>{{ item.filename }}</p>
+                  <small>{{ buildDocumentBrief(item) }}</small>
+                </article>
+                <p v-if="recognitionError" class="risk-sim-error-text">{{ recognitionError }}</p>
+              </div>
+            </section>
+
+            <section>
+              <span>规则边界</span>
+              <strong>{{ requiresAttachment ? '附件和文字合并判断' : '仅使用文字事实判断' }}</strong>
+              <p>{{ boundaryDescription }}</p>
+            </section>
+
+            <section>
+              <span>使用字段</span>
+              <div class="risk-sim-field-list">
+                <b v-for="field in displayFields" :key="field.key">{{ field.label }}</b>
+                <em v-if="!displayFields.length">当前规则未声明字段</em>
+              </div>
+            </section>
+
+            <section>
+              <span>关闭后清理</span>
+              <p>聊天记录、临时附件、识别结果会从弹窗内存中清空。</p>
+            </section>
+          </aside>
+        </div>
+
+        <footer class="risk-sim-foot">
+          <span>{{ lastSimulationHint }}</span>
+          <div>
+            <button type="button" class="risk-sim-secondary-btn" :disabled="busy" @click="resetConversation">
+              <i class="mdi mdi-refresh"></i>
+              <span>清空本轮</span>
+            </button>
+            <button
+              type="button"
+              class="risk-sim-primary-btn"
+              :disabled="busy || !activeSimulationResult?.ready"
+              @click="saveSimulationConclusion"
+            >
+              <i :class="busyAction === 'report' ? 'mdi mdi-loading mdi-spin' : 'mdi mdi-check-decagram-outline'"></i>
+              <span>{{ busyAction === 'report' ? '保存中' : '确认测试通过' }}</span>
+            </button>
+          </div>
+        </footer>
+      </section>
+    </div>
+  </Transition>
+</template>
+
+<script setup>
+import { computed, nextTick, ref, watch } from 'vue'
+
+import {
+  confirmRiskRuleTestReport,
+  runRiskRuleSampleTest,
+  simulateRiskRuleTest
+} from '../../services/agentAssets.js'
+import { recognizeOcrFiles } from '../../services/ocr.js'
+import { useToast } from '../../composables/useToast.js'
+import {
+  createId,
+  formatFileSize,
+  formatTestError,
+  formatTime
+} from './riskRuleTestDialogUtils.js'
+
+const props = defineProps({
+  open: {
+    type: Boolean,
+    default: false
+  },
+  rule: {
+    type: Object,
+    default: null
+  }
+})
+
+const emit = defineEmits(['close', 'report-saved'])
+const { toast } = useToast()
+
+const messageListRef = ref(null)
+const fileInputRef = ref(null)
+const composerRef = ref(null)
+const sessionId = ref('')
+const draft = ref('')
+const messages = ref([])
+const uploadedFiles = ref([])
+const recognizedDocuments = ref([])
+const recognitionError = ref('')
+const activeSimulationResult = ref(null)
+const latestSummary = ref(null)
+const busyAction = ref('')
+
+const requiresAttachment = computed(() => Boolean(props.rule?.riskRuleRequiresAttachment))
+const recognitionBusy = computed(() => busyAction.value === 'recognize')
+const busy = computed(() => Boolean(busyAction.value))
+const hasPendingFiles = computed(() => uploadedFiles.value.some((file) => file.status === 'recognizing'))
+const hasRecognizedFiles = computed(() => uploadedFiles.value.some((file) => file.status === 'recognized'))
+const hasFailedOnlyFiles = computed(() => uploadedFiles.value.length > 0 && uploadedFiles.value.every((file) => file.status === 'failed'))
+const canSend = computed(() => {
+  if (hasPendingFiles.value) return false
+  const hasText = Boolean(draft.value.trim())
+  if (requiresAttachment.value) {
+    return hasText && uploadedFiles.value.length > 0
+  }
+  return hasText
+})
+const fields = computed(() => (Array.isArray(props.rule?.riskRuleFields) ? props.rule.riskRuleFields : []))
+const displayFields = computed(() => fields.value.map((field) => ({
+  key: field.key,
+  label: formatFieldLabel(field)
+})))
+const composerPlaceholder = computed(() => requiresAttachment.value
+  ? '填写测试意图并上传附件，例如：请检查这张酒店发票是否与行程城市一致'
+  : '描述测试事实，例如：酒店发票城市上海，申报目的地北京，金额580元')
+const boundaryDescription = computed(() => requiresAttachment.value
+  ? '附件选择后不会立即识别，点击发送时才会和输入内容一起进入仿真；不会创建报销草稿或写入风险标记。'
+  : '这条规则不需要上传附件，测试窗口只根据输入文字执行规则；不会创建报销草稿或影响审批流。')
+const testVersion = computed(() => props.rule?.workingVersion || props.rule?.displayVersion || props.rule?.version || '-')
+const sessionShortId = computed(() => sessionId.value ? sessionId.value.slice(-8).toUpperCase() : '-')
+const currentTime = computed(() => formatTime())
+const severityTone = computed(() => `tone-${props.rule?.riskRuleSeverity || 'medium'}`)
+const sendBlockedReason = computed(() => {
+  if (hasPendingFiles.value) return '单据识别中，请稍后执行风险识别。'
+  if (requiresAttachment.value && !uploadedFiles.value.length) return '这条规则要求上传测试附件。'
+  if (requiresAttachment.value && !draft.value.trim()) return '请填写测试意图或关键事实，附件会和文字一起判断。'
+  if (!draft.value.trim()) return '请先描述测试单据或测试事实。'
+  return ''
+})
+const simulationSteps = computed(() => [
+  {
+    key: 'recognize',
+    label: '1. 单据识别',
+    description: buildRecognitionStepDescription(),
+    status: resolveRecognitionStepStatus(),
+    icon: recognitionBusy.value ? 'mdi mdi-loading mdi-spin' : 'mdi mdi-text-recognition'
+  },
+  {
+    key: 'fields',
+    label: '2. 字段确认',
+    description: buildFieldStepDescription(),
+    status: resolveFieldStepStatus(),
+    icon: 'mdi mdi-format-list-checks'
+  },
+  {
+    key: 'execute',
+    label: '3. 规则执行',
+    description: activeSimulationResult.value?.ready
+      ? '已使用规则执行器完成判断。'
+      : '字段齐备后才会执行规则。',
+    status: activeSimulationResult.value?.ready ? 'done' : 'pending',
+    icon: 'mdi mdi-shield-check-outline'
+  }
+])
+const lastSimulationHint = computed(() => {
+  if (!activeSimulationResult.value) {
+    return '本窗口是独立临时会话，关闭后会清空聊天记录和上传文件。'
+  }
+  if (activeSimulationResult.value.ready === false) {
+    return `最近一次仿真：${activeSimulationResult.value.summary || '待补充字段'}`
+  }
+  return activeSimulationResult.value.hit
+    ? `最近一次仿真：命中${activeSimulationResult.value.severity_label}`
+    : '最近一次仿真：未命中风险'
+})
+
+watch(
+  () => props.open,
+  (open) => {
+    if (open) {
+      initializeSession()
+    } else {
+      destroySession()
+    }
+  },
+  { immediate: true }
+)
+
+function initializeSession() {
+  sessionId.value = createId()
+  latestSummary.value = props.rule?.latestTestSummary || null
+  draft.value = ''
+  uploadedFiles.value = []
+  recognizedDocuments.value = []
+  recognitionError.value = ''
+  activeSimulationResult.value = null
+  busyAction.value = ''
+  messages.value = [buildMessage('assistant', buildWelcomeMessage())]
+  clearFileInput()
+  nextTick(() => {
+    scrollMessagesToBottom()
+    composerRef.value?.focus()
+  })
+}
+
+function destroySession() {
+  draft.value = ''
+  messages.value = []
+  uploadedFiles.value = []
+  recognizedDocuments.value = []
+  recognitionError.value = ''
+  activeSimulationResult.value = null
+  busyAction.value = ''
+  sessionId.value = ''
+  clearFileInput()
+}
+
+function resetConversation() {
+  initializeSession()
+}
+
+function handleClose() {
+  destroySession()
+  emit('close')
+}
+
+function triggerFilePick() {
+  if (!requiresAttachment.value) return
+  fileInputRef.value?.click()
+}
+
+function handleFileChange(event) {
+  if (!requiresAttachment.value) {
+    clearFileInput()
+    return
+  }
+  const input = event.target
+  const incoming = Array.from(input?.files || [])
+  if (!incoming.length) return
+  const nextFiles = incoming.map((file) => ({
+    id: createId(),
+    name: file.name,
+    size: file.size,
+    contentType: file.type || '',
+    status: 'pending',
+    statusText: '待发送',
+    ocrDocument: null,
+    error: '',
+    file
+  }))
+  uploadedFiles.value = [...uploadedFiles.value, ...nextFiles].slice(0, 12)
+  clearFileInput()
+}
+
+function removeFile(fileId) {
+  uploadedFiles.value = uploadedFiles.value.filter((file) => file.id !== fileId)
+}
+
+async function sendMessage() {
+  if (!props.rule?.id || !canSend.value || busy.value) return
+  const activeSessionId = sessionId.value
+  const text = draft.value.trim()
+  const runFiles = requiresAttachment.value ? uploadedFiles.value.slice() : []
+  messages.value.push(buildMessage('user', text, { attachments: runFiles }))
+  draft.value = ''
+  await nextTick()
+  scrollMessagesToBottom()
+
+  try {
+    let attachments = []
+    if (requiresAttachment.value) {
+      const filesForRecognition = runFiles.filter((file) => file.status !== 'recognized')
+      if (filesForRecognition.length) {
+        busyAction.value = 'recognize'
+        await recognizeTemporaryFiles(filesForRecognition, activeSessionId)
+        if (!isActiveSession(activeSessionId)) return
+      }
+      attachments = runFiles.map(toAttachmentPayload)
+    }
+    busyAction.value = 'simulate'
+    const result = await simulateRiskRuleTest(props.rule.id, {
+      version: testVersion.value,
+      message: text,
+      attachments
+    })
+    if (!isActiveSession(activeSessionId)) return
+    activeSimulationResult.value = result
+    messages.value.push(buildMessage('assistant', result.summary, { result }))
+  } catch (error) {
+    if (!isActiveSession(activeSessionId)) return
+    messages.value.push(buildMessage('assistant', formatTestError(error, '仿真识别失败，请稍后重试。')))
+  } finally {
+    if (isActiveSession(activeSessionId)) {
+      busyAction.value = ''
+      await nextTick()
+      scrollMessagesToBottom()
+      composerRef.value?.focus()
+    }
+  }
+}
+
+async function saveSimulationConclusion() {
+  if (!props.rule?.id || !activeSimulationResult.value?.ready || busy.value) return
+  const activeSessionId = sessionId.value
+  busyAction.value = 'report'
+  try {
+    const sample = await runRiskRuleSampleTest(props.rule.id, {
+      version: testVersion.value,
+      cases: []
+    })
+    if (!sample?.passed) {
+      messages.value.push(buildMessage('assistant', '系统样例复核未通过，暂不能保存测试通过结论。请调整规则后重新仿真。'))
+      return
+    }
+    const report = await confirmRiskRuleTestReport(props.rule.id, {
+      version: testVersion.value,
+      confirm_passed: true,
+      note: '通过独立对话仿真后确认测试通过；聊天记录和临时附件不保存。'
+    })
+    if (!isActiveSession(activeSessionId)) return
+    latestSummary.value = {
+      ...(latestSummary.value || {}),
+      sample,
+      report,
+      test_passed: true
+    }
+    emit('report-saved', latestSummary.value)
+    messages.value.push(buildMessage('assistant', '测试通过结论已保存。这个动作只更新规则生命周期状态，不保存本轮聊天记录或上传文件。'))
+  } catch (error) {
+    if (!isActiveSession(activeSessionId)) return
+    messages.value.push(buildMessage('assistant', formatTestError(error, '测试结论保存失败，请稍后重试。')))
+  } finally {
+    if (isActiveSession(activeSessionId)) {
+      busyAction.value = ''
+      await nextTick()
+      scrollMessagesToBottom()
+    }
+  }
+}
+
+async function recognizeTemporaryFiles(files, activeSessionId) {
+  if (!files.length) return
+  recognitionError.value = ''
+  files.forEach((file) => {
+    const target = uploadedFiles.value.find((item) => item.id === file.id)
+    if (!target) return
+    target.status = 'recognizing'
+    target.statusText = '识别中'
+    target.error = ''
+  })
+  try {
+    const payload = await recognizeOcrFiles(files.map((file) => file.file), {
+      timeoutMs: 90000,
+      timeoutMessage: '单据 OCR 识别超时，请补充关键字段后再执行规则。'
+    })
+    if (!isActiveSession(activeSessionId)) return
+    const documents = normalizeOcrDocuments(payload)
+    recognizedDocuments.value = mergeRecognizedDocuments(recognizedDocuments.value, documents)
+    files.forEach((file, index) => {
+      const document = documents[index] || null
+      const target = uploadedFiles.value.find((item) => item.id === file.id)
+      if (!target) return
+      if (document && documentHasMeaningfulText(document)) {
+        target.status = 'recognized'
+        target.statusText = '已识别'
+        target.ocrDocument = document
+        target.error = ''
+      } else {
+        target.status = 'failed'
+        target.statusText = '识别不足'
+        target.ocrDocument = document
+        target.error = '未提取到足够文本'
+      }
+    })
+    const recognizedCount = files.filter((file) => {
+      const target = uploadedFiles.value.find((item) => item.id === file.id)
+      return target?.status === 'recognized'
+    }).length
+    messages.value.push(buildMessage(
+      'assistant',
+      recognizedCount
+        ? `已完成 ${recognizedCount} 份临时单据识别。请核对右侧识别字段，字段不足时可以直接在输入框补充。`
+        : '上传文件没有提取到足够字段，暂不能直接执行规则。请在输入框补充票据城市、金额、发票号等关键信息。'
+    ))
+  } catch (error) {
+    if (!isActiveSession(activeSessionId)) return
+    recognitionError.value = formatTestError(error, '单据识别失败，请补充关键字段后再执行规则。')
+    files.forEach((file) => {
+      const target = uploadedFiles.value.find((item) => item.id === file.id)
+      if (!target) return
+      target.status = 'failed'
+      target.statusText = '识别失败'
+      target.error = recognitionError.value
+    })
+    messages.value.push(buildMessage('assistant', recognitionError.value))
+  } finally {
+    if (isActiveSession(activeSessionId)) {
+      await nextTick()
+      scrollMessagesToBottom()
+    }
+  }
+}
+
+function buildMessage(role, text, extra = {}) {
+  return {
+    id: createId(),
+    role,
+    text,
+    time: formatTime(),
+    ...extra
+  }
+}
+
+function buildResultFields(result) {
+  const values = result?.field_values && typeof result.field_values === 'object'
+    ? result.field_values
+    : {}
+  return Object.entries(values).slice(0, 8).map(([key, value]) => ({
+    key,
+    label: formatFieldLabel(fields.value.find((field) => field.key === key) || { key }),
+    value: Array.isArray(value) ? value.join('、') : String(value ?? '-')
+  }))
+}
+
+function buildEvidenceItems(result) {
+  const evidence = result?.evidence && typeof result.evidence === 'object'
+    ? result.evidence
+    : {}
+  const items = []
+  if (Array.isArray(evidence.failed_conditions)) {
+    evidence.failed_conditions.slice(0, 3).forEach((condition) => {
+      const left = Array.isArray(condition.left_values) ? condition.left_values.join('、') : '-'
+      const right = Array.isArray(condition.right_values) ? condition.right_values.join('、') : '-'
+      items.push(`${formatFieldName(condition.left)}：${left}；${formatFieldName(condition.right)}：${right}`)
+    })
+  }
+  if (Array.isArray(evidence.missing_fields)) {
+    evidence.missing_fields.slice(0, 5).forEach((field) => {
+      items.push(`${formatFieldName(field)} 缺失`)
+    })
+  }
+  if (Array.isArray(evidence.keyword_hits)) {
+    items.push(`命中关键词：${evidence.keyword_hits.join('、')}`)
+  }
+  if (evidence.condition_summary) {
+    items.push(String(evidence.condition_summary))
+  }
+  return [...new Set(items)].slice(0, 5)
+}
+
+function formatFieldLabel(field) {
+  const key = String(field?.key || '').trim()
+  const label = String(field?.display || field?.label || '').trim()
+  if (!key) return label || '-'
+  if (!label || label === key) return key
+  return label.includes(`[${key}]`) ? label : `${label}[${key}]`
+}
+
+function formatFieldName(key) {
+  return formatFieldLabel(fields.value.find((field) => field.key === key) || { key })
+}
+
+function toAttachmentPayload(file) {
+  const document = file.ocrDocument || {}
+  return {
+    id: file.id,
+    name: file.name,
+    size: file.size,
+    content_type: file.contentType,
+    note: file.error || '',
+    recognition_status: file.status,
+    ocr_text: document.text || '',
+    summary: document.summary || '',
+    document_type: document.document_type || '',
+    document_type_label: document.document_type_label || '',
+    scene_code: document.scene_code || '',
+    scene_label: document.scene_label || '',
+    avg_score: document.avg_score || 0,
+    document_fields: Array.isArray(document.document_fields) ? document.document_fields : []
+  }
+}
+
+function normalizeOcrDocuments(payload) {
+  const documents = Array.isArray(payload?.documents) ? payload.documents : []
+  return documents.map((item) => ({
+    filename: String(item?.filename || '').trim(),
+    summary: String(item?.summary || '').trim(),
+    text: String(item?.text || '').trim(),
+    avg_score: Number(item?.avg_score || 0),
+    document_type: String(item?.document_type || 'other').trim() || 'other',
+    document_type_label: String(item?.document_type_label || '').trim(),
+    scene_code: String(item?.scene_code || 'other').trim() || 'other',
+    scene_label: String(item?.scene_label || '').trim(),
+    document_fields: Array.isArray(item?.document_fields)
+      ? item.document_fields
+        .map((field) => ({
+          key: String(field?.key || '').trim(),
+          label: String(field?.label || '').trim(),
+          value: String(field?.value || '').trim()
+        }))
+        .filter((field) => field.key && field.label && field.value)
+      : [],
+    warnings: Array.isArray(item?.warnings) ? item.warnings : []
+  }))
+}
+
+function mergeRecognizedDocuments(current, incoming) {
+  const next = [...current]
+  incoming.forEach((document) => {
+    const index = next.findIndex((item) => item.filename === document.filename)
+    if (index >= 0) {
+      next.splice(index, 1, document)
+    } else {
+      next.push(document)
+    }
+  })
+  return next
+}
+
+function documentHasMeaningfulText(document) {
+  return Boolean(
+    String(document?.text || document?.summary || '').trim() ||
+    (Array.isArray(document?.document_fields) && document.document_fields.length)
+  )
+}
+
+function buildDocumentBrief(document) {
+  const fields = Array.isArray(document?.document_fields) ? document.document_fields : []
+  if (fields.length) {
+    return fields.slice(0, 4).map((field) => `${field.label}：${field.value}`).join('；')
+  }
+  return String(document?.summary || document?.text || '未提取到结构化字段').slice(0, 120)
+}
+
+function resolveFileStatusLabel(file) {
+  return file.statusText || {
+    pending: '待发送',
+    recognizing: '识别中',
+    recognized: '已识别',
+    failed: '识别失败'
+  }[file.status] || '待识别'
+}
+
+function buildRecognitionStepDescription() {
+  if (!requiresAttachment.value) return '当前规则不需要附件，直接根据文字测试事实抽取字段。'
+  if (recognitionBusy.value) return '正在读取临时附件并提取 OCR 字段。'
+  if (hasRecognizedFiles.value) return `已识别 ${recognizedDocuments.value.length} 份临时单据。`
+  if (hasFailedOnlyFiles.value) return '识别不足，请在对话中补充字段。'
+  if (uploadedFiles.value.length) return '附件已加入本轮，点击发送后会和文字一起识别。'
+  return '需要上传测试附件，并填写测试意图后执行。'
+}
+
+function resolveRecognitionStepStatus() {
+  if (!requiresAttachment.value) return 'done'
+  if (recognitionBusy.value) return 'running'
+  if (hasRecognizedFiles.value) return 'done'
+  if (hasFailedOnlyFiles.value || recognitionError.value) return 'warning'
+  return 'pending'
+}
+
+function buildFieldStepDescription() {
+  if (activeSimulationResult.value?.recognized_fields?.length) {
+    return `已确认 ${activeSimulationResult.value.recognized_fields.length} 个字段。`
+  }
+  if (draft.value.trim()) return '将使用你输入的文字抽取测试字段。'
+  return '识别完成或补充字段后进入确认。'
+}
+
+function resolveFieldStepStatus() {
+  if (activeSimulationResult.value?.ready) return 'done'
+  if (activeSimulationResult.value?.missing_fields?.length) return 'warning'
+  if (hasRecognizedFiles.value || draft.value.trim()) return 'running'
+  return 'pending'
+}
+
+function buildWelcomeMessage() {
+  if (requiresAttachment.value) {
+    return '这条规则要求测试附件。请先上传临时票据并填写测试意图，点击发送后我会统一识别附件和文字，再交给规则执行器判断。'
+  }
+  return '这条规则不需要上传附件。你可以直接输入测试事实，我只会执行风险识别，不创建单据、不写入主工作台会话。'
+}
+
+function clearFileInput() {
+  if (fileInputRef.value) {
+    fileInputRef.value.value = ''
+  }
+}
+
+function scrollMessagesToBottom() {
+  const target = messageListRef.value
+  if (target) {
+    target.scrollTop = target.scrollHeight
+  }
+}
+
+function isActiveSession(activeSessionId) {
+  return props.open && activeSessionId && activeSessionId === sessionId.value
+}
+
+</script>
+
+<style src="../../assets/styles/components/risk-rule-test-dialog.css"></style>
+
--- a/web/src/components/shared/riskRuleTestDialogUtils.js
+++ b/web/src/components/shared/riskRuleTestDialogUtils.js
@@ -0,0 +1,28 @@
+export function createId() {
+  if (typeof crypto !== 'undefined' && crypto.randomUUID) {
+    return crypto.randomUUID()
+  }
+  return `${Date.now()}-${Math.random().toString(16).slice(2)}`
+}
+
+export function formatFileSize(size) {
+  const value = Number(size || 0)
+  if (value < 1024) return `${value}B`
+  if (value < 1024 * 1024) return `${(value / 1024).toFixed(1)}KB`
+  return `${(value / 1024 / 1024).toFixed(1)}MB`
+}
+
+export function formatTestError(error, fallback) {
+  const message = String(error?.message || '').trim()
+  if (/not\s*found/i.test(message)) {
+    return '测试接口暂未加载或规则详情已失效，请刷新规则详情后再试。'
+  }
+  return message || fallback
+}
+
+export function formatTime() {
+  return new Intl.DateTimeFormat('zh-CN', {
+    hour: '2-digit',
+    minute: '2-digit'
+  }).format(new Date())
+}
--- a/web/src/composables/useAppShell.js
+++ b/web/src/composables/useAppShell.js
@@ -29,6 +29,7 @@ export function useAppShell() {
  })
  const smartEntrySessionId = ref(0)
  const smartEntryInvalidatedDraftClaimId = ref('')
+  const selectedRequestSnapshot = ref(null)

  const { activeView, currentView, setView } = useNavigation()
  const {
@@ -60,14 +61,32 @@ export function useAppShell() {
    const rawRequest = requests.value.find(
      (item) => String(item.claimId || '').trim() === requestId || String(item.id || '').trim() === requestId
    )
-    return normalizeRequestForUi(rawRequest)
+    const normalizedRequest = normalizeRequestForUi(rawRequest)
+    if (normalizedRequest) {
+      return normalizedRequest
+    }
+
+    const snapshot = normalizeRequestForUi(selectedRequestSnapshot.value)
+    if (
+      snapshot
+      && (
+        String(snapshot.claimId || '').trim() === requestId
+        || String(snapshot.id || '').trim() === requestId
+        || String(snapshot.documentNo || '').trim() === requestId
+      )
+    ) {
+      return snapshot
+    }
+
+    return null
  })

-  const detailMode = computed(() => route.name === 'app-request-detail')
+  const detailMode = computed(() => ['app-request-detail', 'app-document-detail'].includes(route.name))
  const logDetailMode = computed(() => route.name === 'app-log-detail')
  const detailAlerts = computed(() => (detailMode.value ? buildDetailAlerts(selectedRequest.value) : []))

  const requestsListActive = computed(() => activeView.value === 'requests' && !detailMode.value)
+  const documentsListActive = computed(() => activeView.value === 'documents' && !detailMode.value)
  const workbenchActive = computed(() => activeView.value === 'workbench')

  watch(requestsListActive, (isActive, wasActive) => {
@@ -76,6 +95,12 @@ export function useAppShell() {
    }
  })

+  watch(documentsListActive, (isActive, wasActive) => {
+    if (isActive && !wasActive) {
+      void reloadRequests()
+    }
+  })
+
  watch(workbenchActive, (isActive, wasActive) => {
    if (isActive && !wasActive) {
      void reloadRequests()
@@ -237,21 +262,23 @@ export function useAppShell() {
      smartEntryOpen.value = false
      void refreshApprovalInbox()
      toast(`${claimNo || '该'}单据已完成 AI预审${approvalStage ? `，当前节点：${approvalStage}` : '，并已提交审批'}。`)
-      router.push({ name: 'app-requests' })
+      router.push({ name: activeView.value === 'documents' ? 'app-documents' : 'app-requests' })
      return
    }
    toast(`${claimNo || '该'}单据已保存为草稿，可继续上传票据或补充信息。`)
  }

  function openRequestDetail(request) {
+    selectedRequestSnapshot.value = request || null
+    const routeName = activeView.value === 'documents' ? 'app-document-detail' : 'app-request-detail'
    router.push({
-      name: 'app-request-detail',
+      name: routeName,
      params: { requestId: request.claimId || request.id }
    })
  }

  function closeRequestDetail() {
-    router.push({ name: 'app-requests' })
+    router.push({ name: activeView.value === 'documents' ? 'app-documents' : 'app-requests' })
  }

  async function handleRequestUpdated() {
@@ -268,7 +295,8 @@ export function useAppShell() {

    await reloadRequests()
    void refreshApprovalInbox()
-    router.push({ name: 'app-requests' })
+    selectedRequestSnapshot.value = null
+    router.push({ name: activeView.value === 'documents' ? 'app-documents' : 'app-requests' })
  }

  return {
--- a/web/src/composables/useNavigation.js
+++ b/web/src/composables/useNavigation.js
@@ -3,7 +3,7 @@ import { useRoute, useRouter } from 'vue-router'

 import { icons } from '../data/icons.js'

-export const appViews = ['overview', 'workbench', 'requests', 'approval', 'archive', 'policies', 'audit', 'employees', 'logs', 'settings']
+export const appViews = ['overview', 'workbench', 'documents', 'requests', 'approval', 'archive', 'policies', 'audit', 'employees', 'logs', 'settings']

 export const navItems = [
  {
@@ -22,6 +22,14 @@ export const navItems = [
    title: '个人工作台',
    desc: '聚焦当前待办、快捷操作与助手入口。'
  },
+  {
+    id: 'documents',
+    label: '单据中心',
+    navHint: '统一查看申请、报销、审批与归档',
+    icon: icons.file,
+    title: '单据中心',
+    desc: '统一查看申请、报销、审批与归档。'
+  },
  {
    id: 'requests',
    label: '报销中心',
@@ -91,6 +99,7 @@ export const navItems = [
 const viewRouteNames = {
  overview: 'app-overview',
  workbench: 'app-workbench',
+  documents: 'app-documents',
  requests: 'app-requests',
  approval: 'app-approval',
  archive: 'app-archive',
@@ -106,6 +115,7 @@ const routeNameViews = Object.fromEntries(
 )

 routeNameViews['app-request-detail'] = 'requests'
+routeNameViews['app-document-detail'] = 'documents'
 routeNameViews['app-log-detail'] = 'logs'

 export function resolveAppViewFromRoute(route) {
--- a/web/src/composables/useSettings.js
+++ b/web/src/composables/useSettings.js
@@ -114,6 +114,7 @@ export function useSettings() {
      adminForm: { ...pageState.value.adminForm },
      sessionForm: { ...pageState.value.sessionForm },
      llmForm: buildLlmPayload(pageState.value.llmForm),
+      hermesForm: { ...pageState.value.hermesForm },
      renderForm: buildRenderPayload(pageState.value.renderForm),
      logForm: { ...pageState.value.logForm },
      mailForm: { ...pageState.value.mailForm }
--- a/web/src/router/index.js
+++ b/web/src/router/index.js
@@ -10,7 +10,7 @@ import LoginRouteView from '../views/LoginRouteView.vue'
 import SetupRouteView from '../views/SetupRouteView.vue'

 const appChildRoutes = appViews
-  .filter((view) => view !== 'requests')
+  .filter((view) => view !== 'requests' && view !== 'documents')
  .map((view) => ({
    path: view,
    name: `app-${view}`,
@@ -51,6 +51,24 @@ const router = createRouter({
      path: '/app',
      redirect: { name: 'app-overview' }
    },
+    {
+      path: '/app/documents',
+      name: 'app-documents',
+      component: AppShellRouteView,
+      meta: {
+        requiresAuth: true,
+        appView: 'documents'
+      }
+    },
+    {
+      path: '/app/documents/:requestId',
+      name: 'app-document-detail',
+      component: AppShellRouteView,
+      meta: {
+        requiresAuth: true,
+        appView: 'documents'
+      }
+    },
    {
      path: '/app/requests',
      name: 'app-requests',
--- a/web/src/services/agentAssets.js
+++ b/web/src/services/agentAssets.js
@@ -154,6 +154,72 @@ export function generateRiskRuleAsset(payload, options = {}) {
  })
 }

+export function fetchRiskRuleLatestTest(assetId) {
+  return apiRequest(`/agent-assets/${assetId}/risk-rule-tests/latest`)
+}
+
+export function runRiskRuleSampleTest(assetId, payload, options = {}) {
+  return apiRequest(`/agent-assets/${assetId}/risk-rule-tests/sample`, {
+    method: 'POST',
+    body: JSON.stringify(payload || {}),
+    headers: buildWriteHeaders(options)
+  })
+}
+
+export function simulateRiskRuleTest(assetId, payload, options = {}) {
+  return apiRequest(`/agent-assets/${assetId}/risk-rule-tests/simulate`, {
+    method: 'POST',
+    body: JSON.stringify(payload || {}),
+    headers: buildWriteHeaders(options)
+  })
+}
+
+export function runRiskRuleScenarioTest(assetId, payload, options = {}) {
+  return apiRequest(`/agent-assets/${assetId}/risk-rule-tests/scenario`, {
+    method: 'POST',
+    body: JSON.stringify(payload || {}),
+    headers: buildWriteHeaders(options)
+  })
+}
+
+export function confirmRiskRuleTestReport(assetId, payload, options = {}) {
+  return apiRequest(`/agent-assets/${assetId}/risk-rule-tests/report`, {
+    method: 'POST',
+    body: JSON.stringify(payload || {}),
+    headers: buildWriteHeaders(options)
+  })
+}
+
+export function deleteAgentAsset(assetId, options = {}) {
+  return apiRequest(`/agent-assets/${assetId}`, {
+    method: 'DELETE',
+    headers: buildWriteHeaders(options)
+  })
+}
+
+export function returnRiskRuleAsset(assetId, payload, options = {}) {
+  return apiRequest(`/agent-assets/${assetId}/return`, {
+    method: 'POST',
+    body: JSON.stringify(payload || {}),
+    headers: buildWriteHeaders(options)
+  })
+}
+
+export function publishRiskRuleAsset(assetId, options = {}) {
+  return apiRequest(`/agent-assets/${assetId}/publish`, {
+    method: 'POST',
+    headers: buildWriteHeaders(options)
+  })
+}
+
+export function setRiskRuleAssetEnabled(assetId, enabled, options = {}) {
+  return apiRequest(`/agent-assets/${assetId}/risk-rule-enabled`, {
+    method: 'POST',
+    body: JSON.stringify({ enabled: Boolean(enabled) }),
+    headers: buildWriteHeaders(options)
+  })
+}
+
 export function fetchAgentAssetSpreadsheetChangeRecords(assetId, limit = 30) {
  return apiRequest(
    `/agent-assets/${assetId}/spreadsheet/change-records${buildQuery({ limit })}`
--- a/web/src/utils/accessControl.js
+++ b/web/src/utils/accessControl.js
@@ -1,6 +1,7 @@
 export const DEFAULT_APP_VIEW_ORDER = [
  'overview',
  'workbench',
+  'documents',
  'requests',
  'approval',
  'archive',
@@ -11,7 +12,7 @@ export const DEFAULT_APP_VIEW_ORDER = [
  'settings'
 ]

-const ALWAYS_VISIBLE_VIEWS = new Set(['workbench', 'requests', 'policies'])
+const ALWAYS_VISIBLE_VIEWS = new Set(['workbench', 'documents', 'requests', 'policies'])
 const VIEW_ROLE_RULES = {
  overview: ['finance', 'executive'],
  approval: ['approver', 'finance', 'executive'],
--- a/web/src/utils/documentCenterTime.js
+++ b/web/src/utils/documentCenterTime.js
@@ -0,0 +1,76 @@
+function toDate(value) {
+  if (!value) {
+    return null
+  }
+
+  const nextDate = new Date(value)
+  return Number.isNaN(nextDate.getTime()) ? null : nextDate
+}
+
+export function extractDateText(value) {
+  const matched = String(value || '').match(/\d{4}-\d{2}-\d{2}/)
+  return matched ? matched[0] : ''
+}
+
+export function formatDocumentListTime(value) {
+  const raw = String(value || '').trim()
+  if (!raw || raw === '待补充') {
+    return '待补充'
+  }
+
+  const date = toDate(raw)
+  if (date) {
+    const month = String(date.getMonth() + 1).padStart(2, '0')
+    const day = String(date.getDate()).padStart(2, '0')
+    const hour = String(date.getHours()).padStart(2, '0')
+    const minute = String(date.getMinutes()).padStart(2, '0')
+    return `${month}-${day} ${hour}:${minute}`
+  }
+
+  return raw.replace(/^\d{4}-/, '').slice(0, 11)
+}
+
+export function resolveDocumentSortTime(value) {
+  const date = toDate(value)
+  return date ? date.getTime() : 0
+}
+
+export function formatDocumentDurationSince(value, now = Date.now()) {
+  const startAt = toDate(value)
+  if (!startAt) {
+    return ''
+  }
+
+  const diffMs = Math.max(0, Number(now) - startAt.getTime())
+  const totalMinutes = Math.floor(diffMs / (60 * 1000))
+  if (totalMinutes < 1) {
+    return '刚刚'
+  }
+
+  const days = Math.floor(totalMinutes / (24 * 60))
+  const hours = Math.floor((totalMinutes % (24 * 60)) / 60)
+  const minutes = totalMinutes % 60
+
+  if (days > 0) {
+    return hours > 0 ? `${days}天${hours}小时` : `${days}天`
+  }
+
+  if (hours > 0) {
+    return minutes > 0 ? `${hours}小时${minutes}分钟` : `${hours}小时`
+  }
+
+  return `${minutes}分钟`
+}
+
+export function resolveDocumentStayTimeDisplay(row, now = Date.now()) {
+  const currentStep = Array.isArray(row?.progressSteps)
+    ? row.progressSteps.find((step) => step?.current)
+    : null
+  const stepTime = String(currentStep?.time || '').trim()
+  if (stepTime.startsWith('停留')) {
+    return stepTime.replace(/^停留\s*/, '') || '待计算'
+  }
+
+  const startedAt = row?.updatedAt || row?.submittedAt || row?.createdAt || row?.applyTime
+  return formatDocumentDurationSince(startedAt, now) || '待计算'
+}
--- a/web/src/utils/expenseApplicationOntology.js
+++ b/web/src/utils/expenseApplicationOntology.js
@@ -0,0 +1,158 @@
+const EXPENSE_TYPE_LABELS = {
+  travel: '差旅费',
+  hotel: '住宿费',
+  transport: '交通费',
+  meal: '业务招待费',
+  entertainment: '业务招待费',
+  meeting: '会务费',
+  office: '办公用品费',
+  training: '培训费',
+  communication: '通讯费',
+  welfare: '福利费',
+  other: '其他费用'
+}
+
+const SLOT_LABELS = {
+  expense_type: '费用场景',
+  amount: '申请金额',
+  time_range: '业务时间',
+  reason: '申请事由',
+  attachments: '附件说明',
+  customer_name: '客户名称',
+  participants: '参与人员'
+}
+
+const PRE_APPROVAL_TYPES = new Set(['travel', 'meeting', 'office', 'training'])
+const ATTACHMENT_REQUIRED_TYPES = new Set(['meeting', 'training'])
+
+export const APPLICATION_EXAMPLES = [
+  '申请下周去北京做客户现场验收，差旅预算18000元',
+  '申请上海产品发布会会务费32000元，需要场地和物料',
+  '申请部门集中采购办公用品4800元，用于新员工入职'
+]
+
+export function buildExpenseApplicationOntologyContext(currentUser = {}) {
+  return {
+    document_type: 'expense_application',
+    application_stage: 'pre_approval',
+    conversation_scenario: 'expense',
+    entry_source: 'documents_application',
+    role_codes: Array.isArray(currentUser.roleCodes) ? currentUser.roleCodes : [],
+    is_admin: Boolean(currentUser.isAdmin),
+    name: currentUser.name || '',
+    role: currentUser.role || '',
+    department: currentUser.department || currentUser.departmentName || '',
+    department_name: currentUser.department || currentUser.departmentName || '',
+    position: currentUser.position || '',
+    grade: currentUser.grade || '',
+    employee_no: currentUser.employeeNo || currentUser.employee_no || ''
+  }
+}
+
+export function resolveEntity(ontology, type) {
+  const entities = Array.isArray(ontology?.entities) ? ontology.entities : []
+  return entities.find((item) => item?.type === type) || null
+}
+
+export function resolveConstraint(ontology, field) {
+  const constraints = Array.isArray(ontology?.constraints) ? ontology.constraints : []
+  return constraints.find((item) => item?.field === field) || null
+}
+
+export function resolveExpenseTypeCode(ontology) {
+  const entity = resolveEntity(ontology, 'expense_type')
+  return String(entity?.normalized_value || entity?.value || 'other').trim() || 'other'
+}
+
+export function resolveExpenseTypeLabel(code) {
+  return EXPENSE_TYPE_LABELS[String(code || '').trim()] || EXPENSE_TYPE_LABELS.other
+}
+
+export function resolveApplicationAmount(ontology) {
+  const amountEntity = resolveEntity(ontology, 'amount')
+  const amountConstraint = resolveConstraint(ontology, 'amount')
+  const rawValue = amountEntity?.normalized_value || amountEntity?.value || amountConstraint?.value || ''
+  const numericValue = Number(String(rawValue).replace(/[^\d.]/g, ''))
+  return {
+    raw: String(rawValue || '').trim(),
+    value: Number.isFinite(numericValue) ? numericValue : 0
+  }
+}
+
+export function resolveTimeRangeText(ontology) {
+  const range = ontology?.time_range || {}
+  if (range.start_date && range.end_date) {
+    return range.start_date === range.end_date
+      ? range.start_date
+      : `${range.start_date} 至 ${range.end_date}`
+  }
+  return String(range.raw || '').trim()
+}
+
+export function resolveAttachmentPolicy(expenseTypeCode, amount = 0) {
+  const code = String(expenseTypeCode || '').trim()
+  if (ATTACHMENT_REQUIRED_TYPES.has(code)) {
+    return {
+      level: 'required',
+      label: '必须提交',
+      description: code === 'meeting'
+        ? '需补充会议通知、议程、参会范围或预算说明。'
+        : '需补充培训通知、课程说明、报价或审批依据。'
+    }
+  }
+  if (code === 'office' && amount >= 5000) {
+    return {
+      level: 'required',
+      label: '必须提交',
+      description: '办公采购金额较高，需补充采购清单、报价或预算说明。'
+    }
+  }
+  if (code === 'travel') {
+    return {
+      level: 'optional',
+      label: '说明可选',
+      description: '可先提交出差目的、时间和预算；行程或邀请材料可作为补充说明。'
+    }
+  }
+  return {
+    level: 'none',
+    label: '无需附件',
+    description: '当前申请事项可先不提交附件，后续报销阶段再按票据要求补充。'
+  }
+}
+
+export function buildApplicationFieldsFromOntology(ontology, prompt, currentUser = {}) {
+  const expenseTypeCode = resolveExpenseTypeCode(ontology)
+  const amount = resolveApplicationAmount(ontology)
+  const locationEntity = resolveEntity(ontology, 'location')
+  const documentTypeEntity = resolveEntity(ontology, 'document_type')
+  const workflowStageEntity = resolveEntity(ontology, 'workflow_stage')
+  const attachmentPolicy = resolveAttachmentPolicy(expenseTypeCode, amount.value)
+
+  return {
+    documentType: documentTypeEntity?.normalized_value || 'expense_application',
+    documentTypeLabel: documentTypeEntity?.value || '费用申请',
+    workflowStage: workflowStageEntity?.normalized_value || 'pre_approval',
+    workflowStageLabel: workflowStageEntity?.value || '前置申请',
+    expenseTypeCode,
+    expenseTypeLabel: resolveExpenseTypeLabel(expenseTypeCode),
+    amount: amount.value,
+    amountDisplay: amount.value ? `¥${amount.value.toLocaleString('zh-CN')}` : '待补充',
+    timeRange: resolveTimeRangeText(ontology) || '待补充',
+    location: locationEntity?.normalized_value || locationEntity?.value || '待补充',
+    reason: String(prompt || '').trim() || '待补充',
+    applicant: currentUser.name || currentUser.username || '当前用户',
+    department: currentUser.department || currentUser.departmentName || '待补充',
+    preApprovalRequired: PRE_APPROVAL_TYPES.has(expenseTypeCode),
+    attachmentPolicy,
+    missingSlots: normalizeMissingSlots(ontology?.missing_slots || [])
+  }
+}
+
+export function normalizeMissingSlots(slots = []) {
+  const normalized = Array.isArray(slots) ? slots : []
+  return normalized.map((item) => ({
+    key: String(item || '').trim(),
+    label: SLOT_LABELS[String(item || '').trim()] || String(item || '').trim()
+  })).filter((item) => item.key)
+}
--- a/web/src/utils/hermesEmployeeSettingsModel.js
+++ b/web/src/utils/hermesEmployeeSettingsModel.js
@@ -1,69 +1,26 @@
 /** 数字员工设置：面向管理员的简明任务列表（频率固定，仅可调执行时间） */
 export const HERMES_SIMPLE_TASKS = [
  {
-    id: 'knowledgeAggregation',
-    label: '知识库同步',
-    hint: '同步制度文档与知识索引',
-    frequency: 'daily',
-    frequencyLabel: '每天'
-  },
-  {
-    id: 'ruleReviewDigest',
-    label: '规则待审提醒',
-    hint: '汇总待审规则并推送管理员',
-    frequency: 'daily',
-    frequencyLabel: '每天'
-  },
-  {
-    id: 'riskSummary',
+    id: 'global_risk_scan',
    label: '风险每日巡检',
    hint: '扫描报销、付款等风险信号',
    frequency: 'daily',
    frequencyLabel: '每天'
  },
  {
-    id: 'archiveDigest',
-    label: '归档周报',
-    hint: '汇总已归档报销单',
+    id: 'weekly_expense_report',
+    label: '费控洞察周报',
+    hint: '聚合生成财务总结简报',
    frequency: 'weekly',
    frequencyLabel: '每周一',
    weekday: 1
-  },
-  {
-    id: 'dailyStats',
-    label: '日报统计',
-    hint: '生成昨日报销与审批数据',
-    frequency: 'daily',
-    frequencyLabel: '每天'
-  },
-  {
-    id: 'monthlyStats',
-    label: '月报统计',
-    hint: '每月 1 号生成上月汇总',
-    frequency: 'monthly',
-    frequencyLabel: '每月 1 日',
-    monthDay: 1
-  },
-  {
-    id: 'yearlyStats',
-    label: '年报统计',
-    hint: '每年 1 月 1 号生成上年汇总',
-    frequency: 'yearly',
-    frequencyLabel: '每年 1 月 1 日',
-    month: 1,
-    monthDay: 1
  }
 ]

 function buildDefaultSchedules() {
  const defaults = {
-    knowledgeAggregation: { enabled: true, frequency: 'daily', time: '00:00', weekday: 1, monthDay: 1, month: 1 },
-    ruleReviewDigest: { enabled: true, frequency: 'daily', time: '18:00', weekday: 5, monthDay: 1, month: 1 },
-    riskSummary: { enabled: true, frequency: 'daily', time: '09:00', weekday: 1, monthDay: 1, month: 1 },
-    archiveDigest: { enabled: false, frequency: 'weekly', time: '10:30', weekday: 1, monthDay: 1, month: 1 },
-    dailyStats: { enabled: true, frequency: 'daily', time: '08:30', weekday: 1, monthDay: 1, month: 1 },
-    monthlyStats: { enabled: true, frequency: 'monthly', time: '09:00', weekday: 1, monthDay: 1, month: 1 },
-    yearlyStats: { enabled: false, frequency: 'yearly', time: '10:00', weekday: 1, monthDay: 1, month: 1 }
+    global_risk_scan: { enabled: true, frequency: 'daily', time: '09:00', weekday: 1, monthDay: 1, month: 1 },
+    weekly_expense_report: { enabled: false, frequency: 'weekly', time: '10:30', weekday: 1, monthDay: 1, month: 1 }
  }

  for (const task of HERMES_SIMPLE_TASKS) {
@@ -91,13 +48,8 @@ export function buildDefaultHermesEmployeeForm() {
    masterEnabled: true,
    notifyOnFailure: true,
    capabilities: {
-      knowledgeAggregation: true,
-      ruleReviewDigest: true,
-      riskSummary: true,
-      archiveDigest: false,
-      dailyStats: true,
-      monthlyStats: true,
-      yearlyStats: false
+      global_risk_scan: true,
+      weekly_expense_report: false
    },
    schedules: buildDefaultSchedules()
  }
--- a/Show More
+++ b/Show More