Files
agent_jrxml/frontend/README.md
T
panda bd5bfbac2d fix: band-level windowed refine_layout + programmatic map_fields to prevent 91.5% content loss
Root cause: LLM receiving full 34k-char JRXML would regenerate from scratch
instead of modifying coordinates in-place, shrinking output to ~3k chars.

Solution (programmatic node control, not prompt engineering):

- New agent/jrxml_windower.py: decompose JRXML into header (never sent to
  LLM) + individual bands. Split bands >4000 chars at element boundaries.
  Reassemble with element count validation (>10% change = rollback).

- Rewrite refine_layout: per-band windowed LLM processing (~2-4k chars
  each). LLM cannot "reimagine" the entire report.

- Rewrite map_fields: 100% programmatic regex $F{field_N} -> real name
  replacement. Zero LLM calls, zero content loss.

- _sanitize_field_name: non-ASCII chars escaped to _uXXXX_ format for
  valid JRXML identifiers.

- Tests: 48 new unit tests (windower 28 + map_fields 20). All passing.
  Full suite 385 tests, zero regressions.
2026-05-24 08:55:38 +08:00

1.7 KiB
Raw Blame History

JRXML Agent 前端

Vue 3 + TypeScript + Vite + Pinia — JRXML 报表生成代理的 Web UI。

技术栈

  • Vue 3 (Composition API + <script setup>)
  • TypeScript 6.x
  • Vite 8.x
  • Pinia 3.x (状态管理)
  • SSE (Server-Sent Events) 流式响应

组件结构

src/
├── api/client.ts            SSE 客户端 + fetch 封装
├── stores/
│   ├── chat.ts              Pinia: 消息/流式/节点进度/文件
│   ├── session.ts           Pinia: 会话 CRUD
│   └── kb.ts                Pinia: 多租户知识库管理
├── components/
│   ├── Sidebar.vue          会话列表 + 下载 + 历史版本
│   ├── ChatMessages.vue     消息列表渲染
│   ├── ProcessSection.vue   处理过程折叠区
│   ├── UnifiedInput.vue     统一输入框(文本+文件拖拽/粘贴)
│   ├── SummaryCard.vue      结果摘要卡片
│   ├── KbSelector.vue       KB 下拉选择器
│   └── KbManager.vue        KB 管理面板(创建/上传/构建/删除)
└── utils/format.ts          工具函数

开发

npm install
npm run dev        # 启动开发服务器 (localhost:5173)
npm run build      # 生产构建
npx playwright test # E2E 测试

SSE 事件流

前端通过 api.chat() 发起 POST 请求,后端返回 text/event-stream

事件 说明
node_start 节点开始执行(含 node/label/step_index
node_complete 节点执行完成(含 detail
stream_token LLM 逐字输出
agent_complete 全图执行完成(含 intent/status/jrxml_length/error 等)
agent_error 执行异常