fix: 增强导出下载诊断能力，缩短超时时间

下载超时从 300s 减至 120s，失败时自动保存截图、打印服务端响应内容、
检查页面错误提示和新标签页，便于定位 download 事件未触发的根因。

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

CLAUDE.md

@@ -4,7 +4,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 ## Project Overview
 
-SalesShow is a monolithic Flask web application for analyzing sales data from Excel files. It supports manual Excel uploads and automated daily downloads from secsion.com via Playwright browser automation. There is no database — all data lives as Excel files on disk in `uploads/`.
+SaleShow is a monolithic Flask web application for analyzing sales data from Excel files. It supports manual Excel uploads and automated daily downloads from secsion.com via Playwright browser automation. There is no database — all data lives as Excel files on disk in `uploads/`.
 
 ## Commands
 
@@ -23,6 +23,7 @@ gunicorn -w 4 -b 0.0.0.0:8000 app:app
 
 # CLI automation — download reports from secsion.com
 python -m automation.secsion --start 2026-04-28 --end 2026-04-28
+python -m automation.secsion --start 2026-05-15 --end 2026-05-17 --username 15682076681 --password yourpassword
 ```
 
 No test framework or linter is configured in this project.
@@ -31,17 +32,19 @@ No test framework or linter is configured in this project.
 
 **Backend (Flask, single `app.py`):**
 - Routes handle file upload (`/upload`), file listing (`/files`), data loading/processing (`/load/<filename>`), deletion, and cleanup.
-- `process_sales_data()` (~lines 371-575 in `app.py`) is the core logic. It uses a state-machine approach to handle two Excel formats: "flat tables" (each row has code + product) and "hierarchical tables" (code row is a header, product rows are children). Outputs daily summaries with per-product breakdowns.
+- `process_sales_data()` (~lines 371-575) is the core logic. It uses a state-machine approach to handle two Excel formats: "flat tables" (each row has code + product) and "hierarchical tables" (code row is a header, product rows are children). Outputs daily summaries with per-product breakdowns.
 - `find_header_row()` dynamically detects the header row by scanning first 20 rows for keyword matches.
+- Auto-download routes use a global `download_status` dict and run Playwright in daemon threads via `threading.Thread`.
 
 **Automation module (`automation/`):**
-- `secsion.py` — `SecsionDownloader` uses Playwright headless Chromium to log into secsion.com, navigate to reports, set date range via TDesign date picker, optionally inject `shop_id` via route interception, and download exports.
-- `uploader.py` — copies downloaded files into `uploads/` with timestamp-prefix naming (same convention as manual uploads).
-- `scheduler.py` — APScheduler `BackgroundScheduler` with `CronTrigger` runs daily auto-download (default 01:00).
+- `secsion.py` — `SecsionDownloader` uses Playwright headless Chromium to log into secsion.com, navigate to reports, set date range via TDesign date picker (requires click → select day → Enter confirm → Escape close sequence), optionally inject `shop_id` via route interception on `**/api/bill/export`, and download exports. Has 3-retry logic with exponential backoff.
+- `uploader.py` — copies downloaded files into `uploads/` with `YYYYMMDD_HHMMSS_` prefix naming (same convention as manual uploads).
+- `scheduler.py` — APScheduler `BackgroundScheduler` with `CronTrigger` runs daily auto-download (default 01:00). Uses `misfire_grace_time=3600`.
 
 **Configuration (`config.py`):**
 - Three-tier priority: Web UI settings (`data/config.json`) > environment variables (`.env` / system env) > defaults.
 - `Config` class provides static methods for reading/writing secsion credentials, shop ID, and scheduler settings.
+- Passwords are masked (`******`) when returned via the API.
 
 **Frontend (vanilla JS/CSS, no build step):**
 - `main.js` — all client-side interactivity: file upload (drag-and-drop), AJAX to API, data rendering (card/table view), client-side filtering, sorting, pagination (50 items/page), export.
@@ -52,5 +55,5 @@ No test framework or linter is configured in this project.
 
 - No database — Excel files on disk are the data store.
 - No frontend build step — vanilla JS/CSS served directly via Flask static files.
-- Playwright automation runs in daemon threads with a global `download_status` dict for status tracking.
-- Passwords are masked (`******`) when returned via the API.
+- Playwright automation runs in daemon threads; status tracked via module-level `download_status` dict in `app.py`.
+- The secsion.com date picker uses TDesign's `needconfirm="true"` mode — simply calling `.fill()` won't work; must click cell then press Enter.