diff --git a/skills/archived/miaoshou-collect-box/SKILL.md b/skills/archived/miaoshou-collect-box/SKILL.md
new file mode 100644
index 0000000..a160cae
--- /dev/null
+++ b/skills/archived/miaoshou-collect-box/SKILL.md
@@ -0,0 +1,369 @@
+---
+name: miaoshou-collect-box
+category: cross-border-ecommerce
+description: Explore and interact with Miaoshou ERP's "通用采集箱" (Common Collect Box) — URLs, collection methods, import paths, and login automation.
+---
+
+# 妙手ERP 通用采集箱 (Common Collect Box)
+
+## Overview
+Miaoshou ERP (`erp.91miaoshou.com`) has a **Common Collect Box** (`通用采集箱`) at a fixed URL. This skill documents how to navigate to it, the collection methods available, and how to interact with the import/collection UI programmatically via DrissionPage.
+
+## Login
+- URL: `https://erp.91miaoshou.com/auth/login`
+- Credentials: `xiaochaoren2026 / [REDACTED]`
+- CAPTCHA: **Base64 inline image** (`data:image/png;base64,...`) rendered in `.captcha-img`
+  - Unlike Dianxiaomi (which uses a separate PNG endpoint), Miaoshou embeds the captcha directly.
+  - Extract with JS: `document.querySelector('.captcha-img').src`
+  - Decode base64, save to file, run `ddddocr.classification()`
+  - Apply digit cleaning: `raw.translate(str.maketrans('oliOSsZz', '01105522'))`
+- Form selectors:
+  - Account: `css:.account-input`
+  - Password: `css:.password-input`
+  - Captcha text: `css:.captcha-text`
+  - Captcha UUID: `input[name="captchaUuid"]` (must be set/synced)
+  - Submit: `css:.login-button`
+- Success redirects to: `https://erp.91miaoshou.com/welcome`
+- Cookie file: `~/.hermes/cookies/miaoshou_cookies.json` (autoLoginToken is critical)
+
+## Common Collect Box URL
+Direct link (works after login):
+```
+https://erp.91miaoshou.com/common_collect_box/index
+```
+Title: `妙手-产品采集`
+
+## Collection Methods Discovered
+The page exposes **6 collection methods** as tab-like options:
+
+| Method | Chinese Name | Notes |
+|--------|-------------|-------|
+| Link Collection | 链接采集 | Enter public product URLs (1688, Taobao, etc.). System scrapes the page. **Requires publicly accessible pages** — internal systems (like AtomK behind login) will NOT work. |
+| Import Collection | 导入采集 | **Bulk import via file upload** (template-based). This is the viable path for transferring data from AtomK or other internal systems. |
+| 1688 AI Selection | 1688Ai选品 | AI-assisted product selection from 1688 |
+| Inventory Collection | 货盘采集 | Collect from inventory/supplier pallets |
+| Keyword Collection | 关键词采集 | Search keyword-based batch collection |
+| Whole-Store Collection | 整店采集 | Enter a store URL to collect all products |
+| Plugin Collection | 插件采集 | Browser extension-based collection |
+| Store Cross-Collection | 店铺互采 | Cross-store collection |
+
+## Import Collection (导入采集) — 4 Sub-Methods
+When you click "导入采集", the page switches to `?fetchType=importCopy` and shows **4 sub-tabs**:
+
+| Sub-Method | Chinese Name | What It Does |
+|-----------|-------------|-------------|
+| Import Link Collection | 导入链接采集 | Upload an **XLS file** with product URLs. Miaoshou scrapes each URL. |
+| Excel Spreadsheet Import | Excel表格导入 | Upload a pre-structured Excel with full product data (details TBD). |
+| Local Material Package | 本地素材包导入 | Upload a local asset package (images + data). |
+| Jushuitan Material Package | 聚水潭素材包导入 | Import from Jushuitan ERP format. |
+
+### Import Link Collection Template (导入链接采集)
+This is the **most important** one for AtomK-style workflows.
+
+**Template file**: `导入产品链接模板.xls` (downloadable via "下载导入模板" link in the modal)
+**File format**: `.xls` (older Excel 97-2003, NOT xlsx) — verified April 2026. `.xlsx` triggers `解析文件数据错误`.
+
+**CRITICAL**: The template is **strictly 5 columns**. Any extra column causes failure.
+
+| Column | Field | Required | Notes |
+|--------|-------|----------|-------|
+| A | 链接地址（必填） | ✅ Yes | Product detail page URL. **Must be publicly accessible**. Login-gated pages (like AtomK behind login) will fail. |
+| B | 产品标题 | No | **AI titles ARE preserved** — verified. Overrides scraped 1688 title. |
+| C | 价格（RMB） | No | **AI prices ARE preserved** — verified. Overrides scraped 1688 price. |
+| D | 促销价（RMB） | No | Only used when publishing to **Lazada**. |
+| E | 提示：促销价仅可用于采集到Lazada | No | Literal hint text — include this column to match the official template exactly. |
+
+**Empirically verified findings (April 2026)**:
+- ✅ 5-column `.xls` with exact headers above → `解析成功 N/0/N`
+- ❌ 6+ columns (e.g., adding "产品描述") → `导入文件未识别到:【产品主编号】表头`
+- ❌ Adding "产品主编号" column → `解析文件数据错误`
+- ❌ `.xlsx` format → `解析文件数据错误`
+- ✅ AI title from XLS **overrides** scraped 1688 title in the final product
+- ✅ AI price from XLS **overrides** scraped 1688 price
+- ✅ Sheet name can be `Worksheet` or `Sheet1` (doesn't matter with correct 5-col headers)
+- ✅ Encoding can be `utf_16_le` or `utf-8` (doesn't matter with correct 5-col headers)
+
+**Important**: The error *"导入文件未识别到:【产品主编号】表头"* is a **misleading** catch-all for "your columns don't match the expected 5-column template". Do NOT add a "产品主编号" column — doing so triggers `解析文件数据错误`. Always use the exact 5-column format. The system does not actually use or store a "产品主编号" from Excel; SKU must be filled later via the edit dialog.
+
+**Critical limitation**: The "导入链接采集" method fundamentally relies on **web scraping** each product URL. If the URL requires authentication (e.g., AtomK product pages behind `atomlisting.com` login), Miaoshou cannot fetch the data. The 5-column XLS only passes the link + AI title + AI price; Miaoshou still scrapes the 1688 page for images, description, specs, etc.
+
+#### File Upload Mechanics
+When the import modal is open:
+- The upload area is a hidden `<input type="file" class="jx-upload__input">` inside a `.pro-upload` / `.jx-upload--text` container.
+- **Selector**: `css:input[type=file]` or `css:.jx-upload__input`
+- **Accepts**: `.xls,.xlsx,.csv` (but only `.xls` works reliably)
+- Use `file_input.input('/path/to/file.xls')` to set the file path programmatically.
+- After upload, the file name appears in `.jx-upload-list li`.
+- **The confirm button is inside the modal dialog** — search within `.jx-dialog` for buttons with text `确认`.
+- **After clicking confirm, the modal closes and the file appears in the "已导入的文件列表" table.** Wait ~15s for parsing, then **refresh the page** to check status.
+- **The page does NOT auto-update after confirm.** You must navigate back to `?fetchType=importCopy` or use `page.get()` to refresh and see the parse result.
+
+#### Complete Upload Flow (Verified April 2026)
+```python
+# 1. Navigate to import page
+page.get("https://erp.91miaoshou.com/common_collect_box/index?fetchType=importCopy")
+time.sleep(10)
+
+# 2. Click Excel import button
+page.run_js("""
+    var all = document.querySelectorAll('span, div, button, a');
+    for (var i=0; i<all.length; i++) {
+        var txt = all[i].innerText || '';
+        if (txt.includes('Excel表格导入') && txt.length < 20) {
+            all[i].click(); break;
+        }
+    }
+""")
+time.sleep(5)
+
+# 3. Upload file
+file_inputs = page.eles('css:input[type="file"]')
+file_inputs[0].input('/path/to/import.xls')
+time.sleep(5)
+
+# 4. Click confirm
+page.run_js("""
+    var btns = document.querySelectorAll('button');
+    for (var i=0; i<btns.length; i++) {
+        if (btns[i].innerText && btns[i].innerText.trim() === '确认') {
+            btns[i].click(); break;
+        }
+    }
+""")
+
+# 5. Wait for backend parsing, then refresh to check result
+time.sleep(20)
+page.get("https://erp.91miaoshou.com/common_collect_box/index?fetchType=importCopy")
+time.sleep(10)
+
+# 6. Check result in page body text
+body_text = page.run_js("return document.body.innerText;")
+# Look for: '解析成功' or '解析失败' near the filename
+```
+
+#### Known Error Patterns
+| Error Message | Meaning | Fix |
+|---------------|---------|-----|
+| `导入文件未识别到:【产品主编号】表头` | Column count ≠ 5 or headers don't match | Ensure **exactly 5 columns** with headers: `链接地址（必填）`, `产品标题`, `价格（RMB）`, `促销价（RMB）`, `提示：促销价仅可用于采集到Lazada`. Do NOT add "产品主编号" — it's a misleading error. |
+| `解析文件数据错误` | Wrong format or added "产品主编号" | Use `.xls` (not `.xlsx`). Remove any "产品主编号" column. Ensure exact 5-column headers. |
+| `解析失败` (with `0/N/N`) | URLs accepted but scraping failed | Linked pages are unsupported (e.g., COS-hosted custom HTML, login-gated pages). Use standard 1688/Taobao/Pinduoduo URLs. |
+
+## Collect Box Product Edit Dialog (采集箱产品编辑弹窗)
+After a product is imported and **claimed to a platform** (e.g., Temu全托管), it appears in the **已认领** tab. Each row has an **编辑** button that opens a modal dialog for editing the product's full data.
+
+This dialog is the key to filling AI-generated fields (SKU, description, keywords) that cannot be imported via the 5-column XLS template.
+
+### Dialog Structure
+- **Root dialog**: `.jx-dialog.pro-dialog.collect-box-editor-dialog-V2` (1880×953)
+- **Editor body**: `.collect-box-editor` (1559×953)
+- **Tabs**: 基本信息, 产品属性, 销售属性, 产品图片, 产品视频, 认证说明, 物流信息, 货源链接
+
+### Critical Field Mappings (Verified April 2026)
+The editor uses Vue.js with custom `jx-input` components. Direct `.value` assignment works but **must be followed by event dispatching** for Vue to sync state.
+
+| Field | DOM Element | Selector Strategy | Notes |
+|-------|------------|-------------------|-------|
+| 产品标题 | `input[type="text"]` (multiple) | Find first input whose `.value` contains the title text | Title is pre-filled from import; usually does NOT need updating. |
+| **产品主编号 (SKU)** | `input[type="text"]` at **index 1** | `editor.querySelectorAll('input')[1]` | Placeholder: `请输入产品主编号，不填将自动生成`. This is the **most reliable** field to target. |
+| **简易描述** | `textarea` at **index 0** | `editor.querySelectorAll('textarea')[0]` | Plain text. Fills to Shopee/Wish/Joom/Mercado Libre/Ozon/Walmart product description. |
+| **详细描述** | `iframe` at **index 0** (TinyMCE-like) | `editor.querySelectorAll('iframe')[0]` → `contentDocument.body` | Rich text editor. Used for Lazada/速卖通/TikTok/Coupang/eBay/Allegro. Body is `contenteditable`. |
+
+### Vue State Sync (Mandatory)
+Miaoshou uses Vue.js. Simply setting `.value` is **not enough** — the change must be propagated to Vue's reactive state. After setting any value, dispatch:
+```javascript
+el.value = 'your-value';
+el.dispatchEvent(new Event('input', {bubbles: true}));
+el.dispatchEvent(new Event('change', {bubbles: true}));
+el.dispatchEvent(new Event('blur', {bubbles: true}));
+el.dispatchEvent(new KeyboardEvent('keydown', {key: 'a', bubbles: true}));
+el.dispatchEvent(new KeyboardEvent('keyup', {key: 'a', bubbles: true}));
+```
+
+Without this, clicking **保存修改** will silently discard your changes.
+
+### Opening the Edit Dialog Programmatically
+Products are rendered in a **virtual scroll table** (`.pro-virtual-table__row`), not a standard `<table>`.
+
+```python
+# 1. Navigate to claimed tab
+page.get('https://erp.91miaoshou.com/common_collect_box/items')
+# ... click 已认领 tab ...
+
+# 2. Find the row by collect_box_id and click 编辑
+page.run_js("""
+    var rows = document.querySelectorAll('.pro-virtual-table__row, .pro-virtual-scroll__row');
+    for (var i=0; i<rows.length; i++) {
+        var txt = rows[i].innerText || '';
+        if (txt.includes('3516358992')) {  // your collect_box_id
+            var btns = rows[i].querySelectorAll('button');
+            for (var j=0; j<btns.length; j++) {
+                if (btns[j].innerText && btns[j].innerText.trim() === '编辑') {
+                    btns[j].click();
+                    break;
+                }
+            }
+        }
+    }
+""")
+time.sleep(10)  # Dialog takes time to render
+editor = page.ele('css:.collect-box-editor')
+```
+
+### Filling Each Field
+
+#### SKU (产品主编号)
+```python
+editor.run_js(f"""
+    var editor = document.querySelector('.collect-box-editor');
+    var skuInput = editor.querySelectorAll('input')[1];
+    skuInput.value = {json.dumps(sku)};
+    skuInput.dispatchEvent(new Event('input', {{bubbles: true}}));
+    skuInput.dispatchEvent(new Event('change', {{bubbles: true}}));
+    skuInput.dispatchEvent(new Event('blur', {{bubbles: true}}));
+    skuInput.dispatchEvent(new KeyboardEvent('keydown', {{key: 'a', bubbles: true}}));
+    skuInput.dispatchEvent(new KeyboardEvent('keyup', {{key: 'a', bubbles: true}}));
+""")
+```
+
+#### Simple Description (简易描述)
+```python
+editor.run_js(f"""
+    var editor = document.querySelector('.collect-box-editor');
+    var ta = editor.querySelectorAll('textarea')[0];
+    ta.value = {json.dumps(description)};
+    ta.dispatchEvent(new Event('input', {{bubbles: true}}));
+    ta.dispatchEvent(new Event('change', {{bubbles: true}}));
+    ta.dispatchEvent(new Event('blur', {{bubbles: true}}));
+""")
+```
+
+#### Rich Description (详细描述 — iframe)
+```python
+html_content = f'<p><span style="font-size: 12.0pt;">{description.replace(chr(10), "</span></p><p><span style=\\"font-size: 12.0pt;\\">")}</span></p>'
+
+editor.run_js(f"""
+    var editor = document.querySelector('.collect-box-editor');
+    var iframe = editor.querySelectorAll('iframe')[0];
+    var doc = iframe.contentDocument || iframe.contentWindow.document;
+    var body = doc.querySelector('body');
+    body.innerHTML = {json.dumps(html_content)};
+    body.dispatchEvent(new Event('input', {{bubbles: true}}));
+""")
+```
+
+### Saving
+```python
+page.run_js("""
+    var btns = document.querySelectorAll('button');
+    for (var i=0; i<btns.length; i++) {
+        if (btns[i].innerText && btns[i].innerText.includes('保存')) {
+            btns[i].click();
+            break;
+        }
+    }
+""")
+time.sleep(5)
+```
+
+### Verification
+After saving, re-open the edit dialog and read back the values to confirm persistence:
+```python
+verify = editor.run_js("""
+    var editor = document.querySelector('.collect-box-editor');
+    var inputs = editor.querySelectorAll('input');
+    var textareas = editor.querySelectorAll('textarea');
+    var iframes = editor.querySelectorAll('iframe');
+    var result = {};
+    if (inputs.length > 1) result.sku = inputs[1].value;
+    if (textareas.length > 0) result.desc = textareas[0].value;
+    for (var i=0; i<iframes.length; i++) {
+        try {
+            var doc = iframes[i].contentDocument || iframes[i].contentWindow.document;
+            var body = doc.querySelector('body');
+            if (body) { result.rich_desc = body.innerText; break; }
+        } catch(e) {}
+    }
+    return result;
+""")
+# verify.sku should match, verify.desc should contain AI text
+```
+
+### AtomK → Miaoshou Batch Import Script
+A production-ready end-to-end script exists at `/home/ubuntu/auto_miaoshou_import.py`:
+
+**What it does:**
+1. Logs into AtomK API, fetches claimed products + their AI-generated listings
+2. Generates a 5-column `.xls` file (`utf_16_le` encoding, `Worksheet` sheet name)
+3. Launches headless Chromium, loads Miaoshou cookies
+4. Navigates to `?fetchType=importCopy`, clicks `Excel表格导入`
+5. Uploads the `.xls` file, clicks `确认`
+6. Refreshes the page after 20s to verify `解析成功`
+7. Logs results and updates state file
+
+**State tracking:** `/home/ubuntu/.hermes/miaoshou_import_state.json`
+- Tracks already-imported `product_id`s to avoid duplicates
+- Updated after each successful run
+
+**Cron job:** `miaoshou-atomk-auto-import` (job_id: `2b6587d6895b`)
+- Schedule: `0 * * * *` (every hour on the hour)
+- Batch size: 6 products per run
+- Logs to: `/home/ubuntu/.hermes/miaoshou_import.log`
+
+**Usage:**
+```bash
+# Manual run
+python3 /home/ubuntu/auto_miaoshou_import.py
+
+# Check cron status
+hermes cron list
+```
+| Scenario | Viable? | Recommended Path |
+|---------|---------|-----------------|
+| AtomK product pages are **public** | ❌ No | Even **public** AtomK HTML pages (hosted on Tencent COS) fail with `解析失败 0/3`. Miaoshou's scraper is hardcoded for domestic e-commerce platforms (1688, Taobao, Pinduoduo), not custom HTML pages. |
+| AtomK product pages require **login** | ❌ No | Same as above — plus authentication barrier. |
+| **5-column XLS** import (link + AI title + AI price) | ✅ Yes | `解析成功`. Titles and prices from Excel **override** scraped 1688 data. Verified: AI title `Romantic Heart...` appears instead of 1688 original. |
+| 5-column XLS + extra columns (desc, SKU, images, EAN, etc.) | ❌ No | Triggers `导入文件未识别到:【产品主编号】表头`. Adding "产品主编号" triggers `解析文件数据错误`. |
+| ".xlsx" format | ❌ No | Triggers `解析文件数据错误`. Must use `.xls` (xlwt). |
+| **Full AI data via dual-browser automation** | ✅ **Yes** | **Proven workflow (April 2026)**:<br>1. Import via 5-col XLS (link + AI title + AI price)<br>2. Claim product to target platform<br>3. Open collect box edit dialog<br>4. Auto-fill SKU, 简易描述, 详细描述 (rich text) via DrissionPage + JS<br>5. Save and verify<br>Reference script: `/home/ubuntu/miaoshou_atomk_bridge.py` |
+| Direct platform upload (bypass Miaoshou) | ⚠️ Partial | Generate Temu/AliExpress CSV from AtomK API → upload to platform directly. Bypasses Miaoshou entirely but loses Miaoshou's inventory/order management. |
+
+## Page Structure Notes
+- Left sidebar contains: `通用功能`, `产品采集`, `公用采集箱`, `AI工作台`, `侵权检测`, `货盘中心`, etc.
+- Supported platforms displayed as tags: 1688, 义乌小商品城, 浙宝网, Vvic, 搜款网, 网商园, 货捕头, etc.
+- Bottom action bar: `采集并自动认领`, `采集并自动发布`
+- A modal/popup may appear on first visit with collection settings — dismiss with "我知道了".
+
+## Browser Setup (DrissionPage)
+```python
+from DrissionPage import ChromiumPage, ChromiumOptions
+co = ChromiumOptions()
+co.set_browser_path('/home/ubuntu/.cache/ms-playwright/chromium-1217/chrome-linux64/chrome')
+co.set_argument('--headless=new')
+co.set_argument('--no-sandbox')
+co.set_argument('--disable-dev-shm-usage')
+co.set_argument('--window-size=1440,900')
+co.set_user_agent('Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/121.0.0.0 Safari/537.36')
+co.set_local_port(get_free_port())
+page = ChromiumPage(addr_or_opts=co)
+```
+
+## Pitfalls
+- **Cookie expiration**: The saved cookie (autoLoginToken) expires relatively quickly. Re-login with captcha is often needed.
+- **Base64 captcha session**: Must extract the captcha image from the **same browser session** (via JS `document.querySelector('.captcha-img').src`) — do NOT use `requests` to fetch it separately, or the captcha UUID will mismatch.
+- **Tab clicking**: The collection method tabs (链接采集, 导入采集, etc.) are `<span>` elements inside radio-button groups, not standard `<button>` or `<a>` tags. Standard `.ele('xpath://*[contains(text(), "导入采集")]').click()` may fail. Use JS click via `page.run_js()` or search for the actual `<button>` inside the tab component.
+- **Import dialog**: Clicking "导入链接采集" opens a modal dialog. The template download link "下载导入模板" triggers a browser download to the default Downloads folder. In headless mode, the file goes to `/home/ubuntu/Downloads/导入产品链接模板.xls`.
+- **XLS format only**: The import link template is `.xls` (Excel 97-2003). Using `.xlsx` will likely fail.
+- **"暂无当前模块权限"**: Some accounts may see "You don't have permission for this module" if the collect box feature is not enabled for the sub-account. Contact the main account admin to enable.
+- **SPA navigation**: Clicking menu items may not trigger full navigation. Use `page.get('https://erp.91miaoshou.com/common_collect_box/index?fetchType=importCopy')` for direct access.
+- **Link scraper limitation**: The 导入链接采集 method requires **publicly accessible** product detail pages. Any page behind authentication (AtomK, internal ERPs) will result in empty/failed collection.
+## Next Exploration Steps (when needed)
+1. ✅ ~~Click the "导入采集" tab and inspect the file upload input~~ (Done)
+2. ✅ ~~Download the template to understand required columns~~ (Done — 5-column `.xls`: link, title, price, promo_price, hint)
+3. ✅ ~~Analyze "Excel表格导入" template for full-data import~~ (Done — the XLS template is corrupt/unreadable; full-field bulk import is **not viable**. The working path is 5-col import + edit-dialog automation.)
+5. ✅ ~~Automate the file upload and submission~~ (Done — upload via `file_input.input()` + confirm button in modal + page refresh to check result.)
+6. ✅ ~~Automate the edit dialog post-import fill~~ (Done — SKU, 简易描述, 详细描述 all verified saving correctly.)
+7. ✅ ~~Build end-to-end automation~~ (Done — script at `/home/ubuntu/auto_miaoshou_import.py` fetches AtomK API data, generates 5-col XLS, uploads via browser, and tracks state. Cron job `miaoshou-atomk-auto-import` runs hourly.)
+8. ✅ ~~Set up cron job for hourly batch import~~ (Done — job `miaoshou-atomk-auto-import` runs at `0 * * * *`, imports 6 products per run, skips already-imported IDs via `/home/ubuntu/.hermes/miaoshou_import_state.json`.)
+9. 🔴 **Future**: Extend edit-dialog automation to fill **产品图片** (replace 1688 images with AI-generated ones from AtomK `platform_fields.pictures`), **产品属性** (material, shape, brand), and **关键词** fields.
+10. 🔴 **Future**: Build batch loop that processes multiple `collect_box_id`s end-to-end (import → claim → edit → save).
+11. 🔴 **Future**: Investigate if `快速上货` (Quick Listing) offers a more direct API or form for bulk AI data injection, bypassing the collect box entirely.