tkshop-daren-cleaninglisted

## 触发条件（description 有 200 字符上限，触发词放这里） 任何满足以下之一的请求都走这个 skill： - 上传或提到 `xxxx-xx-xx_tkshop达人信息.xls` / `tkshop达人信息` 这类文件 - 「清洗一下数据」「照上次那样洗」「把这几张表合并」「按达人昵称去重」 - 「初筛」「筛达人」「不符合的行直接删掉」「输出符合条件的」「M列删掉」 - 提到 K₫ / M₫ / Tr₫ / 万 这类越南、中文数量简写要转纯数字 - 提到要把「每个销售渠道的GMV」「粉丝性别占比」「粉丝年龄区间占比」这类 多行百分比字段拆成独立列 ## What this skill is for The user periodically uploads TikTok Shop (tkshop) 达人 (influencer) export files named like `2026-07-14_tkshop达人信息.xls`. These raw exports have messy Vietnamese/Chinese-locale formatting: currency shorthand (K₫/M₫/Tr₫), view-count shorthand (万), and several fields that pack multiple percentages into one multi-line text cell. This skill defines the exact cleaning rules that were worked out (and debugged) with the user across several real files, so future files get the same treatment without re-deriving the rules or re-making the same mistakes. A reference implementation is bundled below as `clean_tkshop.py` — write it out and run it rather than reimplementing the logic from scratch, since it already encodes fixes for real bugs that were found in this data (see the "real bug" notes inside the script). ## Step 0: figure out which mode the user wants There are two distinct things the user asks for, and they are NOT interchangeable — always confirm (or infer from clear phrasing) which one is meant: 1. **合并去重 (`merge_dedupe`)** — merge two or more raw exports (possibly spanning several dates) into one table, dropping duplicate 达人名称 and keeping only the most recent