6 张为输出物示意(按真实产物版式重排),不是成品截图。
Requires
What you need before running this pack. Miss one and it cannot do the work.
- Python python3 runs on your machine — one command, and it does not touch your files
- Report / exported CSV You export a CSV or JSON from the platform admin as the input
Picked most often by
Who buys it most, by role. Click a role to see only that role's skills.
Similar skills
View all →竞品拆解包
下架急救包(变狗 / 搜不到 / 受限商品)
欧盟合规包:上架前的四关自查(GPSR / PPWR / EPR / 欧代)
侵权自查与预警包
What is included
- ✓ sample-scope — sample-scope:先量样本,再谈结论
- ✓ pain-cluster — pain-cluster:差评到底在骂什么
- ✓ motive-extract — motive-extract:买家到底为什么买
- ✓ action-plan — action-plan:从「知道了」到「改哪一行」
Who it is for
- ✓ People who need steady output in a fixed scene (Sourcing / Operations)
- ✓ You already have assets — you need structure
- ✓ Want to set it up once and reuse it, instead of re-explaining every time
Not for
- – Want full automation with no input at all
- – One-off use, no reuse planned
- – Platform-specific ad console actions (do those manually)
Use “评论挖掘包” to make me Web pages — I will paste the input below.
Copy this line to your agent to startAbout this skill
选品数据告诉你卖得怎么样,评论才告诉你为什么买、为什么退。把竞品 ASIN / 商品链接 / 导出的评论 CSV 变成四张可复算的表:差评痛点聚类(带逐字原文引用)、好评购买动机、未被满足的需求、逐行卖点改写与差评预防清单。四关离线、纯标准库、不联网,样本不足会直接说「样本不足」。
Sample output
Real input → real output, unpolished# 一页汇总:评论挖到了什么、明天改什么 - 样本:706 条评论(差评 ≤3 星 506 条 / 好评 ≥4 星 200 条) - 平均星级:2.72 - 改写 profile:beauty-tool(类目覆盖只换措辞,不换阈值与证据) - 达到「未满足需求」阈值的痛点:2 类(阈值:频次≥3、占差评≥10%、严重度≥4) - 触发转折句(好评里也提)的:2 类 ## 先改这三行 listing 1. 【五点第 2 条】硬度与手感不符(证据:az-B00MZT4UEG-0010 az-B00MZT4UEG-0012 az-B00MZT4UEG-0092) - 现在:STIMULATES SCALP: Brushing your hair with boar bristles massages the scalp and stimulates blood flow - 建议(英文草稿):Bristle guide: this model is medium-firm for normal-to-thick hair. The close-up shows bristle length, density and tip finish; a softer model is available for fine hair. 2. 【图片/尺码表】气味、手感、外观(证据:az-B00MZT4UEG-0212 az-B00MZT4UEG-0225) - 现在:(现用 listing 里这一行是空的) - 建议(英文草稿):Close-ups of texture, finish and colour under natural light — no filter. 3. 【A+ 图文】效果看得见(证据:az-B00MZT4UEG-0209 az-B00MZT4UEG-0211) - 现在:(现用 listing 里这一行是空的) - 建议(英文草稿):Results timeline: what you should expect wee
rank,motive_id,label_zh,label_en,freq,share_of_positives,share_of_all,slot,quote_ids,sample_quote 1,sensory-detail,气味、手感、外观,Scent / texture / look,40,0.2000,0.0567,五点第 2 条 + 细节特写图,az-B00MZT4UEG-0212 az-B00MZT4UEG-0225,"It’s not too soft or too hard, it’s on point." 2,effect-visible,效果看得见,Visible results,29,0.1450,0.0411,五点第 1 条 + 前后对比/实测图,az-B00MZT4UEG-0209 az-B00MZT4UEG-0211,It is tiny but it works well for certain purposes. 3,price-anchor,价格锚点与性价比,Price anchor / value,17,0.0850,0.0241,五点第 5 条 + 组合装,az-B00MZT4UEG-0226 az-B00MZT4UEG-0260,"I received today this small hair brush, good for purse, very nice, good manufacture and good price, i received one day before that i expected." 4,scenario-use,具体使用场景,Use scenario,16,0.0800,0.0227,五点第 1 条 + 场景主图,az-B00MZT4UEG-0214 az-B00MZT4UEG-0228,"It will be fine for travel and those occasions when you want a brush, but are carrying a compact purse."
Install guide
One column per agent, each with a check that proves it installed. If the check fails, it is not installed.Claude Code
.claude/skills/review-mining/
mkdir -p .claude/skills
unzip review-mining.zip -d .claude/skills/
How to verify Open a new session and ask it to list the installed skills — seeing this pack's name means it worked.
For a global install, swap .claude/skills/ for ~/.claude/skills/ so every project of this user sees it.
Codex
~/.codex/skills/review-mining/
mkdir -p ~/.codex/skills
unzip review-mining.zip -d ~/.codex/skills/
How to verify Open a new codex session and ask it to list available skills — this pack showing up means it is live.
Codex only reads ~/.codex/skills/, not the project folder; on another machine you place it again.
Cursor
.cursor/rules/review-mining.mdc
mkdir -p .cursor/rules
cp review-mining/SKILL.md .cursor/rules/review-mining.mdc
How to verify In chat, @ that rule file and ask what the pack's first hard rule is — a correct answer means it loaded.
Cursor has no separate skills folder: rules belong to the project, so you place them again per project — the biggest difference from the other two.
If it will not installNot opening a terminal at all? Paste the whole SKILL.md into the chat. Same result, you just re-paste it each new session.
Full steps and common errors live in the pack's playbook.md; the guides section also has ordering and usage guides。
How to use
4 steps; the first run takes about 10 minutes-
1
准备 **环境**:Python 3.8+,只用标准库。不需要 pip、不需要 API key、不联网。 **数据**:四个关卡吃的都是同一张评论表,9 列: | 列 | 必填 | 说明 | |---|---|---| | `review_id` | 是 | 唯一编号,后三关引用它;没有就自己编号(如 `az-ASIN-0001`) | | `site` | 是 | `amazon` / `shopify` / `etsy` / `自建站` 等,原样保留 | | `asin` | 是 | Amazon 用 ASIN,独立站用 handle / product id | | `product` | 是 | 商品名(人类可读,用来按商品合并样本) | | `rating` | 是 | 1–5;`4.0` 也接受,`0`
-
2
装配 没有 shell 的用法:把 `skills/<name>/SKILL.md` 全文贴进对话,按里面的规则手工执行,产出自己写进 CSV——数字仍然可复算,只是慢。
-
3
触发顺序(为什么是这个顺序) 四关是**串行**的,前一关的产出是后一关的输入,顺序不能换: 1. **sample-scope**(先量样本)→ 决定后面所有结论的措辞强度。 因果:样本不足时,任何「多数买家」都是错的。所以第一关先算条数、ASIN、日期、评分分布,给出 `usable / thin / insufficient`。 反例(真实踩过):先聚类再回头看样本,写出「50% 的差评都骂缝线」——结果那 50% 只有 4 条评论,分母是 8。顺序反了就会这样。 2. **pain-cluster**(差评聚类)→ 决定 listing 改哪一行、预防做什么。 因果:频次 + 严重度是「先改哪里」的唯一依据;引用是给人复核用的,不是用来论证情绪。 反例:只挑最刺眼的一条差评改文案,结果那一条是竞争对手刷的(本包不判断真假,但 `sa
-
4
每关的输入输出与验收 | 关卡 | 输入 | 输出 | 验收标准(怎么算过) | 失败信号(怎么算没过) | |---|---|---|---|---| | sample-scope | `before/reviews.csv` | `sample-scope.md` / `.json` | 能在 md 里看到「结论 + 触发原因 + 商品分布 + 不能得出的结论」;`usable_rows` 与 CSV 行数对得上 | 缺列直接退出码 2;日期全坏 → `可用行数为 0` | | pain-cluster | `reviews.csv` + `pain-rules.json` | `pain-clusters.csv`、`quotes.csv`、`pain-clusters.md`、`pain-unmatched.csv` | 每
Reviews
Only verified buyers count; we do not edit, delete or pad them.This pack just launched and nobody has reviewed it yet. Once real buyers do, the stars, text and 1–5 distribution appear here; until then the rating shows an em dash instead of a made-up score.
Changelog
Updates are free while subscribed; a buyout includes later versions of this skillHow to buy
Subscribe or buy outright — a buyout means no more subscription feesMonthly
Try it for a month first
- ✓ “评论挖掘包” — all files, samples and install instructions
- ✓ Every skill unlocked while subscribed (including new releases)
- ✓ Commercial licence: client work allowed, unlimited deliveries
- ✓ Version updates pushed automatically, no repurchase
- ✓ 1:1 support on Telegram, replies within 12 hours on business days
- ✓ Cancel anytime, no next-period charge
Yearly
Pay for a year; prices are never converted to a monthly figure.
- ✓ “评论挖掘包” — all files, samples and install instructions
- ✓ Every skill unlocked while subscribed (including new releases)
- ✓ Commercial licence: client work allowed, unlimited deliveries
- ✓ Version updates pushed automatically, no repurchase
- ✓ 1:1 support on Telegram, replies within 12 hours on business days
- ✓ Cancel anytime, no next-period charge
Lifetime
The founder price rises as the skill library grows.
- ✓ “评论挖掘包” — all files, samples and install instructions
- ✓ Every skill unlocked while subscribed (including new releases)
- ✓ Commercial licence: client work allowed, unlimited deliveries
- ✓ Version updates pushed automatically, no repurchase
- ✓ 1:1 support on Telegram, replies within 12 hours on business days
- ✓ No subscription or renewals — one payment, forever
No price conversion: yearly plans only show the saving, never a per-month figure. A buyout is one payment with no renewals.