首页 / Skills / 电商运营 / Shopify 评价分诊 Shopify Review Triage
电商运营

Shopify 评价分诊 Shopify Review Triage

Shopify App Review Brief (independent)MITSkill 语言: 英文
协议已核对不含脚本低风险
演示真实运行记录 · claude-sonnet-5-5 · 2026年10月
Using the shopify-review-triage skill, triage these fictional public reviews. 1 | Example Upsell App | The cart drawer shows a blank screen and checkout stopped working after the update. | 2 | Example Upsell App | Charged twice on the billing page and the free plan limit was not explained. | 3 | Example Upsell App | Setup was confusing and I had to contact support to find the settings. | 4 | Example Upsell App | Would love an option to hide the widget on certain collections. | 5 | Example Upsell App | Nice app.
Skill: shopify-review-triage

Low-star review brief - Example Upsell App - week of 2026-10-05

Triaged 5 rows supplied: 1 incident risk, 1 repeated friction, 1 pricing confusion, 1 feature request, 1 needs human read. This is a first pass and has not been human-checked.

Scope: Example Upsell App, no competitors watched, 5 rows supplied, no review dates supplied. The brief covers only the rows supplied and makes no claim of exhaustive coverage. Reviews are customer reports, not verified defects. Items marked "first pass" are unverified keyword matches.

I read the leading numbers (1 to 5) as star ratings, as the skill specifies for the three-field form. If they were row IDs, the ratings are unknown, so please tell me.

P0 - Incident risk

  • Example Upsell App - blank cart drawer, checkout stopped working (1 star, date not captured, source: not captured)
    • Reviewer reports: "The cart drawer shows a blank screen and checkout stopped working after the update."
    • Matched keywords: blank screen, stopped working, checkout. The evidence is more than the word "checkout" alone.
    • Status: first pass - not human-checked
    • Reproduced: not yet attempted
    • Next action: try to reproduce on a development store today, on the version that shipped just before the review. Check the error tracker and support inbox for matching signals. If confirmed, treat it as an incident. Owner: unassigned, due: not set.

P1 - Repeated friction

  • Example Upsell App - setup and finding settings (3 stars, date not captured, source: not captured)
    • Reviewer reports: setup was confusing and they had to contact support to find the settings.
    • Matched keywords: confusing, had to contact support. This is one row, so there is no repetition and no escalation.
    • Status: first pass - not human-checked
    • Next action: log it against the matching support theme. Check whether other setup-related tickets exist. Owner: unassigned, due: not set.

P2 - Pricing confusion

  • Example Upsell App - double charge and unexplained free plan limit (2 stars, date not captured, source: not captured)
    • Expected vs. actual: the reviewer reports being charged twice on the billing page and says the free plan limit was not explained.
    • Matched keywords: charged, billing, free plan.
    • Status: first pass - not human-checked
    • Next action: compare the listing's pricing section and in-app upgrade prompts with what the reviewer describes. Clarify the free plan limit copy where they diverge. A double charge could be a billing defect rather than a copy issue, so have a person check the billing records. Owner: unassigned, due: not set.

P3 - Feature requests

  • Example Upsell App - option to hide the widget on certain collections (4 stars, date not captured, source: not captured). Log it in the feature-request log. If collection-level visibility already exists, reply with where to find it.

Needs human read

  • Example Upsell App (5 stars, date not captured, source: not captured). No keyword matched ("Nice app."). There is no detail to act on, so a person should read it and file it manually.

Competitor watch

  • None supplied.

Decisions this week

  • Reproduce row 1 first, since it is the only candidate that may affect the purchase path. Rows 1 and 2 both involve money flow, so the double-charge report in row 2 is worth a billing-records check at the same time.

Limits and notes

  • Every item is a keyword match. Nothing has been reproduced or human-checked.
  • No source links, dates, or owners were supplied. Send the five-field form (rating | app | date | reviews URL | text) and I can fill in sources and dates.
  • Keyword matching is English-only and misses sarcasm and context.
  • Nothing was sent, posted, or published. This is a draft for your team.

Real run in an isolated folder with only this skill installed. Only the Skill and Read tools were enabled, so nothing was fetched from the web and no file was written. The example is fictional. The reviews are invented.

它做什么

把公开的应用商店评价文字分成五类:事故风险(P0)、反复出现的摩擦(P1)、定价困惑(P2)、功能请求(P3)和需要人工阅读。每条评价只归入一个主分类,其余命中的记为次要分类,最终得到一份简报,每个条目都有负责人、下一步动作和来源链接,没有来源的明确写“source: not captured”。

工作方式

  1. 你粘贴若干行数据:评分、应用名、日期、公开评价链接和评价正文。
  2. 模型按一套已公开的关键词规则分类,并遵守判定规则:取最严重的分类;约 60 天内有三条或更多评价反复出现同一问题则升一级;超过一年的评价只作背景;竞品的事故不会变成你的 P0。
  3. 所有结果都标注“first pass - not human-checked”(初筛,未经人工核对),并要求由人阅读原评价、对 P0 候选尝试复现后再提升级别。

适合什么场景

独立 Shopify 应用团队和代运营机构,需要从分散的低分评价里判断该先修什么。

说明与风险

基于你粘贴的数据做纯推理:没有脚本,不联网,不需要账号或密钥。它只接受公开的评价文字,拒绝工单、商家邮件等私有数据,绝不编造评价、评分或链接,也绝不联系任何人或发布回复,只交还草稿。关键词匹配只支持英文,识别不了讽刺,也可能归错类,所以每个条目都只是初筛,必须由人核对。与 Shopify 无关联,也未获其背书。