customer-pulse

作者: anthropic

彙整 PayPal 爭議、HubSpot 回饋與工單、電子郵件情感分析(以及貼上或匯出的 Google/Yelp 評論),生成一份包含逐字引述的主題報告…

npx skills add https://github.com/anthropics/knowledge-work-plugins --skill customer-pulse

Customer Pulse

Quick start

Ask: "How are customers feeling this month?"

Claude pulls disputes, tickets, email threads, and Intercom conversations for the last 30 days, groups them into 3–5 themes with verbatim evidence, and delivers a "do these 3 things this week" action list.

To include Google/Yelp reviews, paste them after triggering — or say "I have some reviews to add."

Workflow

  1. Set the date window. Default: last 30 days. If the user specifies a range, use it.

  2. Pull PayPal disputes. Fetch disputes opened in the window. If the PayPal API returns a rate-limit error, skip and add PayPal: rate-limited — not included to the Sources section. Do not retry; do not error. See reference/gotchas.md for the rate-limit pattern.

  3. Pull HubSpot tickets and feedback. Fetch open and recently closed tickets. If 0 tickets exist, record HubSpot tickets: 0 and continue — do not surface a warning.

  4. Pull Gmail threads. Search for threads in the window containing: refund cancel unhappy issue problem disappointed frustrated broken late slow wrong missing. Extract subject lines and 1–2 sentence excerpts per thread.

  5. Pull Intercom conversations. Call search_conversations to fetch open and recently closed conversations. Then call get_conversation for each conversation ID returned to access the full conversation_parts. Extract parts where author.type === 'user' — these are customer messages. Exclude parts where author.type is admin or bot.

  6. Accept pasted reviews (optional). If the user pastes Google or Yelp review text, include it in the source pool tagged as [Review]. No connector required.

  7. Extract themes. Group all evidence into 3–5 recurring themes. Each theme must include:

    • A one-sentence label (e.g., "Shipping delays causing repeat complaints")
    • 2–3 verbatim quotes with source tags: [PayPal], [HubSpot], [Gmail], [Intercom], or [Review]
    • A signal count (how many items touch this theme)

    Verbatim quotes are non-negotiable — never paraphrase. See reference/gotchas.md for the verbatim anti-pattern.

  8. Generate the "do these 3 things" list. Rank themes by signal count. Pick the top 3 and write one concrete, owner-actionable step per theme. Format as a numbered checklist.

  9. Deliver the report. Structure the output with these sections in order:

    • Header — H2 with "Customer Pulse" and the date range.
    • Sources pulled — Bullet list with signal counts per source (PayPal disputes, HubSpot tickets, Gmail threads, Intercom conversations, pasted reviews). Note any source that was rate-limited and skipped.
    • Themes — For each theme, show a bold numbered theme label with the signal count, followed by two verbatim quotes as blockquotes, each attributed to its source.
    • Do these 3 things this week — Numbered list of three concrete, owner-actionable steps, each tied to one of the top themes.

    For a complete worked example, see reference/examples/example-report.md.

Approval gates

This skill is read-only — it does not post, send, reply, or modify any records. No approval gate is required.

Reference

來自 anthropic 的更多技能

access
anthropic
管理 Discord 頻道存取權限 — 核准配對、編輯允許清單、設定私訊/群組政策。當使用者要求配對、核准某人、查詢誰被允許時使用…
official
session-report
anthropic
從 ~/.claude/projects 的對話記錄中,生成一份可探索的 HTML 報告,內容涵蓋 Claude Code 工作階段的使用情況(包含 token、快取、子代理、技能及高成本提示)。
official
build-mcp-server
anthropic
當使用者要求「建立 MCP 伺服器」、「建立 MCP」、「製作 MCP 整合」、「為 Claude 包裝 API」、「暴露工具給…」時,應使用此技能。
official
cookbook-audit
anthropic
根據評分標準審核 Anthropic Cookbook 筆記本。每當要求進行筆記本審查或審核時使用。
official
handle-complaint
anthropic
處理進線客戶投訴的完整流程——提取背景資訊、草擬回覆,並建議營運改善方案。可接受選填的電子郵件或工單編號……
official
use-case-triage
anthropic
快速判斷某項處理活動是否需要PIA、強制性的GDPR DPIA,或可直接進行——揭露隱私政策衝突,並引導至正確的…
official
board-minutes
anthropic
根據您指定的格式,自動草擬董事會或委員會會議記錄。自動從您的日曆中偵測即將召開的董事會及委員會會議,詢問議程及…
official
renewal-tracker
anthropic
顯示即將到來的取消截止日期的合約,並在通知窗口關閉前發出警告,依據維護中的續約登記表進行操作。當用戶詢問時使用…
official