anthropics/customer-pulse-check
Synthesizes themes from PayPal disputes, HubSpot tickets, and review exports into a top-3 fixable issues list with drafted response templates. Accepts optional since-date argument.
npx skills add https://github.com/anthropics/knowledge-work-plugins --skill customer-pulse-check
Run the customer voice synthesis. Pull feedback signals from all connected sources, identify the themes that are actually fixable, and produce drafted responses the owner can review and send.
Parse arguments:
--since (default: last 30 days) — start date YYYY-MM-DD for the lookback windowUsing the customer-pulse skill workflow:
Cluster all signals into recurring themes. For each theme:
Using the ticket-deflector skill workflow:
Select the top 3 themes by: frequency × impact rating. For each:
Response template format:
Subject: Re: {issue topic}
Hi {first name},
Thank you for reaching out. {Acknowledgment of their experience in 1-2 sentences}.
{What we're doing about it / what happened / resolution offered}.
{Next step or offer}.
{Sign-off}
Format the output as:
Customer Voice — {date range}
Total signals: {n} ({PayPal disputes: n} | {HubSpot tickets: n} | {Reviews: n})
TOP 3 FIXABLE ISSUES
1. {Issue} ({frequency}) — {impact} — Fix: {one-line fix}
2. {Issue} ({frequency}) — {impact} — Fix: {one-line fix}
3. {Issue} ({frequency}) — {impact} — Fix: {one-line fix}
Run with whatever sources are connected — this command degrades gracefully. If PayPal is missing, skip dispute data and note "PayPal not connected — dispute data skipped." If HubSpot is missing, skip ticket data and note it. If no sources are connected at all, stop and tell the owner: "No feedback sources connected. Connect at least one of PayPal, HubSpot, or upload a review export CSV."
Present the summary table, then each response template. Ask the owner which templates they'd like to send, then wait for explicit approval before drafting the send.
Take anthropics/customer-pulse-check from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.