When should you use it?
- Reviews go unanswered
- Your rating dropped and you don't know why
- Repeat customers stopped ordering
Not for: Posting replies without reading them first
What do you need, and what do you get?
You need
- Review exports or pasted reviews from Google, Yelp and Facebook
- Your customer or job list with last order dates (from a CRM, booking tool or storefront)
- Tickets, disputes or complaint emails (optional)
You get
- Themes with quotes
- Draft replies
- At-risk customer list
- Win-back drafts
How does it work, step by step?
- Install the pluginIn Claude Cowork, install the small-business plugin from claude.com/plugins; in Claude Code, add the knowledge-work-plugins marketplace and install it.
- Connect your toolsConnect the apps the skill reads from. Without connectors, most skills work from pasted or uploaded files.
- Customize itGive it your reply style (sign-off, how you handle complaints) and what you can offer in win-backs.
- Run itAsk in plain language or use the the review-reputation skill command, then review the output before anything leaves your hands.
The prompt
Our adapted version, written to work in Claude Cowork and Claude Code. Values in lime are placeholders.
Fill in the blanks (optional; the copied prompt uses your values)
What are customers saying about {business}? Use these reviews: {review_export}, and this customer list with past order or job dates: {customer_list}. For the last {days} days, give the rating picture from these reviews only, then three to five themes ranked by how many customers mention them, each with exact quotes and their source. Draft a reply under 60 words to every review, in my voice. List customers whose gap since their last order is well past their own usual rhythm, ranked by what they have spent, and flag anyone who left a bad review and then stopped buying. Before drafting any win-back message, ask me what offer I can honor. Do not post or send anything.
The original skill is published by Anthropic on Anthropic. Open it there to install or import it; the prompt above is our own adaptation.
What can go wrong?
- Never offer incentives in exchange for reviews; platform rules forbid it.
- Handle complaints that mention safety or legal issues personally, and agree any discount before it appears in a win-back draft.
Our take
Reputation work usually falls to the owner at night, so it doesn't happen. This skill bundles three jobs that belong together: listen (themes with real quotes), respond (drafts), and retain (who's gone quiet). For local service businesses this pairs naturally with SEO and Google Business Profile work. We ran it on its documented fallback: a 90-day export of reviews from two platforms plus a customer list with past job dates, for a fictional heating and cooling company. The useful discipline showed in three places. The rating summary used only the reviews provided, and the missing third platform was named as a gap rather than ignored. Themes were ranked by how many customers raised them and backed by exact quotes, so scheduling problems came out ahead of one angry pricing review. Quiet customers were judged against their own rhythm: a maintenance-plan customer who usually books every quarter and had been silent for eight months made the list, while an annual customer did not. A one-star review describing a visit with no matching job got one calm, factual reply, and the win-back step asked what offer was approved instead of inventing a discount. Our version adds that last check up front.
Where the time goes back: Turns scattered feedback into a short list of themes and ready-to-post replies.
FAQ
Can it reply to reviews automatically?
Does this only work in Claude?
Source and credit
This playbook is based on work by Anthropic. We summarize it and link to it and write our own adapted prompt.
- Original
- review-reputation skill · small-business plugin (anthropics/knowledge-work-plugins)
- Creator
- Anthropic
- Published on
- Anthropic
- License
- Apache-2.0
- Originally for
- Claude Cowork
- AI dry run
- Oct 21, 2026: passed. Dry-run tested by our AI test agent on sample input. Arithmetic and theme counts checked by hand; no invented offer; nothing posted.
Agent dry run of the published SKILL.md: fetched from the source repo, its steps followed on a realistic synthetic input, output checked against the skill's format and our guardrails. Plugin install inside Claude Cowork/Code not exercised. - Verified
- Oct 23, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
The original is licensed under Apache-2.0 by Anthropic. Modified: our adapted prompt and summary are our own wording; the original skill {"was unchanged at the source when we snapshotted it (live source later removed)" if src_removed(s) else "is unchanged at the source link"}.
Are you the creator? Claim or correct this page.