Listing Rewriter

Source: a garden-level guest suite in a US mountain city, from a public open-data snapshot of listings. The title and description were written as an emoji list.

Before: a title of three words and two bed emoji, and a description of short lines such as "15 min walk to Downtown", "Uber is super fast" and "Maximum of 2 guests", each followed by emoji.

After (skill output, unedited, current version):

1. Garden-level suite with private entrance (40 characters)
2. Walk to South Slope restaurants and breweries (45 characters)
3. Self check-in suite, 15-min walk downtown (41 characters)

"Garden-level guest suite with a private entrance, in a neighborhood setting. It's a 15-minute walk to downtown and an easy walk to South Slope restaurants and breweries. Uber is quick. Sleeps up to 2 guests, with a mini fridge, microwave, Keurig coffee maker and free driveway parking (2 spaces)."

First item under "Confirm before you publish": "What sizes are the 2 beds? If they're two twins, add to "Your property": "The suite has two twin beds.""

Every feature in the "after" is in the host's own title, description or amenities list.

Paste your current listing title, description and amenities. You get three titles under 50 characters, a description within 500 characters, and the "Your property", "Guest access" and "Other details to note" fields. Each field comes in its own copy block. Everything is built from what you gave it, so the new listing holds no surprises for a guest.

Tested on 4 real public listings. Final Claude Sonnet scores: 9, 8.5, 9 and 8.5 out of 10.

4 real tests on public listing data: an entire home, two guest suites and a tiny home. One had a one-line description and one was written as an emoji list. Scored by hand against a written 10-point rubric. The first version scored 6, 6, 6 and 9: it called a home "cedar" because of its name and called a fire pit private when the listing did not. Both are now ruled out. Final scores on Claude Sonnet: 9, 8.5, 9 and 8.5 (test 04 re-run on the current version). The earlier Claude Opus runs (9, 9, 8.5, 9) were scored from grader notes, not checked line by line, and one ran with unrelated connectors attached, so we don't count them. On the current version, one Opus run on a fresh public listing from another city scored 8.5, checked by hand. An independent review re-scored two of our tests and ran two fresh listings; its fixes (no bracket placeholders in the confirm list, a sourced rule for homes unsuitable for children) are in this version.

Update 2026-10-04 (second check): run on a listing we wrote with contradictory and hedged details ("maybe 10 minutes", "might close", "sleeps 6 if you are friendly"). The first run turned a hedged beach distance into a firm title; the skill now keeps hedged facts out of titles and a re-run came back clean. One run, not scored.

The tests ran on real public data, never made-up cases, and were scored against a written rubric. A skill in this range ships only when every test scores 8/10 or more.

How we test, what we do not do, and which assistants we ran it on.

• It works from the text you paste. It does not open your listing, see your photos or know your neighbourhood.
• The title guidance and field names follow Airbnb's public host pages as read on 4 October 2026. The 500-character description limit comes from channel-manager documentation and hosts, not from an Airbnb help page. If your editor shows a different limit, follow your editor.
• Its choice of one title over another is a judgement. It has no data on what guests search for and does not claim any.
• It gives no advice on pricing, permits, tax or insurance.
• It does not guarantee results. Read the output before you use it.

Works with any assistant that reads SKILL.md skills (Claude, Codex, Gemini CLI, Cursor and others), plus a paste-in prompt version for ChatGPT or any chatbot. We run our release tests on Claude Sonnet and Opus. Use your assistant's strongest model: in our tests, a small, fast model added facts that were not in the source data. A small script counts the characters. Without code execution the skill still works, but the counts are not verified.

1. Unzip the download. It holds the listing-rewriter folder, ready-made zips for claude.ai in a claude-ai folder, a README and the licence.
2. Put the listing-rewriter folder where your assistant looks for skills. The generic path is .agents/skills/ (or npx skills add from a folder). Tool-specific paths:
   • Claude Code: ~/.claude/skills/ (or .claude/skills/ inside one project)
   • OpenAI Codex, Gemini CLI, Cursor and others: .agents/skills/ or the skills folder their docs name
   • claude.ai: Customize, Skills, then the plus button, Create skill, Upload a skill, and upload the zip from the claude-ai folder without unzipping it (code execution must be on in Settings, Capabilities)
   • ChatGPT or any chatbot: open the skill's prompt.md, paste it into the chat, then paste your own text or link
   • Steps for each tool, with links to its official docs: how to install Agent Skills
3. Ask for the job in plain words, for example: "Rewrite my Airbnb listing. Here is my title, description and amenities: ...". The skill loads by itself.