CommunityBildgithub.com

review-posts

Verwandelt die neuesten 5-Sterne-Bewertungen Ihres Unternehmens (Google, Yelp, BBB, Facebook, Trustpilot, Houzz, Angi, die eigene Website oder kopierten Text) in fertige Instagram-Bilder (1080×1350) im eigenen Markendesign – mit Bildunterschriften und einem Freigabeblatt.

Was ist review-posts?

Diese Fähigkeit hilft Unternehmen und Marketingfachleuten, aus ihren besten Online-Bewertungen schnell optisch ansprechende Instagram-Beiträge zu erstellen, ohne technische Kenntnisse zu benötigen. Unterstützte Plattformen umfassen Google, Yelp, BBB, Facebook, Trustpilot, Houzz, Angi, die eigene Website oder direkt eingefügten Bewertungstext. Beim ersten Einsatz richtet die Fähigkeit das Markenprofil automatisch ein: Sie extrahiert Logo, Farben, Schriftarten, Hintergrundfotos, Bewertungslinks und Layout-Optionen von der Website des Unternehmens über das Skript `brand_kit.py`. Danach generiert sie mit `make_posts.py` Instagram-konforme Bilder im Format 1080×1350 Pixel, inklusive passender Bildunterschriften und einer Übersicht zur Freigabe. Die Fähigkeit postet niemals selbst ins Internet – der Nutzer prüft und veröffentlicht die Inhalte manuell. Sie ist ideal für wiederkehrende Aufgaben wie „Aus unseren Bewertungen Instagram-Posts machen“ oder „5-Sterne-Bewertungen teilen“ und stellt sicher, dass alle Beiträge konsistent im Corporate Design erscheinen.

Funktioniert mit✓Claude Code~Codex CLI~Cursor
npx skills add https://github.com/javiervillag/review-posts/tree/HEAD/review-posts

Installed? Explore more Bild skills: steipete/songsee, affaan-m/frontend-design, mattpocock/improve-codebase-architecture · View all 6 →

In Ihrer bevorzugten KI fragen

Öffnet einen neuen Chat, in dem dieser Agent-Skill bereits geladen ist.

Dokumentation


name: review-posts description: Turns any business's newest 5-star reviews (Google, Yelp, BBB, Facebook, Trustpilot, Houzz, Angi, the business's own website, or pasted text) into ready-to-post Instagram images (1080x1350) in the business's own brand, with captions and an approval sheet. The first time, it sets the brand up from the website: logo, colors, fonts, background photos, review links and a choice of layouts. Use when someone says "make review posts", "turn our reviews into Instagram posts", "post our 5-star reviews", "review graphics", "quote cards from reviews", "set up review posts for my business", or pastes a review and wants a post. It never posts anything itself.

Review posts

You're helping a business owner or marketer who is probably not technical. Keep messages short, warm and plain. Never show code, JSON or file paths unless they ask, and do the technical work quietly.

This skill never posts, publishes or replies online. The person approves the images and posts them themselves.

Tools: everything is in scripts/, next to this file. The scripts are brand_kit.py (reads a website) and make_posts.py (makes the images). Both need Python 3 and Pillow; if you see "Pillow is missing", run pip install pillow. Default fonts are bundled.

The business folder holds everything about one business: settings.json, the logo, photos/, fonts/, and seen.json (reviews already turned into posts). Name it <Business Name> Review Posts/ and keep it in the working folder, which in Cowork is the folder the person picked. If it already has a settings.json, skip setup and go to Every week below.


First time: set up the brand (about 10 minutes)

Tell the person in one line: "First I'll set up your look from your website. Takes about 10 minutes, once."

1. Ask two things, in one message

  1. "What's your business name and website?"
  2. "Where do your reviews live?" Offer: Google, Yelp, BBB, Facebook, Trustpilot, Houzz, Angi, Nextdoor, or testimonials on their website. Say: "Paste the links if you have them. If not, I'll find them on your website."

2. Pull the brand from the website

Make the business folder, then run:

python3 scripts/brand_kit.py "<website>" --folder "<Business> Review Posts"

It saves logo candidates, big photos, any Google Fonts, and brand-kit/brand-kit.json, which lists the colors, fonts, phone number and review links it found.

  • If the script can't open the site (blocked, or no internet in your sandbox): open the website in the browser instead. Screenshot the logo, read the colors and fonts with javascript_tool (below), and ask the person to send their logo file if needed.
  • Fonts not found (common: many sites load fonts with JavaScript): in the browser, run ['h1','h2','p'].map(t => t + ': ' + getComputedStyle(document.querySelector(t) || document.body).fontFamily) Then fetch each family with python3 scripts/brand_kit.py --get-font "<Family>" --folder "<folder>". If it isn't a Google Font, use the closest bundled one: Playfair Display (elegant serif), Poppins (clean sans), or Raleway (labels). The person can also send the .ttf file.

Then look at every file yourself and choose:

  • Logo: the cleanest full logo (name included), PNG, ideally transparent. Skip partner, award, certification and review-site badges. Copy it to logo.png.
  • Colors: start from colors.suggested in brand-kit.json, and check it against the logo.
    • dark sits behind white text, so it must be dark (navy, forest, charcoal, deep teal).
    • darker is a deeper shade of it.
    • accent is the stars, highlights and lines: the brand's bright color.
    • accent_light is a lighter tint of the accent.
    • If the brand has no dark color, use a very dark shade of its main color.
  • Photos: see step 3.

3. Background photos

The images look best with a real photo behind the quote, and the photo changes on every post. Pick the source in this order:

  1. Their own photos: keep good ones from brand-kit/ (their website) and copy them into photos/. Also ask: "Got any photos of your work? Drone shots, finished jobs, your storefront or team? Drop them here."
  2. Free stock photos that match the business and area, if they have few of their own. Follow reference/photos.md. Rules: free for commercial use, credited in photos/PHOTO-CREDITS.md, and showing what the business works on. For example, homes and roofs for a roofer (never downtown skylines), smiles and a calm clinic for a dentist, plated food for a restaurant.
  3. No photos: the "brand color" layout needs none.

Quality bar for every photo:

  • At least 1500px on the long side. Never upscale a small one; it turns blurry.
  • No text or watermarks, nothing dark or cluttered, and no faces of identifiable strangers.
  • Aim for 8 to 15 photos, so the feed doesn't repeat.
  • Name the files by kind (job-01.jpg, area-01.jpg, and so on). The rotation then alternates between the kinds.

4. Show the look options

Write settings.json (see reference/settings.md) and run:

python3 scripts/make_posts.py --folder "<folder>" --preview

It makes the same sample review in each layout, using their logo, colors, fonts and photos. Show all of them, then ask "Which do you like?":

  1. Big quote on a photo: bold, best for short reviews.
  2. Review card on a photo: looks like a real review, best for longer ones.
  3. Brand color, no photo: clean, and needs no photos.
  4. Carousel: 2 slides for very long reviews. Slide 1 has the best line, slide 2 the full review.

Recommend Automatic: short reviews get layout 1, longer ones get layout 2, and only very long ones (70+ words) become carousels. Fix whatever they don't like (logo too small, wrong color, a bad photo crop), run --preview again, and repeat until they're happy.

5. Captions

Ask once: "What should every caption end with? Most people use a phone number or 'Book online' plus a few hashtags." Suggest one using the phone number from brand-kit.json and 3 or 4 local hashtags. Save it as caption.extras and caption.hashtags.

6. Wrap up the setup

Delete brand-kit/ and setup-previews/; they were only for setup. Then tell them, in a short list, what they can change any time just by asking:

  • "Use these photos as backgrounds" (or "remove the photo with the pool")
  • "Make the logo bigger" or "use our color logo"
  • "Change the gold to our orange"
  • "Only use the review card look" or "no carousels"
  • "Add Yelp" (or any other review site)
  • "Change the caption ending"

Then go straight to Every week and make their first posts.


Every week: make the posts

1. Get the newest 5-star reviews

Use review_links in settings.json. If you have a browser (Claude in Chrome, or Cowork's browser), open each link and follow reference/find-reviews.md. Read only: never click Reply, Like, Report or Write a review. If you can't browse, or a page is blocked: ask the person to paste the reviews in.

Rules:

  • 5-star only. Skip anything lower, unless the person asks otherwise in settings.json (only_stars).
  • Same review in two places (a customer often posts on Google and BBB, or a site reposts Google reviews): make one post. Prefer the copy with no redactions.
  • Photos in the review: if the reviewer attached photos of the work, download the largest one into photos/from-reviews/ and set "photo" on that review. It becomes that post's background.
  • Reviews already in seen.json are skipped automatically. Say how many were new.

2. Write reviews.json (you do this; the person never sees it)

The format is at the top of scripts/make_posts.py.

  • quote: the full review, word for word. Never fix spelling or grammar.
  • name: first name plus last initial ("Sarah M.").
  • source: where it was posted ("Google", "Yelp", and so on). It's shown on the image.
  • excerpt (long reviews only): the 1 to 3 strongest whole sentences, word for word.
  • hl: one short phrase to highlight, word for word.
  • Hidden names: some sites (BBB) show staff names as "REMOVED". Don't leave "REMOVED" in, and never guess the name.
    • First, try an excerpt that avoids it.
    • Otherwise, swap it for the plainest role word that reads naturally, with no brackets. For example: "edits": [{"from": "REMOVED(owner)", "to": "The owner"}].
  • caption: under 80 words, in the business's own voice.
    • Thank the reviewer by first name.
    • Mention one specific thing they said.
    • End with caption.extras and the hashtags.
    • Never invent details.

3. Make the images

python3 scripts/make_posts.py --folder "<folder>" reviews.json

Look at every image before you show it. Check for:

  • cut-off text
  • a bad photo crop (fix it with photo_focus in settings.json, then run again)
  • a photo that doesn't fit the business (remove it from photos/)

4. Hand over

  1. Show the images grouped per post, with each caption underneath.
  2. Give them the posts/<date> folder, zipped if you can. It holds the images, captions.md, and approval.csv (opens in Excel or Google Sheets).
  3. Tell them in plain words:
    • how many posts were made, and how many are carousels
    • anything that was skipped or edited
    • that nothing was posted
    • next time, they just say "make review posts".

Verwandte Skills