A price written as $24.00 instead of 24.00 USD, a stock status of "available" instead of in_stock, or one stray tab inside a description is enough for Google Merchant Center to disapprove products or read a whole row into the wrong columns. This skill finds those and eleven other product feed mistakes before you upload, and tells you in plain words how to fix each one.

Your agent runs a scanner on your feed files: tab-delimited .txt and .tsv files and CSV exports with a header row (from your shop, a feed app, Google Sheets or Excel, including Excel's Unicode Text), and RSS 2.0 or Atom XML feeds with Google's g: namespace. It accepts attribute names with or without g: and with spaces or underscores, tells a full feed from a supplemental feed, reads XML safely (DTDs and entities are refused), and names the product id and the line in the file for every problem. When one problem repeats on hundreds of products, it lists the first 20 and counts the rest. It never uploads the feed, opens its links or contacts Google. It returns BLOCK, REVIEW, or NOT FLAGGED, with the file, line, reason, and a safer approach. Secrets are never printed.

What it checks

BLOCK (fix before shipping)

• F01: products without id, title, description, link, image_link, availability or price, and preorder or backorder items without availability_date

• F02: prices that are not a number and a currency code (currency signs, no currency, 15,00 EUR, negative or 0)

• F03: availability other than in_stock, out_of_stock, preorder or backorder, and condition other than new, refurbished or used

• F04: two products with the same id

• F05: product links and image links that are not full http(s) addresses

• F06: broken files: XML that is not well-formed, a wrong g: namespace, rows shifted by a tab or comma inside a value, values that run over several lines, or a feed with no products

REVIEW (check before release)

• F11: titles over 150, descriptions over 5000 and ids over 50 characters, and other values over Google's limits

• F12: links and images on http://, URLs with spaces, and extra image or mobile links that are not full URLs

• F13: new products without GTIN and MPN or without brand, placeholder brands such as N/A or Generic, and identifier_exists set to no next to a GTIN

• F14: GTINs with the wrong length or check digit, 10-digit ISBNs, and in-store or coupon numbers

• F15: sale prices that are not lower than the price or use another currency

• F16: titles with promotional text (free shipping, SALE!, 50% off), ALL CAPS, or emoji and decorative symbols

• F17: Apparel & Accessories products without color, gender or age_group, or clothing and shoes without size

• F18: prices written with the currency code first (USD 15.00) or with more than 2 decimals (1.305 SEK)

What's included

• The scanner (Python 3.9+, standard library only)

• A rules guide with the reason and safer approach for each rule

• A risky and a fixed example feed (tab-delimited and XML) for a made-up shop, and the real report the scanner produces on the risky one

Limits

It checks the feed file as written. It does not open product pages or images, so a price or stock status that differs from the landing page, image quality and policy problems are not checked, and it does not know which countries you sell to (shipping, tax and country rules). Attribute names must be the English ones, and feeds in Google Sheets, the Merchant API or .gz files must be exported or unzipped first. Files over 50 MB are listed as not checked. NOT FLAGGED is not a guarantee of approval.

Contents

One ZIP file with 10 files: SKILL.md, the scanner in scripts/ (3 Python files), references/rules.md (every rule with its reason and safer approach), and examples/ (4 example files plus report.md, the real scanner output on the risky example).

How to use: This is an agent skill: a folder with a SKILL.md file and Python scripts that an AI coding agent such as Claude Code, Cursor, or Codex CLI can run. Unzip it into your agent's skills folder (for Claude Code: ~/.claude/skills/), then ask your agent to check your project. The scripts also run on their own with Python 3.9 or newer, with nothing to install. They read your files only: no network access, and nothing is sent anywhere.