Best Before: what went stale on your site, refreshed honestly

Ask it in your own words

• "Which of my pages need updating?"
• "We changed our pricing. What else on the site is stale?"
• "Should I update, merge or delete this old post?"
• "half my titles still say 2024, can i just swap the year"

What happens next

1. It inventories the perishable facts. find_dated_facts.py scans a folder, a sitemap or a list of URLs and gives each fact a type and the date it rests on.
2. It builds the register and ranks the work. One row per distinct fact, with its source, its next check and its owner. Pages are ordered by importance, staleness and risk.
3. It runs the refresh. Each fact is re-verified at its source, then patched on every page that carries it, and the result is logged.

What you get back

• A freshness report: counts by status, the riskiest findings, ranked pages
• The decay register as a CSV, with check dates and owners
• Date findings: visible date, structured data and sitemap that disagree, an unreliable lastmod, event dates that have passed
• A refresh level per group of pages, from "verify only" to "merge, redirect or retire"

Run on this very site

The home page and the terms of sale, scanned for facts that will go out of date.

pages scanned: 2 [...]; 38 fact sentence(s) found = 40 distinct fact(s) once repeats are merged by entity + attribute + value
status (age against the class cadence): overdue 0, page-dated 18, no date of any kind 18, due-soon 0, ok 2 [...] future-date 0, event-passed 0

Half the facts had no date of any kind, because the home page shows none. The one with a known end, the launch price until 31 October 2026, has its change written down for 1 November.

What it will not do

• The scan cannot know whether a fact is true, only how old the page says it is: truth is checked at the source.
• Without page-level search data (impressions, clicks, links), it will not decide to merge, redirect or retire a page: it says so instead of judging by age.
• It will not move a date when nothing changed, and promises no ranking or traffic gain.

What it needs

Claude Code or the Claude apps, and your content: a folder, a sitemap, a list of URLs or pasted pages. The two scripts need Python 3.9 or later; without them, Claude reads each page by hand. A Search Console export helps but is optional.

How to install a skill in Claude

Where it hands over

• Truth Serum: It re-verifies each fact at the source and finds the old value across the site.
• Gap Hunter: When the search results moved, not the facts, its refresh mode reworks the article.

Questions about this skill