IMPROVE auto+gate refresh

No page goes stale without you knowing

The file

take it and run it.

One markdown file: the goal prompt, the verifier prompt, and the stopping condition. Read it before you run it. MIT licensed, no account, nothing to sign up for.

Download refresh.md View the raw file

curl -fsSL https://vibemarketers.ai/loops/refresh.md -o .claude/loops/refresh.md

The loop

copy it.

Content rots in two ways: the facts go out of date, and the sources they rested on disappear. This loop checks both on a schedule so you find out before a reader does.

Use it when you have more pages than you can personally remember the state of, which is most sites past about twenty pages.

You need your pages and some notion of when each was last verified. A date in the frontmatter is enough.

You get a dated report: every page with its age, what crossed your threshold, and which pages make claims with no source behind them. That second list is usually the uncomfortable one.

The gate is you approving rewrites. The loop reports by default and only edits what you tell it to.

Goal prompt

For every page in <collection>, read the last verification date. Flag every
page older than <threshold> days. For each flagged page, check that every
source link is present and well formed, and report any page that makes claims
with no source link at all. Write the report with a date.

Verifier prompt

Open the report. FAIL if any page in the collection is missing from it, if a
page is flagged without its age, if a dead link is reported as live, or if the
report has no date. Otherwise PASS.

Stopping condition

Every page in the collection has an age and a link status in the report, and the verifier marks PASS.

MCPs it needs

none required

Want everything at once

The installer writes the three starter skills and every published loop into .claude/. Read the script first.

curl -fsSL https://vibemarketers.ai/install.sh | bash

Or have us run it on your site →

The receipt

Receipt refresh 2026-09-25

Baseline: 13 term pages with a last-updated date of 2026-08-19 and no freshness check of any kind. Nobody had checked whether their claims still carried sources.
Hypothesis: A dated freshness check will find pages past the 21 day threshold, and checking source links at the same time will find claims standing on nothing.
Intervention: Read the updated field and body of all 13 term pages. Computed age against a 21 day threshold and counted source links per page.
Result: 13 of 13 pages are 37 days old, past the 21 day threshold. 12 of 13 carry no source link at all, so their claims rest on assertion. No page was rewritten: this run reports, it does not edit.
Commercial signal: none
Attribution limits: Age is not accuracy. A 37 day old page can be entirely correct and a fresh one wrong. The absence of a source link flags a risk, not an error.
Learning: The link check found more than the date check. Staleness was expected; that 12 of 13 pages make claims with no source was not, and it contradicts the evidence rules this site publishes on its own method page.
Next intervention: Add sources to the 12 pages, then rerun so the report distinguishes a page that was checked from one that was merely edited.

0 human touches

3 minutes

0 errors

raw run logopen →

Sources: refresh report, lab/refresh, collected 2026-09-25 , window: 13 pages, 21 day threshold, single run .

One entry per run, append-only, newest first. Corrections append a dated line.

Inside a company

where the lane changes.

VibeMarketers tests the loop. SuperMarketers runs it on a real team.

Problem: Pages age quietly. Nobody notices a claim went out of date until a reader does.
Source of truth: The page collection and the last verification date on each page.
Permitted actions: Read pages, check dates and links, write a report. Rewrite only what a human approves.
Required approvals: The content owner approves each rewrite before it publishes.
Owner: Content lead
Metric: Share of pages verified within the threshold, and pages making claims with no source.

On a real team the threshold differs by page type: pricing and comparison pages age in weeks, definitional pages in months.

Running this on your own site is one person reading a diff. Running it inside a company changes four things, and the guide covers them. When it stops being a file you copy and becomes a build, SuperMarketers does the same loop model inside your stack, on your keys.