SuperAPI AI Directory

crawl-diff

Marketing & growth

View repository

Updated 10-16 08:48

About this skill

Compare crawl snapshots

Use crawl-diff after a deploy, migration, or planned technical change. It crawls a bounded same-origin set, saves the result, and compares it with the previous saved run for the same surface. This turns status, indexability, canonical, metadata, content, and inventory movement into a review queue. It does not prove that a changed page affected search performance.

Run the report

With MCP, call seo_list_reports with {"category":"monitoring"}, then call seo_describe_report with {"id":"crawl-diff"}. Run the exact schema with seo_run_report, for example {"id":"crawl-diff","params":{"startUrl":"https://example.com/","site":"sc-domain:example.com","limit":250,"js":false,"refresh":true}}.

CLI parity uses the same definition:

seo reports describe crawl-diff --json
seo reports run crawl-diff --params '{"startUrl":"https://example.com/","site":"sc-domain:example.com","limit":250,"refresh":true}' --json

The schema types limit as a broad number. Pass a positive whole-page bound and keep it stable between comparison runs.

Interpret the output

Compare run with previousRun before reading summary. A missing previous run means the current crawl is a baseline, not a change result. Check the start URL, limit, URL count, timestamps, rendering choice, and warnings for both runs.

summary counts crawled, added, removed, changed, new errors, indexability flips, and high-priority recommendations. Trace every important count into items, where before, after, and changes preserve observed evidence. recommendations rank review actions; they do not override the raw snapshots. A removed URL may be outside the current discovered set or limit rather than deleted. A fetch or extraction failure cannot support a resolved issue.

Act safely

Verify high-priority status and indexability changes directly, then compare the deployment intent. Restore accidental failures or controls, but leave intended redirects, canonicals, and noindex directives alone. Repeat with comparable scope after a fix. Never call a bounded diff an inventory-wide audit or infer ranking impact from technical change alone.