Site Audit
The polite crawler that scores your site — checks, severities and how runs are triggered.
Technical SEO → Site Audit runs a polite crawler over the project's domain and returns a scored list of technical issues.
How the crawl works
- Seeds: the crawl queue always starts from the homepage; a published sitemap is a supplementary source that adds discovered URLs to the queue (not a replacement for the homepage seed).
- Budget: manual runs default to 40 pages (max 60 per request); the
weekly cron uses
min(30, platform audit page limit)— scheduled sweeps are intentionally smaller than manual ones. - Politeness: ~150 ms between fetches, one domain at a time, standard crawler UA — no aggressive concurrency.
- Safety: private/internal addresses are refused by the SSRF guard (the same rule the REST API applies); non-HTML resources (PDFs, images) are detected, not audited for content checks.
What is checked
Per audited HTML page, with three severity levels:
| Severity | Checks |
|---|---|
| critical | broken link (page returns ≥400) · missing <title> |
| warning | page could not be fetched · missing meta description · no H1 · multiple H1 |
| notice | title > 60 or < 30 chars · thin content (< 300 words, non-homepage) · more than 3 images missing alt · missing canonical · non-HTML resource |
Every issue carries a detail line and a how to fix suggestion — the audit is written to be actionable by whoever ships the fix, not only by an SEO specialist.
The score
The audit run ends with a 0–100 score derived from the issue mix (criticals weigh most). Trends matter more than the absolute number: the weekly job keeps history, so "score after the migration" is a one-glance comparison.
When audits run
- Weekly cron (default Monday 06:00 UTC;
WORKER_*_CRONto change) for every active project — quota-checked like any scheduled job. - Manual run from the page (choose a page budget). A manual run never silently collides with the cron sweep — the two use separate job IDs.
Workflow
- Fix criticals first (broken links, missing titles) — cheap and high impact.
- Re-run the audit to confirm the score moves.
- Feed persistent warnings into the Growth Agent's recommendations — the agent reads audit issues as one of its inputs.