My personal landing site had been live for months and I assumed it was "fine." I ran a multi-agent SEO audit built on the seo-audit skill from Corey Haines' marketingskills collection — five agents in parallel, each owning one dimension: content/E-E-A-T, schema markup, GEO, images, and technical SEO. They came back with roughly 40 findings and category scores between 42 and 78 out of 100. Not fine.
The embarrassing headliner: my favicon was 12.5 MB. One file, downloaded by every visitor, heavier than the rest of the site combined. Rebuilt it down to 25 KB. Then the rest of the critical/high tier: converted images to WebP (~5.5 MB saved), added meta descriptions, canonicals, OG tags and JSON-LD, generated static hreflang pages for the ES/RU versions, added a 404 page, and fixed the www→apex redirect at the CDN (CloudFront) level. A second pass took the medium tier: cache-control headers, ~16 MB of stray deployed files removed, and a few stale copy bugs the content agent had flagged.
What made this work was the split: audit as a fan-out of narrow specialists, fixes as one ordinary session working a ranked list. No single agent would have caught both the favicon and the hreflang gaps.
Reproduce this
Prerequisites: a static site you can redeploy; an agent runner that can fan out subagents; the seo-audit skill installed.
Inputs your agent collects: site URL, deploy method, list of locales (if any).
Steps:
- Fan out one auditor per dimension — content, schema, images, technical, GEO — each returning scored findings.
- Merge into one ranked list.
- Fix critical/high items in a single session, redeploy, re-check.
- Sweep the medium tier separately.
Verify: favicon and image weights in DevTools network tab; redirect with curl -I; structured data with a rich-results checker.