skillfulagents/public-skillset/technical-audit

Technical Audit

Not a report generator — an audit that ends in fixes. Findings without a shipped PR, a drafted ticket, or an applied platform change are unfinished work.

Read /workspace/seo/STATE.md and /workspace/seo/config.json first (site, publishing path, autonomy grants).

Phase 1 — Gather

1. Ahrefs Site Audit (if a crawl exists): pull the latest project's issues via the

API where available, else read the report in the browser. Note crawl date; if stale (>30d), kick off a re-crawl for next time and work with what exists + your own checks.

1. GSC: Index coverage signals — pages with impressions that 404/redirect, big

position/impression drops per page (seo-dashboard snapshot + gsc_query.py).

1. Own crawl of the essentials (curl + parsing; a small script is fine):

• robots.txt — present, sane, not blocking real content; sitemap declared.
• sitemap.xml — present, valid, reachable; URL count by section; every URL 200s

(sample if huge); no redirects/dupes inside; lastmod plausible. Diff vs the last audit's count (catch silent deindexing).

• llms.txt — present? (cheap win if not).
• Home + each key template (blog post, category/listing, product page): status,

canonical, title/meta uniqueness + length, H1, JSON-LD validity (parse it), OG/Twitter tags, noindex accidents, hreflang if multilingual.

• Redirects: http→https, apex↔www single-hop 301s; sample old URLs from

GSC/Ahrefs top pages for chains/302s/404s.

• Internal linking: from the sitemap sample, find orphan-ish pages (few internal

inlinks), check key pages are ≤3 clicks from home, anchor text is descriptive, important pages are linked from high-traffic pages.

• Duplicate content smells: parameter URLs indexed, /page/1 vs base, trailing-slash

duplication.

1. Performance: PageSpeed Insights API (free, keyless at low volume:

https://www.googleapis.com/pagespeedonline/v5/runPagespeed?url=...&strategy=mobile) on home + the 2–3 key templates. Record LCP/CLS/INP + the top opportunities. Mobile first — that's what Google scores.

Phase 2 — Prioritize

Rank findings by (indexing impact, traffic at risk, effort). Three buckets:

• P0 — blocking indexing/crawling (noindex accidents, sitemap 404s, robots blocks,

redirect loops, missing canonicals on duplicated content).

• P1 — ranking drag (slow LCP, broken structured data, thin/duplicate titles,

orphan money pages, 404s with backlinks — reclaim with 301s).

• P2 — hygiene (llms.txt, minor meta lengths, image alts, lastmod).

Phase 3 — ACT (the point of the skill)

Per the publishing path + autonomy grants in STATE.md/CLAUDE.md:

• Site in a repo you can access → open PRs (small, one concern each, with the

finding + evidence in the description). Merge only if the autonomy grant covers it; otherwise request review.

• Engineer will do the work → draft tickets (Linear/Jira/Asana via connected

account, or markdown handoff): symptom, evidence URL(s), suggested fix, priority.

• Hosted platform (Framer/Webflow/Shopify/Wix/Squarespace) → fix directly via the

platform API/MCP where one exists, else offer to drive the UI via browser with the user watching; document what was changed.

• CMS-level fixes (metas, alts, structured data fields) → apply through the

configured CMS path like any surface unit.

Queue anything needing the owner (DNS, billing, agency coordination) in seo/link-backlog.md "Needs owner" or the weekly report's decisions list.

Phase 4 — Record

• Write /workspace/reports/technical-audit-YYYY-MM-DD.md: findings table

(issue / evidence / priority / action taken or ticket/PR link), performance numbers, deltas vs last audit.

• Append a log.md entry; add follow-up [surface] items to the content backlog for

anything that becomes recurring work.

• Deliver the report per the report channel (email via email-report / Slack /

deliver_file) if run standalone; if run inside the weekly session, fold the summary into the weekly report.