A local-first SEO audit + recurring Action Runbook
Turn real site, Search Console, and GA4 evidence into a client-ready audit and an Action Runbook. See what needs fixing, review the work, and measure what changes.
Open the file in Cursor, Claude Code, or Codex. Free, with no account required.
Don't have a website for SEO yet? Make your own website for free — a step-by-step guide to building and publishing one landing page with your own AI assistant.
Made by Chris Cooke · Finalist, The Vibe Coding Games · About
Works with Claude Code · Cursor · Codex
Start with one page and two public scores: PageSpeed Insights Mobile Performance and the SEO Audit Report Builder score. Submit before scores to join the worst-sites board, then come back with after scores to compete for best site and biggest improvement.
A homepage, launch page, pricing page, or product page with PSI Mobile Performance and audit scores.
Rank on worst starting point, best current score, or biggest before-and-after improvement.
The downloadable runbook creates the full audit, then compiles a separate Action Runbook for approved, recurring work.
The expertise lives in a plain-text runbook. Your AI profiles the site, gathers evidence, writes the audit, and prepares bounded action batches on your machine. You remain the approval gate.
A Site Profile activates only the relevant modules and records why international, ecommerce, local, media, large-site, experiment, or stack-specific packs are active or N/A.
Drives your logged-in browser with Playwright to read Search Console and GA4. You handle login and 2FA; credentials are never handed to the runbook.
A dated, client-specific SEO audit explains what matters. A separate dated action runbook maps every recommendation to what the AI can do, draft, guide, research, or leave to a human.
The audit never edits the site. The action document proposes batches of no more than five items, shows research and risk, and waits for approval before making changes.
The Ledger, change events, content decisions, backlink baseline, and fixed SERP panel survive between cycles, so the next run compares evidence instead of starting over.
Run it in Claude Code, Cursor, or Codex. Your working data stays with your chosen local AI workflow, and the markdown files remain readable, editable, and yours.
Most audit deliverables stop at recommendations. This keeps the diagnosis separate, then turns it into a controlled operating document you can return to every week, month, deployment, and quarter.
| What is being compared | Typical SEO SaaS | SEO agency | This tool |
|---|---|---|---|
| Cost | $100–500 / month | $1k–5k per audit | Free |
| Your data | Uploaded to their servers | Shared with the agency | Stays on your machine |
| Customisation | Fixed feature set | High — but you can't touch it | Full — edit the runbook |
| Ongoing work | Dashboard and alerts | Retainer or another project | Recurring, approval-led batches |
| Uses your own AI | No | No | Yes |
| Open methodology | No — black box | No — their secret sauce | Yes — read every step |
Run or commission an audit → receive a long list of recommendations → manually rediscover the context when it is time to act.
Generate the report → open its linked Action Runbook → approve a small batch → verify it live → measure at 2, 4, and 8 weeks.
Get a consultant-grade audit and a practical route from recommendation to verified change.
Keep research, approvals, implementation evidence, and measurement in one repeatable operating record.
Deliver a defensible report plus a client-specific action document without turning your method into hosted software.
If you can download a file and open an app, you can run the audit and return for later action cycles. There's no
git clone, no
bash install.sh, and no terminal
— unless you want one.
An app you install like any other — no terminal needed. Download it, open the folder, and the built-in AI agent runs the audit.
The runbook walks you through adding Playwright — takes 30 seconds. Also works with OpenAI Codex — same runbook, same steps.
If you're comfortable in the terminal, Claude Code gives the fastest agent experience. Same runbook, same result.
npm install -g @anthropic-ai/claude-code Optional: add the
skill pack
for /seo-audit tab-complete. The orchestrator
verifies your tools, invokes analysis skills at the right time, and enforces Ledger
discipline — same runbook, structured execution.
The runbook guides the agent from site profiling and evidence collection to a finished report, then compiles a separate action document. Diagnosis stays diagnosis; implementation needs approval.
Site Profile routing, GSC, GA4, PageSpeed, crawl evidence, live rendering, indexing, and only the conditional modules the site actually needs.
Six core analysis steps plus three public helpers for AI-copy editing that preserves CTAs, traffic trends, and claim verification. Explore and install the skills.
A prioritized 90-day report with a 0–100 health score, then a separate client-specific action map generated only after report QA passes.
Research current guidance, propose up to five actions, implement only what is approved, verify local/deployed/live states, then measure outcomes.
The Action Runbook carries the report's evidence forward. It researches guidance that may have changed, offers help by capability and risk, and keeps recurring work proportionate to the site.
For 1–10 eligible URLs, stage a separate Playwright tab for each Search Console inspection. For larger sets, update a canonical sitemap with truthful lastmod dates and reserve manual requests for the highest-priority pages.
Protect, refresh, merge, differentiate, redirect, noindex, remove, or investigate using demand, conversion, link, SERP, technical, user, and business evidence—not age or traffic alone.
Every new, refreshed, or merged page gets an evidence-backed brief with intent, first-party value, facts still needed, approval gates, acceptance criteria, and measurement.
Record Crawl Stats, but request server logs only for large, rapidly changing, or genuinely crawl-constrained sites. Small healthy sites do not get crawl-budget theatre.
Keep backlink sources separate and preserve a fixed SERP panel—query, market, device, page type, competitors, and features—so the next cycle can compare like with like.
When source access and the stack are known, use a framework-specific adapter, protect unrelated work, build and deploy safely, check affected and unaffected pages, and record what changed.
A genuine audit, anonymised end-to-end — SEO Health Score, 90-day plan, and prioritised, data-grounded fixes. Your own finished report then links directly to its generated Action Runbook.
Drop the runbook in an empty folder, open it in your AI tool, and say:
No site of your own to test on? Reproduce the example report with these sanitised sample inputs:
Two downloads. The audit runbook generates both your client-specific report and its recurring action runbook locally. The methodology explains the why.
Plain markdown · No proprietary formats · No lock-in