SEO audit

Find the on-page and structural problems holding a site back, ranked by what they cost and what they take to fix.

What you need first

• Site url — there is nothing to fetch without it, and the wrong domain wastes the whole run
• What the site is for — whether a thin page is a problem depends on whether it was meant to rank

Check what you already know from this workspace and the conversation. Ask for everything still missing in a SINGLE message, then wait.

If no answer comes, do not guess your way through. Produce whatever is genuinely useful without the missing facts, state at the top which ones you lacked, and say what would change once you have them.

Earlier runs: if seo-audit-<slug>.md already exists here (<slug> is the site's host or path, lowercase, dots and slashes as hyphens — acme-com), say when it was last written (its Updated: line) before you start. This run replaces it; CHANGELOG.md and git keep the history.

What you decide, and what you do not

Decide yourself — which pages to crawl, how findings are grouped and what counts as urgent. Act; do not ask and do not flag.

Decide and flag — the page list when no sitemap was found and any page excluded for size or time. Proceed on a named assumption and put it at the top of the output.

Stop and wait — changing anything on the site itself. Never resolve one of these by assumption, however long the wait.

How you work

Say only what you observed. Every claim in the output traces to something you actually read, fetched or were told. If you did not check it, do not assert it.

Report what you could not do. A page that would not load, a source you could not reach, a step you skipped — these go in the output. Dropping them silently turns partial work into work that looks complete, which is worse than work that looks partial.

Name your assumptions. If you had to assume something to proceed, put it at the top of what you produce, in one line. An assumption stated is corrected in seconds; an assumption buried becomes a fact nobody checked.

Finish or say you did not. Do not pad to length, and do not present a first pass as a final one.

Before you start

Establish what you need before producing anything. A confident answer about a business you have not described is the most expensive kind of wrong: it reads as authoritative and nobody catches it until it is in front of a customer.

Sources

Cite what you used. Name the URL, document or statement behind each finding, inline, where the finding is. A sources list at the bottom that nothing points to is decoration.

Prefer what you fetched to what you recall. When they disagree, the fetch wins and you say so. When you could not fetch, say that instead of filling the gap from memory.

Do not launder a guess through a citation. Linking a plausible source next to an unverified claim is worse than the bare claim, because it borrows credibility the claim did not earn.

Sequencing

Work in the order the procedure gives. Each step exists because the next one is worse without it — research before writing, inventory before fixing, baseline before change.

If you must skip a step, say which and why in the output. A silently skipped step is indistinguishable from a step that found nothing, and the two mean opposite things.

Partial work

Long work gets interrupted. If you cannot complete every step:

• Deliver what is finished, clearly labelled as partial.
• State exactly where you stopped and what remains.
• Never present a subset as the whole. A three-page audit described as a site audit is a false claim about coverage, even when every page in it is correct.

Procedure

1. Scope the crawl. Fetch the homepage and sitemap.xml (or /sitemap_index.xml). Build a page list capped at the 30 most important URLs: homepage, service and product pages, top posts, contact. If there is no sitemap, build the list from the homepage's own navigation and say so at the top of the report — a list you inferred is a different claim from a list the site published.
2. Fetch each page. Request every URL on the list. Record the HTTP status, any redirect chain, and every page that failed. Never guess at the content of a page you could not fetch.
3. Check the on-page elements. For each page, extract the title, meta description, H1 and heading structure. Record the observed text and its character count, then flag: missing or duplicate titles, titles over 60 characters, missing metas, more than one H1, headings that skip a level.
4. Map the internal links. From the fetched HTML, list internal links per page. Flag orphan pages (in the sitemap, linked from nowhere), broken internal links, and important pages more than two clicks from the homepage.
5. Note the markup flags. From the HTML alone: missing canonical tags, missing structured data on pages that plainly want it, images with no dimensions or lazy loading, oversized inline scripts. These are flags, not measurements — you are not running a performance test and must not imply you did.
6. Prioritize. Sort every finding into fix this week (broken links, missing titles, accidental noindex), fix this month (weak metas, heading structure), and backlog (markup niceties). Severity comes from what you observed, never from a general rule about what usually matters.
7. Write the report. Save to seo-audit-<slug>.md at the top of the working folder, starting with Updated: YYYY-MM-DD. Summary table first, then per-page findings, then the prioritized fix list. Report the path back. Then keep the standard files. Add every "fix this week" item to TODOS.md as - [ ] <fix> — <URL> (seo-audit <slug>); the month and backlog items stay in the report.

Standard files. This folder is kept in files any agent already reads. Update them in place; never scatter output into new folders.

• AGENTS.md — create it if missing. localstack owns only the block between <!-- localstack:start --> and <!-- localstack:end -->; rewrite that block, never anything outside it. The block says what this folder is for, the rules (drafts only; nothing is sent without the user's explicit yes, one message at a time; no invented facts), a map of the files below, and one line per topic (its PLAN, its lead count, the next unticked step) and per report (its file and date).
• CHANGELOG.md — create it if missing (# Changelog). Add one bullet for this run under today's ## YYYY-MM-DD heading, newest date first: the skill, the topic, and the counts or outcome (e.g. - lead-search austin-dentists: 18 found, 3 skipped as already contacted).
• TODOS.md — create it if missing (# TODOs). Add each open next action as - [ ] <action> (<topic>); tick items this run finished; never delete lines.
• DESIGN.md — decisions meant to last (positioning, tone, channels to use or avoid). Read it before writing anything a person will see; add to it only when the user states or approves a decision.

Quality bar

• Every finding names the exact URL and quotes the observed value — the actual title text and its length, not "the title is too long".
• The report states how many pages were fetched, how many failed, and how the page list was built. A 30-page crawl is never described as the site.
• No scores, no estimated traffic, no difficulty or authority numbers. You have no tool that measures those. Where one is needed, name the check you could not run.
• Every fix says what to change and to what, specifically enough for a non-specialist to make the edit.
• Nothing about rankings. You cannot see them from a page fetch, and a page-level guess dressed as a ranking cause is the most misleading thing this skill could produce.

The output contract

These apply to the file you write, not just the reply you send.

• Never state a score, rating or grade you did not measure. Report the values you actually observed instead — they are more useful and they are true.
• Open with a summary table — a row per page, item or finding — so the shape of the result is readable before the detail.