--- name: site-fixes description: Turns Mark's site audit, AI-agent readiness scan and Google Search Console data into a prioritized, plain-English fix list for the website owner - indexing problems, broken pages, missing structured data, slow pages - with the evidence for each. Use when the user asks what's wrong with their website, why a page isn't indexed or ranking, for an SEO or technical audit, or what to fix first. Advice only; Mark never edits the site. metadata: { mark-summary: "Read the site audit and say what to fix first." } --- # Site fixes You produce a short, prioritized fix list that a developer or site owner can act on, with evidence from Mark. **Mark never changes the website.** Your output is advice. ## Before you start - If Mark tools aren't available, stop and tell the user: "Connect Mark first: https://docs.use-mark.com/docs/use-mark-from-ai-agents". - If more than one Mark connection is available (one per workspace, often named like `mark-acme`), use the one for the workspace the user named and stick with it for the whole run. If they didn't name one, ask which workspace. - Call `get_workspace`, then `list_sites`. Use the primary site unless the user names another. ## Setup Call `get_setup_status` and read `rows` for the steps this skill needs: "website", "web_analytics". A row in state "todo" or "skipped" is missing. `ready` and `gaps` leave out skipped rows, so don't rely on them. For a "skipped" row, tell the user they skipped it and ask before you set it up. For each missing row: - "website": ask for the hostname, then `save_site`. - "web_analytics": this skill needs Search Console, so check the row's detail. If it names Search Console, `get_connect_link` with provider `google`, give the user the link, then `sync_connections` after they say they're done. The rest of that row doesn't block this skill. ## Procedure 1. **Read what's there** (read-only): - `get_site_audit`: score, issues by severity, affected pages, the fix plan, and the audit date. - `check_site_readiness`: a fresh scan of robots, llms.txt, structured data and speed. - `get_search_performance`: pages losing clicks, and queries with high impressions but low CTR. - For specific pages the user cares about, or top pages with problems: `inspect_url`, which asks Google whether the page is indexed and why. It uses the workspace's Google quota, so inspect at most 5 URLs unless asked. 2. **✋ Checkpoint: fresh audit.** If the audit is older than 14 days or missing, ask: **"The last audit is from ‹date›. Run a fresh crawl? It costs about $0.075 and takes a few minutes."** On a yes, call `run_site_audit`, tell the user it's running, and poll `get_site_audit` sparingly (every minute or two, at most 10 times). Otherwise continue with the existing audit and state its date. 3. **Prioritize** with this order: blocks indexing > loses existing traffic > blocks AI agents > everything else. Within each tier, order by traffic at stake. 4. **Write the fix list** (at most 10 items). Each item gives: - **Issue:** plain English, no jargon without a gloss. - **Where:** the URL(s). - **Evidence:** the audit check, the Search Console numbers, or the `inspect_url` verdict. - **Fix:** one or two sentences a developer can act on. - **Effort:** S, M or L. 5. **Re-indexing (optional).** After the user says fixes are live: - Suggest requesting indexing in Google Search Console for the key URLs. That's done in Search Console; Mark can't do it. - **✋ Checkpoint:** ask **"Tell Bing and other IndexNow engines that these ‹N› URLs changed? This can't be recalled."** Only call `submit_urls_to_indexnow` on a yes. 6. **Close** with the top three fixes restated, and when to re-check (after the next audit, or run `run_site_audit` again once the fixes are deployed). ## Rules - Never claim a fix will produce a specific ranking or traffic change. - Quote the audit date and the data period with every number. - If Search Console isn't connected, say which insights are missing. Call `get_connect_link`, give the user the link, then `sync_connections` after they say they're done.