SEO ideas inside Fimo's agents and routines

Each routine checks the site with scripts first and calls the model only when there is something to fix or ask, so a quiet day costs almost nothing. Fixes wait on the Branch until someone publishes. Blue routines ship today. Orange ones are in the catalog as "Needs data".

"Each run starts from zero" in the first recap was wrong. Every run loads its notes, and what one routine learns, the others can read. A person's answer is saved so the routine never asks twice. Exact data, such as a crawl snapshot, goes to a data file that a script saves and reloads without the model reading it.

A routine can already wake on a site event (an edit, a publish, a merge) when someone writes the trigger in plain words. The default templates still use only schedules: FIMO-2809 adds stored event rules to templates. The catalog also lists three "when we notice" triggers that nothing fires yet.

Each arrow goes from a data source to the routines or triggers it unblocks. Search Console is free for the site owner but needs a Google login and a new integration. The data tools cost credits per call, charged to the run.

The fimo-agents skill says a routine must fix or decide something, and "most runs should end with no message at all". So "track 25 AI questions" or "watch competitors" cannot be routines that only send charts. Each one must end in a fix (auto), a decision (suggest), or silence.

On its first run, the routine proposes about 25 questions a buyer asks (one decision for the person). It saves them in its NOTES.md. Each week a script asks ChatGPT, Perplexity and Gemini through the data tools and saves the answers as a data file. The model looks only at what changed since last week.

Today's catalog entry asks the person for "up to five competitor addresses" and sends "a weekly summary". Both go against the vision. Instead, competitors are found from Google's top 10 for our searches (Ahrefs via the data tools), and a change fires the a competitor changes trigger. The person only sees a decision from Refresh, Writing or Citation.

The SEO template already says: "Once Search Console is connected, you also work the searches each page almost ranks for." The routine takes pages in positions 8–20, compares them with the top 3, and proposes the missing section. A title-only rewrite for a page with few clicks is the Meta tags routine's job, on the traffic trigger.

Refresh finds past years, expired offers and retired product names. Add one project memory topic file, facts.md, with prices, numbers and claims. Then Refresh can also find "$29" on a page when the fact is "$39". Prices are team-owned, so it is always suggest with the exact words.

This is the vision's "Later" step: acting on traffic, not only on what the page says. It needs Search Console numbers. The routine that made a change writes the baseline in its NOTES.md and judges it after four weeks. A win becomes a lesson in the agent's MEMORY.md, so the agent asks less over time.

Each box is one routine change. The effort label counts the files the fimo-agents skill says must change together (template, skill, catalog, docs, E2E), plus any new integration. If we skip step 1, the team waits weeks for Search Console before any customer sees a new result.

The skill asks for an estimate before adding a routine to a default agent. AI visibility is weekly, script-first, and calls the model only when an answer changes. The data calls are the fixed cost. Free projects could get a smaller question list or none.

Each item gives our recommendation. A different answer changes the scope of the first step: for example, choosing Profound adds a paid integration before AI visibility can ship. So we should decide these before anyone starts building.