Next steps on the project page

The card sits on the Overview tab, under Needs you and the project's numbers. A suggestion that Needs you already shows on the same screen (requests waiting on us) is left off the card, so nothing is listed twice. It's hidden while the editor window is open in its full split-screen mode, since the live site is taking up the same space then.

Every suggestion on this card comes from a single, plain function (computeNextSteps, in the codebase's next-steps library) that reads fields the project record already has — whether a goal is on file, whether discovery questions were sent and answered, the real build status, whether it's live, how many feedback items are open, how many open questions the plan has, and a few more — and turns them into a ranked list through fixed rules. There is no model call anywhere in it: the exact same project state always produces the exact same suggestions, and nothing here is ever a guess or an invented idea.

• No goal written yet — "Tell the AI what this project is for," since every build and edit is a guess without one.
• Discovery questions never sent, or on hold — "Send the client discovery questions." Sent but not yet answered — a lower-priority "Waiting on the client's answers" note instead.
• Nothing built yet and nothing in progress — "Start the first build," naming how many kit modules are already picked and ready to build in, if any.
• A build actively running — a status row saying so, pointing at the Overview tab where the live log is.
• Real files exist but the plan has zero kit modules — "Add a quote or contact form," plus one suggestion matched to the project's own kind (a website is pointed at local SEO; a marketing-automation project at marketing automation; an app at booking; an admin system at staff tools).
• Built but not yet live — "Go live on a temporary address," since nothing is public until that happens.
• Live, but still on a temporary aiKit address — "Connect their own domain."
• Live with no access code set yet — "Make the client's feedback link," which opens the Settings tab, where copying the feedback link makes the code.
• Any feedback items waiting on us — a row naming exactly how many, linking straight to the Feedback tab.
• Open questions still sitting in the plan — a row naming how many, linking straight to the Site plan.
• Live, with nothing else currently open — a lower-priority suggestion to ask the AI for a general quality pass (broken links, missing alt text, contrast, phone layout).

Every suggestion carries one of three priorities, each with its own colored dot: Now, Soon, or Later. The full rule set can return up to six ranked suggestions; the card itself only has room for five, and because the list is already sorted by priority before it's trimmed, cutting it to five never drops something more urgent than what's shown.

Depending on the suggestion, its button does one of a handful of real things: open a real page (like the full Site plan), jump to a specific tab on the same project page (like Questions, Feedback, Hosting, or Settings), open the editor window on a specific view (Chat, Feedback, Versions, or Brief & rules) — and for the AI-quality-pass suggestion specifically, the exact prompt is quietly filled into the chat box waiting for you, rather than you having to write it yourself — retry or start the build, or open the Kit browser, sometimes scoped straight to the one relevant category (forms, local SEO, marketing automation, booking, or staff tools, depending on which suggestion it was).

Nothing on this card blocks anything else on the project — it's a suggestion list, not a checklist you have to clear. When the rules genuinely find nothing worth flagging, the card says so plainly ("Nothing pressing. Keep improving the site from the chat.") instead of inventing busywork just to have something to show.