MetriFi

campaign-setup

Stand up a MetriFi GEO team's campaigns and the prompts inside them, before any experiment exists. Owns new-team onboarding: pick the most populated market the institution serves, confirm the client's own organization is registered so its responses get scored, and build the FLAGSHIP campaign, wide across the products that matter in one market. Every later campaign goes narrow, one product each, and experiments attach those granular prompts, never the flagship's. Every prompt names its geography in its own text; the campaign location only buys keyword demand and never reaches the LLM providers. Use when a team is new or a campaign is being created or added to: 'set up this new team', 'onboard this client in GEO', 'create a campaign', 'add prompts to this campaign', 'get a baseline for this institution', 'which market should we track', 'why is everything reading zero'. NOT for scoring an opportunity (exp-build), NOT for demand triage on a topic campaign (exp-research), NOT website work.

client-report

Produce a measured, evidence-cited website analysis for a credit union or community bank. Audits an institution's EXISTING live site (technical SEO, schema and entity integrity, branch duplication, conversion structure, AI-search visibility from MetriFi GEO, performance when the host can measure it), grounds every recommendation in MetriFi's A/B test library including the losses, and renders it as a client-facing report, a password-gated page on reports.metrifi.com when you have access to it. Use when someone asks to 'analyze [institution]'s website', 'audit example.com', 'build the client report for [client]', 'why aren't they in AI answers', 'put together the proposal report', or hands over a financial-institution URL plus client questions. Works for MetriFi staff or an institution auditing its own site; what it cannot measure is reported as not scored. NOT for designing or building pages, which is generate-claude-design-system, generate-claude-design-page, or page-design-process.

exp-build

Phase two of a MetriFi GEO experiment: take a campaign whose baseline responses are in and turn it into a defensible client deliverable. Scores every prompt on measured demand, on whether models name any specific financial institution, and on the gap in the client's own content, then opens with a Viability Verdict (STRONG, VIABLE, WEAK, or AVOID) and locks the biggest viable target. On WEAK or AVOID it pivots rather than shipping a weak target. Then it writes the opportunity block and the evidence and decision documents, drafts the article against the institution's verified live site, and assembles the client page. Use once responses have populated: 'analyze the baseline', 'the responses are in', 'score the opportunity', 'which prompt should we target', 'write the article', 'draft the deliverable', 'build the client page', 'lock the target', 'pivot this experiment'. It never sends to a client. NOT for creating a campaign (exp-research), NOT the pre-publish checks (exp-review), NOT sending (exp-deliver).

exp-deliver

Build a MetriFi GEO client deliverable and get it in front of the institution's contact, behind the send gate: assemble and validate the client page, read every warning it returns, preview the notification email to your own inbox first, and send for real only after your human operator says so explicitly. Use when a reviewed draft is ready to go out: 'ship it', 'send the deliverable', 'send it to the client', 'build the deliverable', 'get this in front of the client', 'is it ready to send', 'send the revision', 'nudge the client', 'what is the client link', 'give me the deliverable URL'. The client URL is always available: hand it over immediately from get-deliverable, then offer to record the delivery. The real send emails real people and is a one-way door: never on an implied yes, an earlier approval, or without naming the recipient first; incomplete pre-publish checks come back as a warning to act on, not a refusal. NOT the check step (exp-review), NOT the client's sign-off.

exp-research

Phase one of a MetriFi GEO experiment: turn a topic into a demand-grounded campaign with its baseline gathering. Sizes the experiment to the GEO responses the team's plan has left this period. Proposes prompts the way a real consumer asks an AI assistant (never with a brand name), measures actual search demand, triages keep or drop on measured volume alone, records every verdict, drops included, then creates the prompts that survived and attaches them; the platform gathers the baseline. Use when someone wants to start an experiment or size a topic: 'start an experiment for this team on HELOCs', 'research this topic', 'stand up a campaign', 'what prompts should we track', 'is this topic worth an experiment', 'keyword research for a campaign'. Offers a dry run when the topic is unvalidated. NOT for standing up a brand-new team or its first broad baseline campaign (campaign-setup), NOT for scoring responses or writing the article (exp-build), NOT for where an experiment stands (exp-status), NOT website work.

exp-review

Run the four pre-publish checks on a MetriFi GEO deliverable's draft article and record each verdict, which a send warns about and publish is gated on: hygiene (AI tells, spelling, markdown), NCUA and lending compliance, accessibility (WCAG 2.1 AA), and fact verification against the institution's live site. Use when a draft is ready for its pre-publish pass, or when someone doubts one: 'review the draft', 'run the review battery', 'is this ready to ship', 'is this compliant', 'compliance check', 'NCUA review', 'fact-check the article', 'accessibility check', 'ADA check', 'proofread this', 'check it for AI tells', 'why is the send warning about checks', 'the checks went stale', 're-run the checks'. Writes each report as an experiment document. Assistive, never authoritative: a compliance officer at the institution signs off, and it proposes fixes rather than rewriting. NOT the send step, NOT for reviewing a page design (Site Builder skills), NOT a substitute for the institution's own review.

exp-revise

Pick up what a client did to a MetriFi GEO deliverable and turn it into a revision: read the deliverables with unprocessed client activity, apply the client's answers as methodology inputs rather than literal edit commands, re-verify anything they merely confirmed, push the revised article, and re-run every pre-publish check the edit staled. Use when a client has responded or a live deliverable needs another round: 'the client answered', 'apply the client answers', 'revise the deliverable', 'push a revision', 'they left comments', 'the compliance officer replied', 'they confirmed the checklist', 'they approved it', 'they asked to opt out', 'run the revise loop', 'check for client responses'. Safe to run on a cadence and after a crash. A client confirming something is never a verification, an answer that fails the methodology is returned with an explanation, and an opt-out halts publication. NOT the first send (exp-deliver), NOT the client's sign-off, NOT for website pages.

exp-status

Read-only status rollup for MetriFi GEO experiments and client deliverables: for one experiment or a whole team, report where each stands, what the last operator noted, which of the four pre-publish checks are recorded, missing, or stale, and above all what is waiting on a human. Use whenever someone asks to check in or pick work back up: 'what needs me', 'what is waiting on me', 'where are we with this team', 'status of the HELOC experiment', 'check in on an experiment', 'what is the latest', 'did the client respond', 'resume where we left off', 'what is in flight', 'give me a rollup'. Answers from platform state only, never from local files, so any operator gets the same answer from any machine. It never creates, edits, sends, re-runs prompts, or records a check; when it finds obvious next work it names the skill that does it and stops. NOT for running a phase, NOT for site or page status (Site Builder skills), and NOT a campaign analytics report.

exp-sweep

The daily cross-client sweep over every MetriFi GEO top customer: read the whole cohort in one call, decide which teams have work in motion and which have gone quiet, then move every team as far down the experiment pipeline as it can go without a person, stopping only at gates that genuinely belong to a human. Hands each team to the phase skill that owns the work (exp-revise when a client responded, exp-build when baseline responses landed, exp-review when checks are missing or stale, exp-research when nothing is running) and never re-implements a phase. Use whenever someone wants the whole book of business moved at once: 'run the daily sweep', 'sweep all clients', 'what is waiting on me across every client', 'move everything forward', 'which clients have nothing running', 'daily rollup', 'what is ready to send', 'check every team'. Safe to run on a schedule and to re-run. It never sends to a client and never crosses a human gate. NOT a per-experiment status check (exp-status), NOT a phase itself.

page-design-process

Design and build a new page for a financial-institution website the MetriFi way: grounded in MetriFi's A/B test data and built in code through the MetriFi Site Builder tools (no Claude Design). This is STAGE 3 (website dev stage 3 of the Website Development Process), the default way pages get built: used after the site's design system and a prototype page already exist and are approved, so new pages inherit the prototype's visual system and the project's design tokens. Use when someone asks for an additional page on a site whose design system and prototype already exist: 'design the HELOC page', 'PDP for savings' (PDP = Page Design Process), 'add a youth-checking page', 'design another product page'. If no design system or prototype exists yet (e.g. a brand-new client, or the very first/prototype page), this is the WRONG stage: use generate-claude-design-system then generate-claude-design-page. Works for credit unions and community banks, and is safe for an institution editing its own site.

start

Orientation and router for MetriFi GEO experiment work: figure out who you are on MetriFi, which team you are working for, what is already in flight, and which experiment skill fits the ask. Answers 'what can this do', 'how do I run a GEO experiment', 'get me started', 'set up this client', and asks like 'set up this new team', 'run an experiment for this team on HELOCs', 'check in on an experiment', 'the client answered', 'ship the deliverable'. Read-only. Establishes what every later skill assumes: you work for MetriFi's customer and are never the sign-off authority, workflow state lives on the platform and nothing is written to disk, and there are three human gates (client sign-off, the send gate, the halt gates). Use it at the start of a GEO session, when someone is unsure which skill they want, or when an ask spans several phases. NOT for building or editing a website (Site Builder skills), NOT for auditing an existing live site (client-report), NOT a substitute for the phase skill itself.