Five skills, one discipline. Your accounts, transcripts and deal notes go in. Prep sheets, call scores, deal diagnoses, prospecting plans and business cases come out, all built around the one question that decides whether a deal is real: what is forcing this buyer to act?
A follow-up that summarizes the call the buyer just sat through. A business case with numbers nobody said out loud. The model is not the problem. Nobody told it what a real deal looks like.
Every skill in the pack checks first whether something is making the status quo unacceptable to the buyer. If nothing is, it says so. A deal with no forcing function is a conversation, whatever the CRM stage says.
A buyer number is quoted with the speaker and the date. A projection cites a range you verified. Anything else ships as [VERIFY] with the exact question that closes it. A blank is a finding. A guess is a liability.
Pipeline Build ranks the list before a word of copy gets written.
Discovery Prep walks in with a hypothesis. Discovery Score grades whether you tested it.
Deal Review rates all eight fields and maps five tiers of the committee.
Business Case ships only figures someone said or you can source.
Each skill takes your real inputs and refuses to fill a blank with a guess. Open any card to see a full sample output on a fictional account.
Three pains per account, cost of inaction with the math shown, a committee map with every contact labeled Verified or Unverified, and the Day 1 email.
The forcing function to test, a three-level pain map for the person in the chair, every question in Waterfall and Pivot form, and a timed call flow.
A 100-point score with a quote behind every mark, the one habit costing you the most, the exact words for each pivotal moment, and the recap the buyer forwards to their boss.
Forcing function confirmed, assumed or absent. Eight MEDDPICC fields with evidence. Five committee tiers. The stage it belongs in. Plays with owners and dates.
A one-page Economic Impact Statement where every number is quoted from the buyer or cited from your verified ranges, cost of inaction as a monthly figure, and every missing input listed with the question that closes it.
At 06:12 you asked about budget, then added "or is that still being figured out" before she could answer. She took the out. Same shape at 14:40 and 19:05. Each vague answer came from the escape hatch you built into the question.
One pattern per call, not twelve. The earliest one wins, because it compounds.
"If the audit lands in Q1 with this backlog, it's my name on it." VP Claims Operations, Oct 14.
The CFO has never attended, so Economic Buyer can't rate above Moderate, whatever the champion says. CRM says stage 4. The evidence says stage 2.
"Industry benchmarks suggest a 30 percent backlog reduction is achievable, which would save roughly $500K annually."
No speaker. No source. No arithmetic. The word "suggest" doing the work of a citation. This is the sentence that loses the deal in the CFO's office.
Eighteen months building the AI operating system for a healthcare technology sales org from scratch. Adopted across five departments without a mandate. A diagnostic built with the VP of Sales Ops surfaced $13.3M in stalled pipeline.
Then a second build at a workforce automation company, where the same discipline had to work for a different product, buyer and vertical. This pack is the parts that survived the move.
None of it was built by someone watching sales from the outside. All of it was built by the person running the calls.
ff-context.md. What you sell, who buys it, the numbers you are allowed to say, the words you never say. Every skill reads it first. This is the step people skip and then wonder why the output is generic.
Claude: add it to a Project or install the five skills. ChatGPT: a Project or a custom GPT. Anything else: paste the skill and the context file at the top of the chat. No logins, no platform.
"Prep me for my call with Dana at Northgate." "Score this call." "Review the Northgate deal." "Build my plan for this list." "Build the business case."
Yes. The skills are plain text files. Add them to a Project or a custom GPT. They were written and tested in Claude first, and the Claude install is the smoothest.
No. The skills run it and show their work. Most people learn it by reading the outputs for two weeks.
No. That is the point. Anything not quoted from your buyer or cited from your context file ships as [VERIFY] with the exact question that closes it.
It uses whatever is connected in your AI tool. Where nothing is, it labels contacts Unverified and tells you. It never pretends to have looked something up.
No. It assumes you can run a call. It makes sure the call, the follow-up, the review and the case all point at the same question.
Edit one file. Every skill reads it at the start of every run.