One skill per stage of the GROUND Method. Install them in Claude, ChatGPT, or Copilot this morning and walk one AI use case from "we want to use AI for X" to a brief your security, legal, or procurement reviewer can approve, without a consultant in the room.
.claude/skills/ folder. Each skill triggers by name or by what you ask for.Every skill labels its output Verified, Unverified, or Original, so a reviewer can tell what you measured from what the assistant concluded. They are working aids, not legal, security, or compliance advice.
Turn a rambling AI idea into one plain sentence a security, legal, or budget reviewer will accept, plus the accountable-owner question. Use when someone says 'we want to use AI for X' and nobody can state X in one line.
use-case-one-liner/SKILL.md
A 20-minute interview script for the person who does the job today, built to surface the exceptions, workarounds, and side spreadsheets that the official process hides. Use before designing any automation.
watch-the-work-interview/SKILL.md
Paste the reports, spreadsheets, and systems a team uses for one workflow; this skill asks the questions that reveal which one is the real source of truth and where two of them disagree. Use before pointing any AI at data.
system-of-record-finder/SKILL.md
Score each data source an AI will use on owner, refresh rate, reconciliation, and audit trail, and output the table a reviewer will ask for. Use when the question is 'is our data ready for AI.'
data-freshness-scorecard/SKILL.md
Walk a team through finding the private spreadsheets, personal notes, and side tools that feed a workflow but appear in no system diagram. Use before automating anything, because the shadow system is where the real process lives.
shadow-system-hunter/SKILL.md
Paste up to ten AI ideas; this skill sorts each by autonomy risk and payoff and names the three that are pilot-safe. Use when a team has a long wish list and needs to choose.
decision-or-task-sorter/SKILL.md
Draft the human review step for one AI use case: who reviews, the checklist they use, what happens when the output is wrong, and how it is switched off. Use when a reviewer asks 'who checks this.'
human-review-gate-designer/SKILL.md
The mirror test. The owner explains the AI use case in their own words; this skill grades the explanation against the material and names every gap, without praise. Use before any review meeting.
explain-it-back/SKILL.md
Shrink any AI project to one step of one workflow with a 30-day stop rule and one success condition. Use when a proposal is too big to approve and too vague to measure.
one-clean-move/SKILL.md
The questions to ask any AI vendor before a pilot, grouped by what a reviewer will want answered, with the answers that should worry you. Use before a demo, a trial, or a contract.
ai-vendor-question-pack/SKILL.md
Take one metric and its baseline and produce the measurement calendar, the fixed review windows, and the proof report template. Use before an AI pilot starts, never after.
30-60-90-proof-plan/SKILL.md
Paste notes, checklist answers, or outputs from the other GROUND skills; this skill writes the one-page Use-Case Readiness Brief a security, legal, or procurement reviewer reads in two minutes. Use when it is time to ask for approval.
reviewer-brief-writer/SKILL.md
Run them in order. Each skill ends by naming the next one. Gather gives you a one-liner and the real process. Reveal tells you whether the data has an owner and a source of truth. Order finds the shadow systems and sorts decisions from tasks. Unify designs the human review gate and tests whether the owner can explain it back. Narrow cuts the pilot to one clean move and arms you for the vendor call. Deploy and Demonstrate fixes the proof windows and writes the brief. Then run the same answers through the AI Readiness Checklist for a scored result with a data-risk tier.
A skill is a plain-text instruction file that tells an AI assistant how to do one job well: what to ask you, what steps to follow, what to output, and what it must never invent. You paste it in or add it to your assistant once. No code, no setup beyond that.
Claude (the app and Claude Code) reads the SKILL.md format directly. In ChatGPT, paste a skill's text into a Project's instructions. In Copilot or any other assistant, paste it at the start of a chat. The README in the download has the steps for each.
Because most AI projects fail at a stage nobody ran: the data had no owner, the side spreadsheet was never found, or there was no human review step. Running the twelve in order walks one use case through the whole method, and each skill ends by naming the next one.
The checklist scores one use case and gives you a data-risk tier and a brief in twelve minutes. The skills pack is the working toolkit behind each of those answers. Use the checklist to find out where you stand and the skills to do the work the checklist says is missing.
A team that has run the GROUND Method on its own data knows exactly what it still needs verified. Some of them ask BBI to verify it. That is the AI Pre-Flight, the paid diagnostic, and it is the only thing these pages sell.
No. The skills run inside your own AI assistant under your own account. BBI never sees your inputs or outputs. Check your assistant's data terms for what it retains; the AI Vendor Question Pack skill tells you what to ask.
The pack and the checklist are self-assessments. The AI Pre-Flight is BBI's fixed-fee diagnostic: a readiness scan of your data, workflow, and risk with a 0 to 100 score you can defend and a remediation plan in GROUND order.