Tariff Exposure — Full System Test
I loaded all four skills, pulled live tariff data, invented a skeptical CFO, and had ChatGPT tear the whole thing apart. The counterfactual is the proof.
This was the test: could the skills system actually change what gets built, or was it just documentation that made me feel organized?
The scenario: a PE portfolio company ($200M revenue, 400 employees) needs tariff exposure analysis after the February 2026 SCOTUS ruling invalidated IEEPA tariffs. The client’s real question: "Which suppliers do we renegotiate first, which products do we reprice, and what’s the 12-month cost impact if we do nothing?"
I loaded four skills before starting: Stakeholder Validation (how a senior analyst validates data), AI Adoption (five failure modes), Testing AI Output (proof chains and oracle validation), and Cross-Model Review (using ChatGPT as adversarial reviewer).
Without skills, here’s what I would have built: a heatmap of tariff exposure by country of origin, a 40-page methodology report, and a final presentation at a big reveal meeting. That’s what a naive consulting engagement produces. The CFO would say "we already know China is expensive" and the $200K engagement would be a PowerPoint.
With skills, here’s what actually got built: a three-tab Excel workbook. Tab 1: suppliers ranked by tariff cost with renegotiation priority and estimated savings. Tab 2: product lines ranked by margin erosion. Tab 3: the 14 suppliers missing country-of-origin data that the client needs to fix in their ERP. No report. No methodology section. The deliverable IS the action plan.
The validation was layered. The stakeholder skill demanded divide-and-reconcile at every step (412 suppliers in, 412 accounted for). The testing skill defined forbidden strings ("consider," "potentially," "it may be worth exploring" — words that mean nothing in a renegotiation playbook). The cross-model review caught 34 products where Claude and ChatGPT disagreed on HS code classification — those got flagged for the customs broker instead of guessed.
I invented a skeptical CFO persona ("David Chen") to stress-test the output. His objections were the ones I’ve heard in real engagements: "We already know China is expensive." "Your HS codes don’t match our broker’s." "Last quarter’s consulting engagement gave us a 40-page report and no action items."
The skills didn’t make the output prettier. They changed what the output was. That’s the difference between a tool and a framework.