Spend Policy Cabinet

A Research agent answers "can I expense this?" questions instantly by reasoning over the live policy doc, flags unusual spend each week, and summarizes department expenses. A QA agent keeps the policy cards accurate and catches edge-case answers before they go out. Fewer Slack questions. Faster reimbursements. No surprises at month-end.

Keep Ramp, Brex, or Expensify as your card platform. Replace the policy PDF nobody reads, the Slack thread with six different answers, and the manual anomaly hunt at close.

The team

• [[.agents/research]] — Policy Research Agent. Answers employee spend questions by reasoning over data/policy.md; flags anomalous transactions from the weekly card export; summarizes department spend by category. The employee-facing member of this cabinet.
• [[.agents/qa]] — Policy QA Agent. Reviews the Research agent's answers for accuracy, checks flagged transactions against policy thresholds, and keeps data/policy.md current when Finance updates limits or categories.

Recurring rhythm

Cadence	Job	Owner	Output
Weekly (Mon 08:00)	[[.jobs/weekly-anomaly-flag]]	Research Agent	Anomaly flag in .cabinet-state/
Weekly (Mon 08:00)	[[.jobs/weekly-policy-qa-digest]]	QA Agent	Q&A digest in .cabinet-state/

How to run the demo

1. Open [[hub]] — the spend policy hub: policy cards by category with limits, example Q&A pairs, flagged transactions from the last sweep, and department spend summary.
2. Browse data/policy.md — the structured spend policy the Research agent reasons over.
3. Check .cabinet-state/anomaly-flag.md — this week's flagged transactions with policy section reference and recommended action.

Connectors

Required: Ramp, Brex, or Expensify (card transaction export), expense policy doc (current version in Drive or Notion). Recommended: Slack (policy Q&A delivery, anomaly alerts), Google Sheets (department manager spend summaries), HR system (employee cost-center mapping).