Expense Reconciliation on Autopilot

/loop 1h Start the "Expense Reconciliation on Autopilot" loop.

Inspired by Ramp (https://ramp.com).

Goal: every expense coded, matched to receipts, and exceptions triaged with audit trail
Max iterations: 24
Between iterations run: Report uncoded expenses, unmatched receipts, approval backlog, and GL sync errors
Exit when: zero uncoded expenses, all material items approved or escalated, and GL export current

## Before you start
- Connect Stripe (optional) — Corporate card and billing events
- Connect Gmail (required) — Receipt capture from email
- Connect Google Sheets (required) — Exception inbox and coding audit
- Connect Slack (required) — Notify finance on material exceptions
- Attach Loop runner (required) — Self-pace iterations and run checks between passes.
- Attach Human approvals (required) — Queue external sends and high-impact changes for sign-off.
- Attach Operations brief (required) — Summarize queue state and blockers each pass.
- ERP / GL (write via api) — Ledger sync with approval
- Ramp (read via browser) — Spend policies and card limits

## Steps
1. Ingest spend: Pull card transactions, receipts, and employee submissions. [tools: Ramp, Gmail, Stripe]
2. Code expenses: AI categorizes spend against policy and GL codes. [tools: Google Sheets]
3. Match receipts: Link receipts to transactions; flag unmatched items. [tools: Gmail, Google Sheets]
4. Approval queue: Route material or ambiguous items for human sign-off. [tools: Slack]
5. Sync ledger: Export approved entries to ERP/GL with audit trail. [tools: ERP / GL]

Self-pace this loop. After each iteration, run the check command, read the output, and only continue if the exit condition is not met. Stop when the exit condition passes or max iterations is reached. Give a short status update each pass.

Guardrails:
- Never post GL entries without explicit human approval
- Do not weaken coding rules to hit exit condition
- Use test/sandbox cards only when validating new flows