Today's Agent Skill: Expense Report Reconciler

What It Does

Every month you dig through a credit card statement, a folder of blurry receipt photos, and a half-finished spreadsheet trying to make three numbers agree. The mismatches are almost always small — a tip added after the fact, a duplicate charge, a receipt you never got — but finding them eats an afternoon. This skill does the matching pass for you and hands back only the exceptions that need a human.

How It Works

The agent ingests a transaction export and a set of receipt records, then matches them on amount, date proximity, and merchant name similarity rather than requiring exact string equality. Anything that matches cleanly gets categorized and dropped into the reconciled bucket; anything that doesn't gets flagged with a specific reason — missing receipt, amount mismatch, possible duplicate, or unmatched receipt with no corresponding charge. The output is a reconciliation table plus a short exception list you can act on in minutes.

How to Deploy It

Save the SKILL.md block below to your agent's skills directory (for Claude Code, `~/.claude/skills/expense-reconciler/SKILL.md`) and restart the session so the trigger phrases register. Then drop in a CSV export from your card and a folder of receipts, say "reconcile my expenses," and the agent will pick it up from there.

SKILL.md — Ready to Deploy

## Description

Reconciles a set of card or bank transactions against a set of receipts, producing a matched ledger, a categorized expense summary, and an exception list of items requiring human review. Designed for monthly expense report prep, corporate card reconciliation, and pre-audit cleanup.

## Trigger

Activates on: `reconcile my expenses`, `expense report`, `match receipts to charges`, `reconcile the card statement`, `find missing receipts`, `expense reconciliation`, `close out expenses`. Also activates when the user supplies a transaction export alongside receipt files or receipt text without naming the task explicitly.

## Input

- **Transactions** — CSV, OFX, or pasted table. Required columns: date, amount, merchant/description. Optional: card last-4, posted vs. transaction date.
- **Receipts** — image files, PDFs, pasted text, or a folder path. Extract date, amount, merchant, and payment method from each.
- **Policy context** (optional) — per-diem caps, disallowed categories, required-receipt threshold.

## Steps

1. Normalize both sets: parse dates to ISO, strip currency symbols, uppercase and de-noise merchant strings (drop store numbers, city suffixes, `SQ *`, `TST*

Copy the full SKILL.md and drop it into your agent's skills directory to activate this skill.

← Back to All Posts | Home — WI AI Tools