Today's Agent Skill: Auto-Drafting Expense Reports From Receipts
What It Does
Nobody got into their job to hunt down a week's worth of crumpled receipts and match them to credit card lines by hand. Expense reports are tedious, error-prone, and always due the day after you stopped tracking anything. An agent that reads receipt images and drafts the report for you turns a 45-minute chore into a two-minute review.
How It Works
The skill takes a folder or batch of receipt images/PDFs plus a CSV export of card transactions, extracts vendor, date, amount, and category from each receipt via OCR-style parsing, then matches each one to its corresponding transaction line. It outputs a filled expense report table with any unmatched or ambiguous items flagged for the human to resolve.
How to Deploy It
Drop the SKILL.md into your agent's skills directory (e.g., `~/.claude/skills/expense-report-drafting/`) alongside a `SKILL.md` file — any agent that reads SKILL.md-formatted skill definitions will pick it up automatically on next load. Point the agent at a receipts folder and a transaction export, and it runs the extraction-and-match pipeline without further setup.
SKILL.md — Ready to Deploy
## Description
Automatically drafts a complete expense report by extracting data from receipt images/PDFs and reconciling each receipt against a bank or credit card transaction export. Reduces manual expense reporting to a single review-and-approve step.
## Trigger
Activates when the user provides a batch of receipts (images, scans, or PDFs) along with a transaction export, or asks to "build my expense report," "reconcile these receipts," or "draft this month's expenses."
## Input
- A folder or list of receipt files (JPG, PNG, PDF)
- A transaction export (CSV or similar) covering the relevant date range
- Optional: expense category list or policy rules (e.g., meal caps, mileage rate)
## Steps
1. Parse each receipt to extract vendor name, transaction date, total amount, and line-item detail where available.
2. Normalize vendor names and dates to match the format used in the transaction export.
3. Match each parsed receipt to the closest transaction by amount and date proximity (±3 days tolerance).
4. Assign an expense category to each matched pair using either the provided category list or common-sense defaults (meals, travel, supplies, software).
5. Flag any receipt with no matcCopy the full SKILL.md and drop it into your agent's skills directory to activate this skill.