lov-expense-report

expense-report — Invoice → Categorized Excel

Extract invoice information from images or text, classify expenses, and generate a professional Excel report with subtotals per category.

When to Use

• User has invoice photos / scanned receipts to process
• User describes expenses in text and wants them organized
• User needs a categorized reimbursement report (报销单)
• User mentions 发票报销, 报销汇总, 发票整理, expense report

Expense Categories

Category	Keywords / Examples
Business entertainment	Client meals, business banquets, gifts, entertainment
Travel - Transportation	Air tickets, train tickets, high-speed rail, taxi, ride-hailing, ride-hailing, refueling, tolls, parking fees
Travel - Accommodation	Hotel, accommodation, guesthouse
Travel - Meals	Meals during business trips, work meals
Office supplies	Stationery, printing, office consumables, office equipment
Communication fees	Phone charges, data, internet fees, broadband
Other	Expenses not belonging to the above categories

Workflow (MANDATORY)

Step 1: Collect Invoice Data

Accept input in any of these forms:

• Images: Read invoice photos using the Read tool. Extract: date, vendor, amount, item type, notes.
• Text descriptions: Parse the user's text for the same fields.
• Mixed: Multiple images + supplementary text.

For each invoice, extract these fields:

{
  "date": "2026-04-15",
  "vendor": "海底捞(国贸店)",
  "item": "客户餐费",
  "amount": 486.00,
  "category": "业务招待",
  "note": "与XX公司李总晚餐"
}

Step 2: Classify

Assign each invoice to a category from the table above. Rules:

1. If the user explicitly states the category, use it.
2. If the item/vendor clearly matches a category keyword, auto-assign.
3. For ambiguous items (e.g., "餐费" could be 业务招待 or 差旅-餐饮):
   • If the note mentions a client/customer → 业务招待
   • If the context is a business trip → 差旅-餐饮
   • If unclear, ask the user.

Step 3: Confirm with User

Before generating, show the extracted data as a table:

| # | 日期 | 商户 | 项目 | 金额 | 分类 | 备注 |
|---|------|------|------|------|------|------|
| 1 | 2026-04-15 | 海底捞 | 客户餐费 | 486.00 | 业务招待 | 与XX公司李总 |
| 2 | 2026-04-14 | 滴滴出行 | 打车 | 45.50 | 差旅-交通 | 机场→酒店 |

Use AskUserQuestion to ask: "以上信息是否正确? 需要修改或补充吗?"

Step 4: Generate Excel

Write the confirmed data to a temp JSON file, then run:

python expense-report-skill/scripts/generate_report.py \
  --input /tmp/invoices.json \
  --output "发票报销汇总-YYYYMMDD.xlsx"

JSON format (array of objects):

[
  {"date": "2026-04-15", "vendor": "海底捞", "item": "客户餐费", "amount": 486.0, "category": "业务招待", "note": "与XX公司李总"},
  {"date": "2026-04-14", "vendor": "滴滴出行", "item": "打车", "amount": 45.5, "category": "差旅-交通", "note": "机场→酒店"}
]

Step 5: Deliver

Tell the user:

• Output file path
• Total amount and breakdown by category
• Remind them to review the "分类汇总" sheet for the summary

Output Format

The Excel file contains two sheets:

1. 发票报销汇总 — Full detail, grouped by category with subtotals
2. 分类汇总 — Summary table: category, count, subtotal

Style: Skill Publisher warm-academic (terracotta headers #4F46E5, warm cream accents).

Edge Cases

• Blurry/unreadable image: Tell the user which fields couldn't be extracted; ask them to provide manually.
• Foreign currency: Note the currency; convert to CNY if user provides rate, otherwise keep original with note.
• Duplicate invoices: Flag potential duplicates (same date + vendor + amount) before generating.
• No date on invoice: Use the date the user provides, or mark as "日期不详".

Runtime context (shared)

Before running, read this Skill package's skill.yaml, with the host providing the skill-runtime/v1 context. The field resolution order is: current request, project context, personal Preferences, brand Profile, general defaults.

• Use only the fields declared in the Manifest; Profile stores public brand facts, Preferences stores personal work preferences.
• required: true When a field is missing, ask the user one focused question according to the Manifest's question configuration; save the answer only after the user explicitly agrees.
• Errors provide a copyable context_id, field path and source; diagnostic content avoids secrets, full private paths and raw configuration.

General feedback loop

When the user raises modification suggestions during a Skill-driven task, before continuing with the current artifact you must:

1. First determine whether the suggestion is task-specific(this time only) or reusable(reusable across tasks).
2. task-specific Only modify the current task, do not change the Skill.
3. reusable First determine the scope: domain rules first update the corresponding canonical Skill; rules applicable to all Skills first update the shared specification.
4. After completing rule updates, versioning, lint and distribution verification, then apply the modification to the current task.
5. reusable The modification invalidates previous "confirm", "continue", "send it" responses; after completing the current artifact modification and read-back you must stop and wait for the user's next instruction, without automatically proceeding to publishing, submission or other external writes.