See where the money goes by category

# Spend by expense category

## Why this skill exists

`spendesk_analyze_spend` has **no "category" dimension** in `groupBy`. Expense categories come out under `analyticalCode`, mixed with the values of custom analytical fields (region, project…). Unsorted, the table is wrong. This skill isolates the categories.

## Step 1 — Get the categories

```json
{ "tool": "get_expense_categories", "params": {} }
```

Returns `id`, `name`, `isArchived`. Keep the list of `id`s: it filters step 2.

- Archived categories are included: keep them for history, and flag them.
- These are employee-facing labels, **not accounting accounts**: do not present them as GL codes (use `get_chart_of_accounts` for that).
- `costCenters: []` means the category applies to every cost center, not to none.

## Step 2 — Analyse and keep only the category rows

```json
{
  "tool": "spendesk_analyze_spend",
  "params": {
    "fromDate": "YYYY-MM-DD",
    "toDate": "YYYY-MM-DD",
    "groupBy": ["analyticalCode", "month"],
    "limit": 50
  }
}
```

In the response, **keep only the rows whose `id` is one of the category IDs** from step 1. The other rows are values of other analytical fields: ignore them.

- `analyticalCode` first in `groupBy`, `limit` at 50 so that no category falls into "Others".
- State the exact dates used. Maximum period: 366 days.
- Do not reuse `share`: it is computed across all rows, other fields included. Recompute each share on `summary.totalAmount`.

## Step 3 — Consistency check

Add up the category rows and compare with `summary.totalAmount`:

- **Equal**: all spend has a category. Verified live over 12 months: IT & Technical fees EUR 87,235.36 + Announcements & Ads EUR 25,984.90 + Travel & Expense EUR 500.00 = EUR 113,720.26, exactly the total.
- **Lower**: the gap is spend **with no category**. Show it on an "Uncategorised" row.
- **Higher**: this should not happen, since each row carries only its own category's amounts. Check that no non-category row was kept; if the gap remains, say so rather than forcing the total.

## Step 4 — Read the trend

The `month` level under each category gives the monthly series. To spot what is accelerating:

- compare the latest month with the average of the earlier months of the same category;
- flag a category concentrated in a single month (on the demo account, 94% of "IT & Technical fees" fell in July): that is a one-off, not a trend.

`compareWithPreviousPeriod` compares with the window of equal length just before, not with the previous calendar month or quarter: read `summary.previousPeriod.period` and show it as returned.

## Recommended output format

```
Spend by category — from <date> to <date>
Total: X EUR — categorised: Y EUR — uncategorised: Z EUR

| Category | Amount | Share of total | No. of expenses | Highest month | Trend |
```

Amounts in functional currency, including VAT, on the document date. Credit notes are included and netted off. The MCP does not read budgets: for a variance against budget by category, the user supplies the figures.

## Example user prompts

- "Where does the money go by expense category over the last 12 months?"
- "Which expense categories grew the most this quarter?"
- "What share of our spend has no category?"