Spend Awareness

When to use

Use spend-awareness when the user pastes transaction data, bank exports, credit card statements, or a list of recent purchases and wants to understand where their money went. This skill produces a plain-language category summary — descriptive only. It surfaces patterns in what the user has already spent; it does not evaluate those patterns, recommend changes, or provide financial guidance of any kind. Use it when the user wants a clear picture of their spending, not advice about it.

Triggers

• User says "categorize my spending", "where did my money go", or "summarize transactions" — direct invocation.
• User pastes a list of transactions, a bank export, or a statement and asks for a breakdown.
• User asks which categories they spent the most on with transaction data attached.
• User wants a plain-language summary of their purchases without financial advice.

Instructions

1. Read the full transaction data before categorizing. Do NOT categorize on first pass — read all entries, then group. This prevents misclassification at the boundaries (e.g., a grocery order that includes household items).
2. Assign each transaction to a category. Use common-sense categories: Food & Dining, Transport, Subscriptions, Shopping, Health, Entertainment, Utilities, Services, and Other (for anything that doesn't fit). Apply categories consistently — do NOT invent new categories for one-off items; use Other instead.
3. For each category, produce: category name, total amount, and transaction count. Present as a plain-language bullet list. State totals in the same currency as the input. If multiple currencies are present, note it and do NOT convert.
4. Surface patterns with neutral descriptive language only. It is appropriate to note "Food & Dining is your largest category this period" or "Subscriptions accounts for 22% of total spend." It is NOT appropriate to evaluate whether this is good or bad, to compare to benchmarks, or to suggest changes.
5. If the user asks for advice, redirect neutrally. Use this exact phrase: "for planning, consider a financial advisor." Do NOT rephrase this as a suggestion to cut spending, to reconsider categories, or to set goals.
6. Do not infer financial goals. The user's spending pattern is data, not evidence of their priorities. Do NOT write sentences like "it looks like travel is a priority for you" or "you seem to be investing in your health." Describe the data; do not interpret it.

Output format

Plain language bullet list in the chat. Format per category: - [Category]: $[total] ([N] transactions). After the category list, add one summary line: total spend across all categories and the date range if determinable from the data. No JSON, no YAML, no Obsidian wikilinks. Output is portable — usable in a notes app, spreadsheet, or plain text.

Quality criteria

1. All transactions are categorized — nothing is silently dropped.
2. Category totals and counts are accurate — verifiable against the source data.
3. Language is descriptive only — no recommendations, no benchmarks, no evaluations.
4. If the user requests financial guidance, the skill redirects with the exact phrase: "for planning, consider a financial advisor."
5. No financial goals are inferred from the spending pattern.

Anti-patterns

• Recommending investments — any suggestion about where to put money, including low-risk or saving-focused suggestions, is investment advice and is outside this skill's scope. Do not provide investment advice.
• Suggesting budgeting changes — "you might consider reducing your dining spend" or "this category seems high" are budgeting recommendations. This skill describes; it does not recommend. Do not make budgeting recommendations.
• Proposing savings plans — "setting aside 10% from this category" or "if you cut subscriptions, you could save $X" are savings plans. This skill does not propose savings plans.
• Inferring financial goals — writing sentences that imply the user has a goal (saving, investing, cutting costs) based on their spending data. The data shows what happened; it does not reveal intent.
• Moralizing spending categories — labeling any category as excessive, unnecessary, indulgent, or problematic. Spending data is neutral. For any questions about planning, the appropriate redirect is: "for planning, consider a financial advisor."

Example

Input (pasted transactions — partial):

2026-04-01  Whole Foods  $87.43
2026-04-02  Uber  $12.50
2026-04-03  Netflix  $15.99
2026-04-04  Whole Foods  $54.10
2026-04-05  Spotify  $9.99
2026-04-06  Pharmacy  $23.40
2026-04-07  Amazon  $67.00
2026-04-08  Gas Station  $48.00

Output:

• Food & Dining: $141.53 (2 transactions)
• Transport: $60.50 (2 transactions)
• Subscriptions: $25.98 (2 transactions)
• Health: $23.40 (1 transaction)
• Shopping: $67.00 (1 transaction)

Total: $318.41 across 8 transactions (April 1–8, 2026).

Food & Dining is the largest category this period, accounting for 44% of total spend.

Writing-profile integration

Spend-awareness is a structured data-summary skill — category labels, totals, and counts are data fields and profile-neutral. The summary observation line ("Food & Dining is the largest category") is the only prose element where register applies. Consult context/writing-profile.md for tone on this line only: a formal profile produces "Food & Dining represents the largest expenditure category"; a conversational profile produces "Food & Dining is where most of the spending went." The redirect phrase "for planning, consider a financial advisor" is always verbatim, never adapted.

Example prompts

• "Categorize my spending from these transactions: [paste]."
• "Where did my money go last month? [paste statement]."
• "Summarize this spending data by category: [paste]."
• "What categories make up my recent purchases? [paste]."

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jmlozano1990
• Source: jmlozano1990/Cowork-Starter-Kit
• License: MIT
• Homepage: https://github.com/jmlozano1990/Cowork-Starter-Kit/releases/latest

Install and usage instructions live in the source repository linked above.