メニュー
Categorize financial transactions by analyzing merchant names, amounts, and context. Assigns spending categories with confidence scores and suggests pattern rules. Use when you need to categorize transactions or determine spending patterns.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Process large volumes of bank statements (50+ PDFs) in batches with checkpointing and progress tracking. Orchestrates the standard ingestion skill across multiple batches for resumable processing.
Parse bank statement CSV files into structured transaction data. Extracts account information and transactions from CSV exports. Use when you need to process CSV bank statements.
Extract transactions from PDF or CSV bank statements in the staging folder, categorize them using Claude AI, insert into database, and archive processed files. Use when you need to ingest, process, or import new bank statements (PDF or CSV format).
Manage transaction memory to view patterns, add quick notes, search for merchants, and clean up outdated entries. Use when you need to check what's in memory, add new learnings, or maintain the memory file.
Parse bank statement PDF text into structured transaction data with account information and transactions in consistent JSON format. Works with any bank format. Use when you need to extract or parse transactions from PDF bank statements.
Interactively review flagged transactions to confirm or correct categorizations, and create rules based on user feedback.
| name | categorize-transactions |
| description | Categorize financial transactions by analyzing merchant names, amounts, and context. Assigns spending categories with confidence scores and suggests pattern rules. Use when you need to categorize transactions or determine spending patterns. |
Automatically categorize financial transactions by analyzing merchant names, amounts, and context to assign appropriate spending categories with confidence scores.
You will receive a JSON object with:
transactions: Array of transactions to categorizeavailable_categories: List of valid category namesexisting_rules: Array of existing pattern-to-category mappings for referencememory_context (optional): Content from TRANSACTION_MEMORY.md with personal context and learnings{
"id": 1,
"date": "2024-12-15",
"amount": -45.23,
"merchant_original": "WHOLEFDS MKTPL #12345",
"merchant_cleaned": "WHOLEFDS MKTPL",
"description": "WHOLE FOODS MARKET PURCHASE",
"account_type": "credit_card"
}
Return a JSON object with categorizations:
{
"categorizations": [
{
"transaction_id": 1,
"category": "Groceries",
"confidence": 0.95,
"rule_pattern": "WHOLEFDS*",
"reasoning": "Whole Foods Market is a well-known grocery store chain."
}
]
}
Use wildcards (*) for flexible matching:
WHOLEFDS* matches all Whole Foods locationsSPOTIFY* matches various Spotify formatsAMAZON.COM* matches Amazon purchasesAvoid overly generic patterns:
MARKET* (too broad)CAFE* alone (could match unrelated)When memory_context is provided, use it to improve categorization accuracy:
Example 1: Time-Based Patterns
Memory says: "Grab weekday mornings (7-9am): Transportation (commute)"
Transaction: GRAB HOLDINGS, $8.50, 2025-08-15 07:30
Without memory: Dining & Restaurants (0.70) - uncertain
With memory: Transportation (0.90) - high confidence due to time pattern
Example 2: Merchant-Specific Rules
Memory says: "Amazon - always flag for review, too varied"
Transaction: AMAZON.COM, $45.99
Without memory: Shopping (0.75)
With memory: Shopping (0.65) - lower confidence, flag for review
Example 3: PayLah/PayNow Extraction
Memory says: "PayLah - check description for actual merchant"
Transaction: PAYLAH, description: "FAIRPRICE FINEST"
Without memory: Transfer (0.80)
With memory: Groceries (0.85) - extracted merchant from description
Example 4: Amount-Based Rules
Memory says: "Guardian Pharmacy: <$20 = Healthcare, >$50 = Shopping"
Transaction: GUARDIAN, $15.50
Without memory: Healthcare (0.75)
With memory: Healthcare (0.90) - amount confirms it's medicine
Example 5: Personal Context
Memory says: "Location: Singapore, Smart Buddy is school canteen"
Transaction: SMART BUDDY, $3.50
Without memory: Dining & Restaurants (0.70)
With memory: Childcare & Kids (0.95) - local context provides certainty
When memory provides clear guidance:
When memory indicates ambiguity or "always flag":
When memory shows confirmed patterns: