israellias/sync-bisa

Sync BISA Pipeline

All python commands MUST use .venv/bin/python (not python or python3).

When the user runs this skill:

1. Run the export + convert pipeline: .venv/bin/python -m bisa --dry-run [--since-date YYYY-MM-DD if provided] (This opens the browser, exports CSV, converts transactions, saves to bisa/transactions.json)

2. Load the converted transactions: .venv/bin/python -c "import json; from bisa import load_pending_transactions; print(json.dumps(load_pending_transactions()))"

3. Check if bisa/rules.md exists:

   • If it exists: read it for deterministic rules
   • If it doesn't exist: ask the user to provide their transaction categorization rules, explaining the expected structure:
     • Credit Card Payment Rules: list of memo patterns that indicate a credit card payment (these get assigned to a transfer account)
     • Amount-Based Rules: a markdown table with columns Pattern (in memo) | Amount Range | Payee | Category for when transaction amounts determine categorization (e.g. loan payments by amount range)
     • AI Guidance: free-form notes to help categorize unmatched transactions (e.g. "POS transactions are usually stores", "TRASP.CTAS.TERCEROS are bank transfers") Write the user's input to bisa/rules.md with proper markdown formatting, then continue

4. Fetch BOB Budget categories (returns [{"id": "uuid", "name": "Category Name"}, ...]): .venv/bin/python -c "import json; from bisa import get_bob_categories; print(json.dumps(get_bob_categories()))"

5. For each transaction: a. Apply payee+category rules from bisa/rules.md (pattern matching on memo) b. Apply amount-based rules (e.g. SYSPago ranges) c. For unmatched: use AI judgment based on memo text + AI Guidance section in bisa/rules.md d. Fuzzy-match the rule's category name against actual YNAB category names (ignore emojis) e. Set category_id on the transaction using the matched category's id (YNAB requires the UUID, not a name)

6. If --auto-accept was passed: skip to step 8

7. Show the user a table: Date | Amount | Payee | Category | Memo (truncated) | Source (rule/AI) Ask user to approve or adjust

8. Write the final categorized transactions back to bisa/transactions.json, then import to YNAB: .venv/bin/python -c "import json, sys; from bisa import import_to_ynab; txns = json.loads(sys.stdin.read()); result = import_to_ynab(txns); print(json.dumps(result))" (pipe the JSON via stdin)