Codex to reconcile tax CSVs after imports
Sanagira used Codex to complete the work for a 2025 Japanese white tax return after leaving the year's bookkeeping unfinished. The author reports about two hours of work, with a relatively simple business and previous bookkeeping knowledge. Journal data, the trial balance and financial statements were still inspected by a human.
note.com
Original prompt
Shared by the source author. See source notes for context.
MF取り込み後の仕訳帳CSVで、同じチェックを再実行して完了判定して
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- さなぎ羅 (@sanagi_l)
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Repeatable Python checks, missing-entry import CSVs and reconciliation reports used to complete the reported tax work.
- Human involvement
- Sanagira chose accounting rules, performed CSV transfers and inspected journals and statements.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Agree account, date and currency rules with the human.
- Generate and reconcile CSVs, isolating uncertain classifications.
- Import manually into MoneyForward and repeat the same checks.
Sanagira used Codex to complete the work for a 2025 Japanese white tax return after leaving the year's bookkeeping unfinished. The author reports about two hours of work, with a relatively simple business and previous bookkeeping knowledge. Journal data, the trial balance and financial statements were still inspected by a human.
The MoneyForward personal plan did not provide usable API or MCP access, so the author exported and imported CSV files manually. Codex handled repeated processing around those handoffs. Five Python scripts inventoried monthly folders, created a receipt manifest, checked journals, reconciled statements and generated missing card-use entries.
The journal checker handled different CSV encodings and column names, calculated debit and credit differences, and exported anomalies. Reconciliation separated bank transactions and card settlements from card-use entries, matching dates and amounts instead of treating every appearance of a charge as the same record. It exposed matches and records present on only one side.
Merchant rules generated import-ready card journals, while uncertain classifications went into a separate review CSV. The author chose account, date and foreign-currency treatment rules. After importing into MoneyForward, the same checks ran again on the newly exported journal. The source reports completing the work, but does not provide an official acceptance notice or refund outcome.
Prompt
Original prompt
Source: note.com · さなぎ羅 (@sanagi_l)
Frequently asked questions
Why were imports manual?
The author's MoneyForward personal plan lacked usable API and MCP access.
How were uncertain classifications handled?
They were exported into a separate CSV for human review.
What happened after import?
The newly exported journal underwent the same checks again.
More details
About this account
Codex built scripts and processed statements, journals and anomaly reports.
Firsthand source opened and read. Prompt is a verbatim excerpt of the published original instruction.
Credit: さなぎ羅 (@sanagi_l)
Source published
Added to Agents In Use
Source checked
Tools and topics
MoneyForward Cloud · Codex · Python · CSV
Tax CSVs · Card reconciliation · Missing entries
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.