Accounting Reconciler

Accounting Reconciler

帳務核對助理

Every number tells a story. This Skill makes sure the stories match. 每個數字都在說一個故事。這個 Skill 確保故事吻合。

Identity 身份

• Name: accounting-reconciler
• Type: Transform (raw financial data → reconciliation report with discrepancies)
• For: Accountants, bookkeepers, finance assistants, and anyone responsible for verifying that two sets of financial records agree — bank statements vs. ledgers, invoices vs. payments, budgets vs. actuals.
• 對象: 會計、帳務人員、財務助理，以及任何負責驗證兩組財務記錄是否一致的人——銀行對帳單 vs. 帳本、發票 vs. 付款、預算 vs. 實際。

Description 描述

Reconciliation is the backbone of financial accuracy. It's the process of comparing two sets of records to ensure they match, identifying discrepancies, and resolving them. It's repetitive, detail-intensive, and error-prone when done manually — exactly the kind of work where a structured Skill shines.

核對是財務正確性的骨幹。它是比較兩組記錄確保吻合、辨識差異並解決的過程。人工執行時是重複的、細節密集的、容易出錯的——正是結構化 Skill 發揮的地方。

This Skill handles the systematic comparison, flags discrepancies, categorizes them by likely cause, and produces a clean reconciliation report that an accountant can review and act on.

這個 Skill 處理系統性的比較，標記差異，按可能原因分類，並產出乾淨的核對報告讓會計可以審閱和處理。

Trigger Conditions 觸發條件

• User provides two sets of financial data to compare 使用者提供兩組財務資料比較
• "Help me reconcile these accounts" / 「幫我核對這些帳目」
• "Compare the bank statement with our ledger" / 「比較銀行對帳單和我們的帳本」
• "Check budget vs. actuals" / 「對照預算和實際」
• "Find the discrepancies between these two reports" / 「找出這兩份報告的差異」
• "月結對帳" / "核帳" / "對帳"
• End of month / quarter / year reconciliation cycle 月/季/年底核對週期

Process Logic 流程邏輯

1. Identify what's being reconciled — understand the two data sets: 辨識核對標的：

   • Data Set A: [source, format, period] 資料集 A
   • Data Set B: [source, format, period] 資料集 B
   • Reconciliation type: bank reconciliation, intercompany, budget-vs-actual, invoice matching, other 核對類型：銀行對帳、公司間、預算 vs. 實際、發票配對、其他

2. Standardize the data — before comparing, ensure both sets use: 標準化資料：

   • Same date format 相同日期格式
   • Same currency and decimal precision 相同貨幣和小數精度
   • Same account/category naming 相同帳戶/類別命名
   • Same period boundaries 相同期間界線

3. Perform line-by-line matching using multiple matching strategies: 逐行配對使用多重配對策略：

   • Exact match 精確配對: Amount + date + reference number all match
   • Amount match 金額配對: Amounts match but dates or references differ (timing difference)
   • Partial match 部分配對: Reference matches but amount differs (partial payment or error)
   • One-to-many 一對多: One record in A matches multiple records in B (split transactions)
   • Many-to-one 多對一: Multiple records in A match one record in B (batch deposits)

4. Categorize unmatched items by likely cause: 將未配對項目按可能原因分類：

   • Timing differences 時間差異: Transaction recorded in different periods (e.g., check issued Dec 31, cleared Jan 2)
   • Missing entries 遺漏登錄: Present in one set but absent in the other
   • Amount differences 金額差異: Same transaction, different amounts (possible error)
   • Classification differences 分類差異: Same transaction, different categories
   • Duplicates 重複: Same transaction recorded twice in one set

5. Calculate summary statistics: 計算摘要統計：

   • Total in Data Set A / Total in Data Set B
   • Difference (absolute and percentage) 差異（絕對值和百分比）
   • Number of matched / unmatched items 配對/未配對項目數
   • Breakdown of unmatched by category 未配對項目按分類細分

6. Generate reconciliation report. 生成核對報告。

7. Flag items requiring action — sorted by impact (largest discrepancies first). 標記需要行動的項目——按影響排序（最大差異優先）。

I/O Contract 輸入輸出合約

Input 輸入

• Data Set A: Financial records (spreadsheet, CSV, or structured text) 資料集 A：財務記錄
• Data Set B: Financial records to compare against 資料集 B：比對用的財務記錄
• Optional: tolerance threshold (e.g., ignore differences under NT$10), matching rules, period 可選：容差閾值、配對規則、期間

Output 輸出

# Reconciliation Report: [A] vs. [B]
# 核對報告：[A] vs. [B]
**Period**: [date range] | **Date**: [report date]

## Summary 摘要
| Metric 指標 | Data Set A | Data Set B | Difference 差異 |
|---|---|---|---|
| Total amount 總金額 | [amount] | [amount] | [diff] |
| Item count 項目數 | [count] | [count] | [diff] |
| Matched items 已配對 | [count] ([%]) | | |
| Unmatched items 未配對 | [count] ([%]) | | |

## Reconciliation Status 核對狀態
[🟢 Reconciled / 🟡 Minor discrepancies / 🔴 Significant discrepancies]

## Matched Items 已配對項目
[Summary or full list depending on size]

## Discrepancies 差異項目

### Timing Differences 時間差異 ([count] items, [total amount])
| # | Description 描述 | Set A 集A | Set B 集B | Diff 差異 | Likely Cause 可能原因 |
|---|---|---|---|---|---|
| 1 | [item] | [amount/date] | [amount/date] | [diff] | [cause] |

### Missing Entries 遺漏登錄 ([count] items, [total amount])
| # | Missing From 遺漏於 | Description 描述 | Amount 金額 | Action 行動 |
|---|---|---|---|---|

### Amount Differences 金額差異 ([count] items, [total impact])
| # | Description 描述 | Set A | Set B | Diff | Action |
|---|---|---|---|---|---|

### Other 其他
[Classification differences, duplicates, etc.]

## Recommended Actions 建議行動
1. [Most impactful action first]
2. [Next action]
3. [Next action]

## Adjusting Entries Needed 需要的調整分錄
[If applicable — specific journal entries to correct discrepancies]

Validation Rules 驗證規則

1. Math is exact — totals must be verifiable; no rounding errors in the reconciliation itself. 數學精確。
2. Every unmatched item is categorized — "miscellaneous" is not an acceptable category for more than 5% of items. 每個未配對項目都要分類。
3. Discrepancies are ordered by impact — largest dollar amounts first. 差異按影響排序。
4. No items lost — total items in A + total items in B = matched + unmatched from A + unmatched from B. 沒有項目遺失。
5. Actions are specific — "investigate further" is only acceptable with a specific person and deadline. 行動要具體。
6. Tolerance is disclosed — if a threshold was applied, state it clearly. 容差有揭露。

Knowledge 知識庫

Common Reconciliation Types 常見核對類型

• Bank reconciliation 銀行對帳: Company ledger vs. bank statement. Key issues: outstanding checks, deposits in transit, bank fees not yet recorded, interest not yet recorded.
• Accounts receivable 應收帳款: Customer ledger vs. aging report. Key issues: unapplied payments, disputed invoices, write-offs pending.
• Budget vs. actual 預算 vs. 實際: Planned budget vs. actual spending. Key issues: timing of expenditures, re-categorized expenses, approved overspends.
• Intercompany 公司間: Transactions between related entities. Key issues: transfer pricing, currency conversion, timing of recording.

Domain-Specific Rules 領域特定規則

• Amounts must NEVER be rounded during comparison — compare to the cent/penny/分 比較時金額絕不能四捨五入
• When matching by date, allow for ±3 business days for bank transactions 銀行交易日期配對允許 ±3 個工作日
• A "reconciled" status requires discrepancies below the materiality threshold set by the organization 「已核對」狀態要求差異低於組織設定的重大性門檻
• Always preserve the original data; never modify source records 永遠保留原始資料；絕不修改來源記錄

Red Flags 紅旗

• Same amount appearing multiple times on the same date (potential duplicate) 同日期同金額多次出現
• Round-number differences (potential manual adjustment without documentation) 整數差異
• Recurring unmatched items month after month (systemic issue) 月復一月重複出現的未配對項目
• Differences just below the review threshold (potential manipulation) 差異剛好低於審查門檻

Composition Hooks 組合接口

• Upstream 上游: Data extraction Skills (reading bank statements, parsing invoices), data import tools
• Downstream 下游: Report generation (formatted reconciliation report), journal entry generation, escalation workflows
• Pairs with 搭配: document-reviewer (review the reconciliation report before filing)

Part of the Agentic Substrate Example Skills Gallery.