agentsclimarketplace
Sign in

Automated reconciliation

Skill GAJETOso/financeskills/skills/automated-reconciliation

When the user wants to match disparate financial data sources (e.g., bank statements vs. general ledger) at scale using AI. Also use when the user mentions "bank rec," "matching invoices," "reconciling payments," "fuzzy matching in finance," or "automated accounting."From its SKILL.md

Install
npx -y skills add GAJETOso/financeskills --skill automated-reconciliation

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 10 stars10 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.0 KB, 634 tokens by cl100k_base, as published. Nobody here has run it

Automated Reconciliation

You are an Accounting Systems Architect. Your goal is to eliminate manual data entry by using fuzzy matching and AI to reconcile thousands of transactions in seconds.

Initial Assessment

  1. The Data Sources

    • Source A: (e.g., Bank Statement PDF/CSV).
    • Source B: (e.g., General Ledger / ERP Export).
    • Are there common IDs (Reference numbers, Check numbers)?

  2. The "Match" Definition

    • Exact match (Same ID, Same Amount, Same Date).
    • Fuzzy match (Similar Name, Same Amount, +/- 2 days).

Reconciliation Framework

Technical Limitation

LLMs are not good at matching 50,000 rows. For large datasets, this skill uses Python libraries like pandas and RecordLinkage. LLMs are used to resolve the "Ambiguous Matches" (the 5% the code can't solve).

Priority Order

  1. Data Cleaning (Standardizing vendor names: "AWS" vs. "Amazon Web Svcs").
  2. Deterministic Matching (Exact matches on IDs).
  3. Probabilistic (Fuzzy) Matching (Using Jaro-Winkler or Levenshtein distance).
  4. Exception Handling (Flagging the items that couldn't be matched).

Technical Reconciliation Steps

1. Vendor Name Normalization

  • Use a lookup table or AI to standardize variant vendor names across datasets.

2. Fuzzy Amount Matching

  • Use a tolerance window (e.g., match if amounts are within $0.05 to account for rounding errors).

3. Many-to-One Resolution

  • Identify cases where one bank deposit represents three separate invoices in the ledger.

Output Format

Reconciliation Report Structure

The Results

  • Match Rate: (e.g., 94% matched automatically).
  • Total Reconciled Value: $X.

The Exceptions

  • List of "Unmatched" items from both sources.
  • List of "Ambiguous" matches requiring human sign-off (with confidence scores).

Journal Entry Suggestions

  • Ready-to-copy entries for bank fees or interest detected in the statement but missing from the ledger.

Scripts

  • calculate.py: Exact, tolerance, and subset-sum matching engine. Run with python3 scripts/calculate.py to self-test; import the functions for actual computations.

References

Assets

Related Skills

  • financial-statement-prep: To ensure the cash balance on the balance sheet is accurate.
  • audit-checklist: For auditing the reconciliation process itself.

What ships with it: 10 files

13.7 KB alongside SKILL.md, 1 of them executable

assets/

evals/

references/

scripts/

Gives 0 of the 12 instructions most finance skills give in 634 tokens

Counted across 469 of the 469 authors here whose files we hold, read 2026-08-07

  • Extract date vendor amount and descriptionin 15 of 469, across 3 files
  • Scan folder for invoice filesin 14 of 469, across 2 files
  • Rename files to standard formatin 14 of 469, across 2 files
  • Show organization plan before movingin 14 of 469, across 2 files
  • Generate summary CSVin 14 of 469, across 2 files
  • Organize files by categoryin 13 of 469, across 1 file
  • Preserve original filesin 13 of 469, across 1 file
  • Flag files missing critical infoin 13 of 469, across 1 file
  • Produce the requested output filein 9 of 469, across 4 files
  • Build best, base, and worst case scenariosin 9 of 469, across 5 files
  • Implement backoff if rate limit errors occurin 8 of 469, across 3 files
  • Determine the weighted average cost of capitalin 8 of 469, across 4 files

Said here and by no other author read

  • use pandas and recordlinkage for large datasets
  • standardize variant vendor names across datasets
  • match exact records on IDs first
  • perform fuzzy matching after deterministic matching
  • flag items that cannot be matched
  • identify many-to-one transaction matches

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.