Smart Spreadsheet Merge Semantic Dedup

Name: Smart Spreadsheet Merge Semantic Dedup
Availability: InStock
Author: Agensi

by Nihillon

Consolidates folder-wide spreadsheets into a single file using four-layer semantic and fuzzy deduplication.

/mo

Subscription · cancel anytime

Secure checkout via Stripe

Always the latest version
Delivered live to your agent via MCP
Cancel anytime, access ends at period end

0 installsSecurity scanned

See it in action

You say

Consolidate all the spreadsheets in the /client-data folder. There are a lot of duplicates but some names are abbreviated or have typos.

Your agent does

I found 5 files with 1,200 total rows. I've mapped the inconsistent 'CustName' and 'Client' headers to a unified 'Customer' column. I propose using 'Customer' + 'Email' as the dedup key. Once you confirm, I'll run the four-pass semantic detection.

What you get

Consolidate inconsistent spreadsheets into a single source of truth.Identify duplicates that use synonyms, abbreviations, or reordered words.Map disparate headers across multiple files into one unified schema.Generate an audit trail of every merge decision and source file origin.

About this skill

Surfaces: Claude Code ✅ · Cowork ✅ · claude.ai ⚠️ uploaded files only, no local folder path · Chrome ❌ no shell to run cluster.py or write a workbook · API ⚠️ needs Bash plus a turn-taking loop for the two approval gates

Runtime: Python 3.7+ — ordered dict.fromkeys dedup at scripts/cluster.py:73 (inferred); nothing else pins a floor, imports are stdlib only

Packages: openpyxl + pandas for the workbook I/O this skill delegates to the public xlsx skill (SKILL.md:12) — preinstalled in the claude.ai sandbox, elsewhere: pip install openpyxl pandas. scripts/cluster.py needs nothing

Tools: Bash / Python execution (cluster.py, workbook read-write) · directory listing and file read across a whole folder (step 1) · file write and delivery to the user, /mnt/user-data/outputs/ then present (SKILL.md:48) · multi-turn dialogue — steps 4 and 6 are blocking approval gates

Network: none at runtime. pypi.org at install time only if openpyxl / pandas are absent

Model: any. Pass D judges synonyms, translations and brand-vs-legal-name pairs by meaning (SKILL.md:38) — a stronger model produces fewer bad groups to veto at the gate

Companions: requires the public xlsx skill, read first (SKILL.md:12) · spreadsheet-data-auditor routes cross-file dedup here

Degrades to: without shell, Passes A–C fall back to manual grouping and no workbook is written — only trivial folders are feasible. Without a turn-taking session the step-4/6 gates cannot run, and the merge must not proceed without them. LibreOffice recalc is not needed — the output sheets carry values, not formulas

Generated from SKILL.md + scripts/ + references/ on 2026-08-14. Items marked (inferred) follow from a code signal rather than a declaration.

spreadsheet-automation