invoice

Invoice Tools — Claude Code Plugin

A Claude Code plugin for homeowners managing large residential construction projects. Audits monthly draw requests from your general contractor, extracts invoice data from PDFs, cross-references against your budget, and drafts emails to your GC.

What It Does

Every month your GC sends a draw request — a PDF with a cover page, a detail summary of every charge, and a stack of supporting invoices (typed invoices, handwritten receipts, progress billings, rental invoices, etc.). This plugin reads all of it, verifies the math, checks every invoice against the summary, cross-references your budget spreadsheet, and tells you what to pay, what to question, and what to dispute.

Skills

/invoice:invoice-analyzer

The main skill. Give it the draw PDF and your actuals spreadsheet, and it produces:

• A structured audit spreadsheet (.xlsx) with tabs for line items, budget cross-reference, math verification, and flagged issues
• A narrative summary with specific questions for your GC
• Flags for budget overruns, missing COs, invoice typos, cost code mismatches, out-of-period charges, and missing documentation

> /invoice:invoice-analyzer

Then point it at your files:
  data/feb/Draw Request.pdf
  data/feb/Actuals.xlsx

/invoice:extractor

Extracts structured data from a single PDF invoice — typed invoices, photographed receipts, handwritten invoices, progress billings, rental invoices, pay applications. Handles wildly different formats.

> /invoice:extractor

Upload or point it at any construction invoice PDF.

/invoice:matcher

Matches line items from the GC's detail summary to their supporting invoices. Flags mismatches in amounts, vendor names, and invoice numbers. Uses greedy and bipartite matching algorithms.

> /invoice:matcher

Give it the parent summary and a set of supporting invoices.

/invoice:boundary-detector

Splits a multi-document PDF into individual documents. Construction draw PDFs typically concatenate a cover page, detail summary, and 20-40 supporting invoices into one file. This skill finds where each document starts and ends.

> /invoice:boundary-detector

Point it at a concatenated draw request PDF.

/invoice:cleanup

Normalizes extracted invoice data — fixes vendor name variations, handles credits in parentheses, distinguishes subtotals from line items, cleans up progress billing math, and standardizes date formats.

> /invoice:cleanup

Run it on raw extraction output that needs normalization.

/invoice:email-drafter

Turns audit findings into a plain-text email to your GC. Calibrates tone to severity — self-deprecating about penny discrepancies, direct about missing change orders, professional about disputes.

> /invoice:email-drafter

Run it after the invoice-analyzer to draft your response.

Installation

1. Clone this repo into your Claude Code plugins directory:

   git clone https://github.com/scottmsilver/invoice-tools-marketplace.git \
     ~/.claude/plugins/marketplaces/invoice-tools
   

2. Install Python dependencies:

   pip install -r ~/.claude/plugins/marketplaces/invoice-tools/plugins/invoice/lib/requirements.txt
   

3. Restart Claude Code. The skills will appear as invoice:invoice-analyzer, invoice:extractor, etc.

Project Setup

For best results, keep a project-context.md file alongside your draw PDFs with project-specific details:

• GC name and contacts
• Builder's Comp rate
• Cost code mappings
• Prior dispute history
• Special billing rules

The invoice-analyzer will pick this up automatically. If it's not there, it extracts what it can from the draw itself.

What Gets Flagged

• Math errors (line item sums, BC rate, retention, totals)
• Invoice number mismatches between summary and supporting docs
• Cost code mismatches (summary vs PO numbers)
• Budget overruns by cost code (against both original and revised budgets)
• Missing change orders for over-budget items
• Out-of-period invoices
• Missing vendor invoices (e.g., only a credit card receipt)
• 50% deposits and future payment exposure
• Internal GC labor charges without external backup
• Invoices billed to jobsite personnel rather than the homeowner

Python Library

The lib/ directory contains tested implementations:

• PDFExtractor — PDF text extraction and page splitting (PyMuPDF)
• LLMService — multi-provider LLM calls (Gemini/Claude/OpenAI) with rate limiting, caching, JSON repair
• ValidationService — line item matching with greedy and bipartite algorithms
• ImageConverter — convert photographed receipts to PDF

License

MIT