Property-Report-Skill-Judges

Property Investigator Skill

Turn any bundle of property documents into a professional Property Investigation Report — and regenerate an updated report whenever you add more documents.

👉 New here? Start with the How-To-Use Guide

Open docs/How-To-Use-Guide.html in a browser for a step-by-step walkthrough of every agent: what to give it, where to drop the files, what type, why each input matters, and the outputs you get. This skill is an orchestrator (SKILL.md) plus five advisor subagents in agents/ — Neighborhood, Loan, Inspection, Insurance, and Comparison. Each runs on its own or as part of the guided home-buying journey.

What it does

• Reads any property docs: seller disclosures, surveys/affidavits, inspection reports, roofing/repair receipts, HOA docs, title/appraisal, listing sheets, and photos — including scanned PDFs (it renders pages to images and reads them visually).
• Builds a normalized data model, resolves conflicts (most recent signed doc wins), and enriches with online market data when available (comps, ZIP medians, builder, outlook).
• Neighborhood/appraisal-district comparison (adaptive): every initial online property investigation pulls the subject’s comp set from the correct county CAD and produces reusable data/ CSV+JSON, a summary report, and an interactive HTML dashboard (filter/sort/scatter). The comp set, appraisal district, and portal are re-resolved per property (every house is different). These outputs are skipped only when the user opts out or online/CAD data is unavailable or blocked, and the report records why.
• Produces Property-Investigation-Report.html with: snapshot, risks (🔴/🟡/🟢 + how to verify), benefits, comparative value, neighborhood comparison, financials, a mandatory yes / conditional yes / no resale-buy verdict, mandatory entry-facing direction + Indian Vastu section with evidence/confidence, due-diligence checklist, open items, and a bottom line.
• Incremental: a manifest tracks processed files, so re-runs only ingest new/changed docs and append a dated Change Log (including any risk-rating changes).
• Compare & pick the best buy: point it at two or more property folders and it produces a structured comparison (side-by-side + weighted scorecard + Vastu/community-resale factor) that ends with one clear recommendation and the condition under which to switch. Saved outside the individual property folders in the parent properties/ folder.

Privacy guardrail: uses public property/parcel records only. It will not use people-search / background-check sites or compile personal profiles of owners/residents. Owner name as printed on the public appraisal roll is the only personal field used, purely as a property attribute.

Files

Property-Report-Skill/
  SKILL.md                              # the skill definition & workflow (orchestrator + entry point)
  README.md                             # this file
  agents/                               # domain subagents (each its own SKILL.md) + README index
    README.md                           # index of the orchestrator + 5 advisors
    neighborhood-advisor/ comparison-advisor/ loan-advisor/ inspection-advisor/ insurance-advisor/
  templates/
    report-template.html                # self-contained HTML report scaffold
    property-data-schema.json           # normalized data model
    neighborhood-dashboard-template.html # reusable interactive comp dashboard
    comparison-template.md              # multi-property best-buy comparison scaffold
  scripts/
    extract_documents.py                # extractor + scanned-page renderer + manifest
    process_narration.py                # demo-video narration helper
  docs/                                 # design & user docs
    AGENT-ARCHITECTURE-DESIGN.md        # full subagent architecture design
    How-To-Use-Guide.html               # step-by-step user walkthrough
  demo/                                 # standalone interactive HTML demos
    Hackathon-Demo-Cockpit.html  Mobile-App-Demo.html
  assets/                               # cover art, video, and presentation/ (deck source + built PPTX)
  app/                                  # runnable reference web app
  properties/                           # local-only property workspaces plus best-buy comparisons
    <Address-Slug>/
      documents/                        # drop raw docs here
      data/                             # neighborhood CSV/JSON outputs (reusable, year-over-year)
      .property/                        # manifest + extraction cache (auto)
      Property-Investigation-Report.html  # the generated main report
      Neighborhood-Comparison-Report.html # generated appraisal-district comp report
      neighborhood-dashboard.html         # generated interactive dashboard
      data/neighborhood-*.csv/.json       # reusable neighborhood parcel data

Quick start

1. Create properties/<Address-Slug>/documents/ and drop the property’s documents there.
2. Ask the assistant: “Create a property report for <Address-Slug>.” The skill will extract, analyze, enrich, and write Property-Investigation-Report.html, Neighborhood-Comparison-Report.html, neighborhood-dashboard.html, and neighborhood CSV/JSON in that property folder when public online/CAD data is available.
3. Later, drop in more documents (e.g., an inspection report) and say “Update the property report.” It re-processes only the new files and regenerates the report with a change log.

Judge delivery

• Meeting launch page: open Judges-Delivery/START-HERE.html for clickable presentation, demo, fallback, and video links. Use Judges-Delivery/README.md for the timed talk track, FAQ, and recovery plan.
• Presentation: open assets/presentation/AI-Property-Due-Diligence-Advisor-Judges-Deck.pptx.
• Live demo (recommended): run npm start from app/, then open http://localhost:5173.
• No-server fallback: open demo/Hackathon-Demo-Cockpit.html; its phone flow is fully synthetic and links to demo/Mobile-App-Demo.html.
• Two-minute video: play assets/video/AI-Property-Due-Diligence-Advisor-Demo.mp4.

The live demo needs no API key and defaults to local, deterministic analysis over fictional fixtures. Rebuild the presentation with npm run build from assets/presentation/.

Manual extraction (optional)

pip install pdfplumber python-docx PyMuPDF openpyxl
python Property-Report-Skill/scripts/extract_documents.py "<property-folder>"

The script prints which files are new/changed and lists any scanned pages to read visually, then writes a manifest under <property-folder>/.property/.

Notes

• Isolation: each property folder is a separate, confidential engagement — the skill never mixes documents across properties.
• Distribution safety: properties/ and comparisons/ are ignored by Git. Keep contracts, quotes, inspection reports, generated reports, and other client data out of distributable branches and judge packages.
• No hallucinations: every claim traces to a document or a cited source; unknowns are flagged “verify.”
• Not professional advice: the report always tells the buyer to confirm material items with licensed professionals.