property-developer-research

Research Indonesian property projects and developers using Webwright/browser evidence. Use when the user asks to investigate a property, housing cluster, developer legality, official website, Sikumbang, Sireng, or generate a buyer due-diligence report from a property/developer name.

Use this skill to investigate an Indonesian housing/property project and developer with screenshot-backed evidence. Default sources for now:

1. Official developer/project website
2. Google Maps coordinate for project/location pinpointing
3. SIKUMBANG (https://sikumbang.tapera.go.id)
4. SIRENG (https://sireng.pkp.go.id)
5. BNPB InaRISK flood raster (https://gis.bnpb.go.id/server/rest/services/inarisk/layer_risiko_banjir/ImageServer)

Use the Webwright workflow: create a task workspace, write plan.md, drive Playwright, save screenshots/logs under final_runs/run_<id>/, verify evidence, then summarize or generate PDF/ZIP if asked.

Prerequisites and setup

Before starting, verify required tools are available:

bash

python3 --version
node --version
npm --version
npx playwright --version

If Playwright or Chromium is missing, install it before browsing:

bash

npm install -g playwright
npx playwright install chromium

For Python helpers/report generation, install missing packages as needed:

bash

python3 -m pip install --user requests beautifulsoup4 lxml pillow reportlab weasyprint pypdf

Optional but useful for polished outputs: zip, pandoc, and ffmpeg. If unavailable, proceed with HTML/PDF generated from Python and clearly note any skipped artifact.

If the webwright skill is installed, follow it for browser automation and screenshot verification. If it is not installed, use direct Playwright scripts and preserve the same evidence/log folder structure described below.

Inputs

Ask only if missing and not inferable:

• property_name — e.g. Clarity House
• developer_name — e.g. Easton Urban Kapital / PT Easton Urban Kapital
• optional: known location, official URL, Google Maps URL/place name, requested output (summary, PDF, evidence ZIP)

If user gives only one of property/developer, proceed with web search to infer candidates, but label inferred values clearly.

Workspace convention

Use:

text

outputs/property_research_<slug>/
├── plan.md
├── final_script.py
└── final_runs/run_<id>/
    ├── final_script.py
    ├── final_script_log.txt
    ├── screenshots/
    ├── data/
    ├── report.pdf          # default for Telegram/group-chat reports unless chat-only requested
    ├── flood_map.png       # if InaRISK checked
    └── evidence.zip        # if requested

slug should combine normalized property + developer when available.

Critical points for plan.md

Always include these unless explicitly out of scope:

• CP1: Identify official project/developer website result and capture source screenshot.
• CP2: Extract official project facts: address/location, unit count, house types, facilities, price/pricelist statement if visible.
• CP3: Extract/pinpoint project coordinate from Google Maps and save the Maps URL/screenshot.
• CP4: Query InaRISK flood raster at/around the Google Maps coordinate and save raw samples/map image.
• CP5: Search SIKUMBANG by exact Nama Perumahan.
• CP6: If exact SIKUMBANG search fails, run parallel fallback queries and rank candidates by similarity.
• CP7: Search SIKUMBANG by Nama Pengembang for the developer.
• CP8: Search SIRENG for developer registration/status and capture registration rows/certificate links.
• CP9: State gaps/caveats clearly: missing project in SIKUMBANG, multiple SIRENG rows, no public price, exact coordinate NoData, etc.
• CP10: Save screenshots, raw text/API responses, and final log under final_runs/run_<id>/.

Source workflow

1) Official website

Preferred path:

1. Search web for "<property_name>" "<developer_name>" official and exact property/developer terms.
2. Prefer developer-owned domains over portals/marketplaces.
3. Capture screenshots for:
   • top/title section
   • project facts/type section
   • price/pricelist/availability statement
   • official brochure/download link if available
4. If e-brochure is linked, open/download it when public. Capture relevant pages and OCR if needed.

Extraction targets:

• project name
• developer/legal entity shown
• address/location
• unit count
• house types and specs
• official prices, or exact statement if prices are not publicly published
• source URLs

Do not quote marketplace prices as official. If price is only on marketplace, label it as third-party listing data.

2) Google Maps coordinate + InaRISK flood risk

For location-based risk, prefer Google Maps place coordinates over broad listing coordinates or address geocoding.

Coordinate extraction priority:

1. User-provided Google Maps URL.
2. Official website embedded Maps/link.
3. Google Maps search result for exact project name + city/area.
4. SIKUMBANG/listing coordinate only as fallback; label it clearly.

Extraction rules:

• Parse coordinates from Maps URLs when available, especially @<lat>,<lon>,<zoom>, !3d<lat>!4d<lon>, q=<lat>,<lon>, or ll=<lat>,<lon>.
• If only a place page is available, open it with browser/Webwright, capture a screenshot, and save the canonical URL containing coordinates.
• Record lat, lon, coordinate source, Google Maps URL, place title, and confidence (exact place, entrance/POI, address-level, fallback).
• If a coordinate conflicts with official/SIKUMBANG address, preserve both and explain the mismatch.

InaRISK flood lookup:

• Endpoint: https://gis.bnpb.go.id/server/rest/services/inarisk/layer_risiko_banjir/ImageServer
• Query the exact Google Maps coordinate and surrounding samples/rings. Use a small local grid around the point, with summaries for ~100 m, 250 m, 500 m, and 1 km when practical.
• Treat pixel values as continuous flood-risk index 0–1. Report classes: Low < 0.333, Medium 0.333–0.666, High >= 0.666.
• Exact coordinates may return NoData. If so, do not stop; use nearby valid samples and report exact pixel as NoData plus ring average/max and high-risk sample share.
• Save raw JSON responses/samples, query URLs/parameters, and a map image/legend screenshot when possible.
• Caveat every report: InaRISK is screening data, not parcel-level proof. Recommend field checks: resident flood history, drainage, elevation, river/canal proximity, and access-road flooding.

3) SIKUMBANG

Base endpoint often works for structured lookup:

text

https://sikumbang.tapera.go.id/ajax/lokasi/search

Parameters:

text

keyword=<query>
selectedSearch=data-lokasi
skalaPerumahan=semua
sort=terbaru
searchBy=nama-perumahan | nama-pengembang
page=1
limit=50

Required searches:

1. Exact property search: searchBy=nama-perumahan, keyword=<property_name>.
2. Developer search: searchBy=nama-pengembang, keyword=<developer_name> and common variants with/without PT.

If exact property search returns 0, run fallback fan-out in parallel:

• exact phrase
• lowercase normalized phrase
• no-space variant
• first two words
• adjacent bigrams
• meaningful tokens length >= 5
• location token if known
• developer/project aliases found from official website

Deduplicate by idLokasi. Rank candidates using phrase similarity + token overlap. Report:

• original query
• exact query result count
• fallback query that found top candidate
• similarity score
• idLokasi
• namaPerumahan
• developer name
• province/city/district/village if available
• commercial/subsidy unit counts
• house types and prices if present in SIKUMBANG

For detail/siteplan pages, capture screenshots when candidate is strong enough or when user asks for deeper extraction.

4) SIRENG

Use official site:

text

https://sireng.pkp.go.id

Search exact developer name and variants:

• without PT
• with PT
• normalized company words

Capture:

• result table screenshot
• raw visible text
• API response if available
• certificate links (/certificate/...) when present

Extract:

• developer name
• registration number (SRG-...)
• association
• address
• status (AKTIF/other)
• certificate URL/PDF if public

If multiple rows exist, do not collapse them. List each row and explain that the buyer must confirm which legal entity/NPWP/signatory is used for the transaction.

Report structure

Default to generating a PDF report for Telegram/group-chat property research unless the user explicitly asks for chat summary only. If time is tight, create a concise PDF from the verified findings and include references/evidence file paths; do not leave the result as chat-only.

For a PDF or final written report, use this order:

1. Executive verdict: Low / Moderate / High public-source risk
2. What was checked
3. Official website findings
4. Google Maps coordinate and InaRISK flood-risk findings
5. SIKUMBANG findings
6. SIRENG findings
7. Key gaps and buyer questions
8. Recommended documents to request before booking fee / DP
9. Evidence appendix with screenshots, source URLs, and raw evidence paths

Risk heuristics:

• Lower risk: official site exists, SIKUMBANG project match, SIRENG active developer, consistent names/locations.
• Moderate risk: official + SIRENG positive, but exact SIKUMBANG project missing or names mismatch.
• Higher risk: no official source, no SIKUMBANG match, no active SIRENG row, conflicting developer/legal names, or high InaRISK flood-risk surroundings without convincing mitigation evidence.

Always say this is public-source due diligence, not a substitute for AHU, OSS/NIB, NPWP, land certificate/BPN, PBG/SLF, PPJB/AJB review, or notary/PPAT verification.

Evidence standards

• Every major claim needs a screenshot, URL, raw text/API response, or downloaded PDF.
• Save exact query URLs and result counts.
• When data conflicts, preserve both values and name the source for each.
• Before sending, verify generated PDFs with page count + text extraction and verify screenshots/maps are readable.
• For InaRISK findings, save both raw numeric samples and the visual map/screenshot used for interpretation.

Telegram delivery preference

When delivering artifacts to Telegram, send:

1. concise summary message
2. PDF report if generated
3. evidence ZIP if generated

Keep the chat summary short; details belong in the PDF/evidence pack.