メニュー
Add support for a brand-new county that auto-discovery does not already resolve, API-first, verified with a golden record. Use when a needed county isn't in the registry and probing/discovery can't map it to a usable source.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Diff an appraiser's MLS export (subject + comps) against county assessor public records to flag GLA, beds, baths, year-built, and basement discrepancies before a report goes out. Use when given one or more MLS CSV files to check.
Configure the assessor sources for the counties in an appraiser's area so they are set up once and reused. Use when a user says where they work, lists counties they need, or wants to add their local counties to assessor-lookup.
| name | add-county |
| description | Add support for a brand-new county that auto-discovery does not already resolve, API-first, verified with a golden record. Use when a needed county isn't in the registry and probing/discovery can't map it to a usable source. |
Goal: a new county client that returns the standard record dict, wired into the registry, and verified with a golden record — added in a bounded change (one client module + one registry entry + one golden case).
Check discovery first. discover_county(county, state) (MCP) or
assessor-lookup discover "<County>" --state <st>. If it resolves to a
tier-1 source, verify with a real lookup_property and you may be done. If
it only hits the tier-2 baseline (owner/legal/value, no building data),
continue.
Map the site. Delegate to the explorer agent with the county and a sample property. It returns, API-first, the platform, exact requests, and the field-name mapping — and confirms whether the building fields (GLA/beds/baths/ year) are available. Honor its verdict: prefer a JSON API; scrape only if no API carries the building data.
Implement the client at platforms/<name>.py (config-driven platform)
or jurisdictions/<country>/<state>/<name>.py (bespoke driver exposing
build(entry, timeout=, verbose=)): a class whose
lookup(address) (and ideally lookup_by_parcel(parcel_id)) returns the
standard record dict with a status key, never raising on failure. Mirror
the closest existing client: jurisdictions/us/co/adams.py (ArcGIS JSON), platforms/spatialest.py
(Spatialest JSON), or platforms/eagleweb.py (scrape).
Wire it in. For a new platform, add one line to the PLATFORMS dict
in platforms/__init__.py; then add the county_registry.json entry
(a config-only jurisdiction needs just the entry).
Pin a golden. Add a case to DEFAULT_CASES in assessor_lookup/harness.py, then
python tests/harness.py --capture --filter <case-id>. Use a real
single-family home (not a vacant lot or HOA tract) so building fields are
exercised.
Verify. Hand off to the reviewer agent (live site + golden + parser
correctness). Then confirm pytest -m "not network" and
python tests/harness.py --filter <case-id> are green.
Bounded change, API-first, verified against ground truth. Don't add frameworks; one client module, one registry entry, one golden case.
