| name | onboard-locale |
| description | 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. |
Onboard a locale
Goal: take the counties an appraiser works in and configure each one for
repeated use — cached source + a pinned golden record — reporting honestly which
are usable for the full discrepancy check.
Steps
-
Gather inputs. Get the county list and one sample property per county
(address or parcel). If the user has an MLS export, pull a sample from it;
otherwise ask for one per county — coverage can't be measured without a real
property. Delegate the run to the county-onboarder agent if you want it
handled end-to-end.
-
Probe each county with the MCP tool probe_county(county, state, address|parcel) (or python tests/harness.py --probe "<County>" --address "..."). Read check_ready:
- true → GLA/beds/baths/year come through; full discrepancy check works.
- false → only owner/legal/value available; the county's public data is
thinner. Say so plainly — it is the source's data, not a defect.
-
Onboard the keepers with onboard_county(...) (or
python tests/harness.py --onboard "<County>" --parcel ...). This caches the
assessor source and pins a golden record to the user config
(~/.config/assessor-lookup/), so the county is re-checked on every run.
-
If a county doesn't resolve at all, switch to the add-county skill.
-
Verify and report. Run run_regression (or python tests/harness.py) to
show every onboarded county reacting. Present a table:
county | platform | check-ready | latency | notes. Then tell the user
they can point MLS exports at the appraisal-check skill.
Principle
API-first, and honesty about coverage. Never present a not-check-ready county's
blanks as if the check succeeded.
