-
v0.81.0
Stablejoeyr released this
2026-10-09 23:29:11 -05:00 | 0 commits to main since this releaseAdded
get_ensemble_tracks, the weather tenant's twenty-first tool and the last of the
tropical-storm plan. The spaghetti plot for an active storm, from NHC's own
model-track file: NHC's official forecast; the American, UK, Canadian and US Navy
global models; the American and Canadian ensemble means; the HAFS-A, HAFS-B, HWRF, HMON
and Navy hurricane models; and NHC's track, intensity and corrected blends -- each
under its own name, with its track and its strength, every 12 hours to 5 days.- Each model at its own latest run. A run's models arrive over hours, so the newest
run held 33 to 35 models where a full one held 74 to 82; each model says whether it is
behind the newest cycle. The model's own run is read rather than NHC's time-shifted
copy, which carries no pressure. - The European model is named as missing: ECMWF's tracks are not in NHC's public
file. Where NHC writes missing as zero -- a blend with no track at0N 0W, one with no
strength at wind 0 -- the answer says not given, never zero. - Checked live against all three active storms: all 45 model runs matched NHC's file
point for point, and NHC's official track matchedget_storm_forecastat 12 of 12
times.
Changed
- The boot line's
upstreamsnamesftp.nhc.noaa.gov.
Removed
docs/weather-plan.md, emptied: this release finishes the tropical-storm tools'
third phase. Its live checks are 24 to 33 indocs/operations.md.
Reconnect the weather connector after deploying, or Claude keeps the old tool list.
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads
-
v0.80.0
Stable2026-10-09 23:17:17 -05:00 | 1 commits to main since this releaseAdded
get_rainfall_outlook, the weather tenant's twentieth tool and the third of the
tropical-storm plan's third phase. The Weather Prediction Center's Excessive Rainfall
Outlook at a coordinate for each of the next five days: Marginal, Slight, Moderate or
High, as WPC labels them, with each day's valid period and issue time.- What a category means is WPC's sentence: the probability that rainfall exceeds
flash flood guidance within 40 km (25 miles) of the point. Nothing here computes a
percentage; a point in no area is below Marginal, WPC's lowest, never zero. - WPC's issue time carries no zone, and it is UTC. Every day's stamp fell 7 to 24
minutes before the map file's own UTC time, and WPC's day 1 page said 0128 UTC beside
the map's 01:24. - Each day has its own stale line: day 1 is reissued three times a day (its
discussion at most 9.9 hours apart in a week), days 2 and 3 twice, and WPC states no
schedule for days 4 and 5. One day past its ceiling costs that day. - Contiguous United States only, as WPC draws it. Alaska, Hawaii, Puerto Rico and
everywhere abroad get no outlook, said as not the same as no risk. - Checked live: at eight places, all 35 day categories matched an independent test of
the point against each area's own outline, and a point inside day 1's High area came
back High.
Reconnect the weather connector after deploying, or Claude keeps the old tool list.
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads
-
v0.79.0
Stable2026-10-09 23:02:31 -05:00 | 2 commits to main since this releaseAdded
get_marine_forecast, the weather tenant's nineteenth tool and the second of the
tropical-storm plan's third phase. The National Weather Service's marine forecast for
the waters at a coordinate, word for word: the forecast for that stretch of water, any
headline such as a small craft advisory, and the product's synopsis of the weather
pattern. Coastal waters, bays, the Great Lakes, and the offshore waters NHC and the
Ocean Prediction Center forecast.- A point on land gets the nearest waters, within about 5, 12 or 25 miles, decided
with the owner: the tool looks around the point at those distances and says how near
the waters it found are. An exact distance would need each zone's outline, and one
bay's was 630 KB. Inland, abroad and on the open ocean the answer says there is no
marine forecast there, and why. - A zone names an office, not always its product. Coastal waters in the lower 48 and
Great Lakes nearshore waters are filed under the zone's own office; NHC's and the Ocean
Prediction Center's offshore waters, the Great Lakes' open waters, Alaska and the
Pacific islands are not. A measured table finds all 694 zones a point lookup can
return, each in exactly one product. - Great Lakes nearshore forecasts carry no synopsis, so their lake's open-lake
forecast's is given, named insynopsis_from. - Official warnings stay with
get_alerts, by decision; it was checked to return a
small craft advisory at a point on the water.
Reconnect the weather connector after deploying, or Claude keeps the old tool list.
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads
-
v0.78.0
Stable2026-10-09 18:50:54 -05:00 | 3 commits to main since this releaseAdded
get_buoy_observations, the weather tenant's eighteenth tool and the first of the
tropical-storm plan's third phase. What NOAA's National Data Buoy Center measures at
sea near a coordinate: wind, gusts, wave height and period, pressure, and air and water
temperature, from moored buoys and offshore platforms, and from coastal stations on
piers, harbor tide gauges and lighthouses.- Two lists, decided with the owner. Up to five buoys and platforms within 150 miles,
and up to three coastal stations only within 25 miles, so a coastal city gets both and
the open sea gets buoys only. Nothing near is a bounded answer that points to
get_observationsfor weather on land. - Each value carries its own time. NDBC's latest-observations file had no wave height
for 136 of its 290 moored buoys, most of which had measured one minutes earlier, so each
reading comes from the station's five-day file: the newest report, with what it lacks
taken from the three hours before it and that value's time named inmeasured_earlier. - One station's history on request: pass
stationfor one reading per hour over up
to five days. An id NDBC does not list, a station with no weather sensors and one
silent for five days are three different answers. - Checked live against NDBC's own files: near a Gulf coastal city, in the open Gulf
and inland, every value the tool returned matched the station's file, time included,
and five buoys' earlier wave heights came back with the times they were measured.
Changed
- The boot line's
upstreamsnameswww.ndbc.noaa.gov.
Reconnect the weather connector after deploying, or Claude keeps the old tool list.
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads
-
v0.77.0
Stable2026-10-09 18:03:09 -05:00 | 4 commits to main since this releaseAdded
get_storm_surge, the weather tenant's seventeenth tool and the last of the
tropical-storm plan's second phase. For a coordinate on the US coast, three National
Hurricane Center surge forecasts, each under its own name: the peak surge range for the
bay, sound or lake the point is in, or for the nearest named stretch of open coast with
its distance; the chance of more than 5 ft of water above ground at the point; and the
height with a 1-in-10 chance of being topped there.storm_idis optional. Without it every active storm NHC publishes surge maps for
is read, and storms without one are named as such, never as no risk.- Checked live against NHC's own advisory: every open-coast stretch the tool named
carried the range the advisory printed for it, and the two surge-odds maps agreed with
each other at 100 of 101 sampled points. - Official surge watches and warnings stay with
get_alerts, by decision.
Fixed
- A map area with several holes in one boundary element lost all but the first.
NHC's peak-surge file writes Mobile Bay that way; earlier maps never did, so no released
answer was affected.
Changed
tests/test_no_pii.pynow scans.kmlfiles, and NHC's own downloads in
tests/fixtures/nhc/are exempt from its coordinate and host rules only while each
file's SHA-256 matches the list in that test.
Reconnect the weather connector after deploying, or Claude keeps the old tool list.
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads
-
v0.76.0
Stable2026-10-09 17:44:00 -05:00 | 5 commits to main since this releaseAdded
get_wind_probabilities, the weather tenant's sixteenth tool and the first of the
tropical-storm plan's second phase. For a coordinate: the National Hurricane Center's
chance of tropical-storm-force (39 mph), 50-knot (58 mph) and hurricane-force (74 mph)
winds over the next five days, as NHC's own band, such as "10-20%".- When tropical-storm-force winds could arrive, for each storm whose map reaches the
point: NHC's earliest reasonable and most likely arrival, each as the window between two
of NHC's six-hourly lines, never a single time. NHC draws these only where the odds
reach 5%, so lower odds skip them. - The odds cover every active storm combined, as NHC publishes them, and the answer
says so. A point outside every band is below NHC's lowest band, not zero. - Checked live against NHC's own per-town bulletins for three storms: 34 of 36 printed
numbers fell in or at the edge of the band returned. The two that did not also
disagreed with NHC's own half-degree grid, which matched the map, so the answer now
says NHC's town table and its map can differ. - NHC's own files for one advisory are kept unchanged in
tests/fixtures/nhc/and
tested against directly.
Reconnect the weather connector after deploying, or Claude keeps the old tool list.
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads
-
v0.75.0
Stable2026-10-09 15:27:43 -05:00 | 6 commits to main since this releaseAdded
get_storm_forecast, the weather tenant's fifteenth tool and the last of the
tropical-storm plan's first phase. The National Hurricane Center's forecast track for
one active storm: where its center is forecast to be at each forecast hour, with the
valid time, maximum wind, gusts and stage (hurricane, tropical storm, post-tropical),
plus the six-hourly track it has already taken.include_coneadds the cone's outline.- Every answer says how to read the cone: it is where the center may go, not the
storm's size, the whole track stays inside it only 60 to 70 percent of the time, and
being outside it does not mean being safe. - Checked against NHC's forecast advisories: all 18 forecast points of three live
storms matched on time, position, wind, gust and stage. - An intermediate advisory's track is labelled as one. NHC re-issues the track at
each intermediate advisory with the current position moved and the forecast points
unchanged, and the answer says so. - A missing pressure is unknown, not 0 mb, and a storm with no forecast still returns
its past track and says why, pointing toget_storm_advisory. - Map links from NHC's storm list are followed only on NHC's own host.
Phase 1 of
docs/weather-plan.mdis complete. Reconnect the weather connector after
deploying, or Claude keeps the old tool list.Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads
-
v0.74.0
Stable2026-10-09 15:15:46 -05:00 | 7 commits to main since this releaseAdded
get_tropical_outlook, the weather tenant's fourteenth tool. Where the National
Hurricane Center expects a tropical cyclone may form: each area it is watching in the
Atlantic, East Pacific and Central Pacific, with NHC's chance of formation within 48
hours and within 7 days, its category (Low, Medium or High), NHC's own paragraph about
the area, and the box around the area NHC drew.basinnarrows it to one basin.- Chances are passed on as NHC writes them, and the tool says a formation chance is
not the chance of a storm reaching a place. Checked live against NHC's text outlook for
all three basins. - A quiet basin is described in NHC's words: "Tropical cyclone formation is not
expected during the next 7 days." - Each basin carries its own issue time and age, and the answer is dated by the
oldest, so one stale basin cannot hide behind two fresh ones. The stale line is eight
hours, measured from a week of outlooks. One basin failing costs only that basin and is
named inbasins_unavailable. - Out of season an old file is reported as old, not refused. NHC issues outlooks
routinely only from June 1 (May 15 in the East Pacific) to November 30. - NHC's map files are read with Python's standard library, so no new dependency. The
same reader will serve the track, cone and surge tools.
Reconnect the weather connector after deploying, or Claude keeps the old tool list.
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads
-
v0.73.0
Stable2026-10-09 15:00:53 -05:00 | 8 commits to main since this releaseAdded
get_storm_advisory, the weather tenant's thirteenth tool. The National Hurricane
Center's latest public advisory and forecast discussion for one active storm, verbatim:
watches and warnings, position, strength, movement, hazards, and the forecaster's
reasoning. It takes thestorm_idthatget_active_stormslists, such asal092026.- The products are the ones
get_active_stormsnames. NHC files its bulletins under a
slot it reuses for later storms, and re-sends a corrected bulletin under the same
number, so each product is matched by the storm and advisory number in its own header,
never taken as the newest. The two tools agree on the number and the issue time,
checked live against three storms. issuedis the time NHC prints, not the time the bulletin was sent, which ran 7 to
20 minutes earlier.- One product missing costs only that product, named in
products_unavailablewith
the reason and NHC's own page for it. Both missing is a refusal, never an empty answer. - An intermediate advisory and its discussion are told apart. Beside advisory 12A the
discussion is still number 12, because NHC writes one only with the full advisories, and
the answer says so.
Reconnect the weather connector after deploying, or Claude keeps the old tool list.
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads
-
v0.72.0
Stable2026-10-09 14:00:06 -05:00 | 9 commits to main since this releaseAdded
get_active_storms, the weather tenant's twelfth tool and its first tropical-storm
tool. It lists every storm the National Hurricane Center is issuing advisories on,
in the Atlantic, East Pacific and Central Pacific, strongest first: name,
classification, wind, pressure, position, movement, the latest public advisory's
number and issue time, and NHC's links to its advisory, track, cone and surge
products.basinnarrows it to one basin. It takes no coordinate.- Each speed is converted from its own unit. NHC's file gives the wind in knots and
the movement in miles per hour, and says neither. Checked against each storm's own
advisory with three storms active; read as knots, a storm moving 16 mph would have been
reported at 18.4. - Winds are rounded to the nearest 5, as NHC's advisories print them: 120 mph, not
120.8, from 105 knots. The advisory number reads as printed,12Arather than the
file's012a. - An empty list says what it covers. It means no storm has advisories now, not that
none is coming; an empty basin names the storms active elsewhere. A product NHC does not
issue for a storm is named inproducts_unavailablerather than dropped. - The list is stale past seven hours, measured: 62 consecutive advisories were never
more than 6.42 hours apart. It is dated by its newest storm update, and a day without
one is refused as a broken feed. timezone: "local"is UTC on this tool, labelledtimezone_source: "no_coordinate"
so it is not mistaken for a failed lookup.docs/weather-plan.mdis open again, for the remaining nine tropical-storm tools.
Reconnect the weather connector after deploying, or Claude keeps the old tool list.
Downloads
-
Source code (ZIP)
0 downloads
-
Source code (TAR.GZ)
0 downloads