Stats
Actions
Tags
From rhiza-forecasting
Fetch CHIRPS precipitation observations for a date range — the validated final product back to 1998, with a preliminary fallback for very recent days — and write a weather-skills envelope Zarr. Use when a task needs CHIRPS rainfall, recent or historical, e.g. to compare against a forecast or station data, or to build a reference period.
How this skill is triggered — by the user, by Claude, or both
Slash command
/rhiza-forecasting:chirps-fetchThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Downloads CHIRPS v3.0 daily `sat` precipitation for the requested date range and writes a global-grid Zarr store. Each day is taken from the validated **final** product (a per-year archive covering 1998 to present) when available, falling back to the **preliminary** product for very recent days the final has not finalized yet. When both exist for a day, final is used.
Downloads CHIRPS v3.0 daily sat precipitation for the requested date range and writes a global-grid Zarr store. Each day is taken from the validated final product (a per-year archive covering 1998 to present) when available, falling back to the preliminary product for very recent days the final has not finalized yet. When both exist for a day, final is used.
Coverage starts in 1998 (CHIRPS v3.0 sat); dates before 1998 are unavailable and exit non-zero.
uv run --script ${CLAUDE_SKILL_DIR}/scripts/fetch.py --start YYYY-MM-DD --end YYYY-MM-DD --output <path.zarr>
--start, --end — inclusive date range. Each value is one of:
YYYY-MM-DD;now or today — the current UTC date;latest — the newest available CHIRPS prelim day (found by probing the data
server backward day-by-day over HTTPS to the first published day);now-<int>{d|w} or latest-<int>{d|w} — the base minus N (w = 7
days, so 3w = 21 days). The offset is capped at 36525 days; a larger value,
a future + offset, a month/year unit, or any malformed value exits 2 before
any network call.Boundary handling: absolute endpoints and ordinary relative ranges are
inclusive of both ends. The duration idiom — start B-<int>{d|w} paired
with end exactly its own base B (both now, or both latest) — yields an
N-day window inclusive of B, with the far edge shifted in by one. So
latest-3w .. latest resolves to [latest-20d, latest] = 21 days incl.
latest, and now-1w .. now resolves to 7 days; 2026-05-01 .. 2026-05-10 is
10 days. Tokens stay literal (latest-3w = latest − 21d); only the
B-N .. B shape moves the far edge. After resolution, start <= end or the
run exits 2 (pre-network). For any invocation using a relative token, the
resolved concrete window is echoed to stderr before fetching, e.g.
resolved "latest-3w".."latest" -> 2026-05-10..2026-05-30 (21 days; duration mode: 3-week window inclusive of latest)
or ... (inclusive both ends). latest discovery runs at most once and only
when a token references latest. The cache key records the resolved absolute
dates, never the relative token.
--output, -o — output Zarr path (overwritten if it exists).
--workers — max concurrent per-day download threads (default 2). Bounds the thread pool that fetches each day's TIF over HTTPS. A concurrency knob only, not data: it is excluded from the cache key. The default is deliberately conservative: CHC's data server can throttle and temporarily block IPs under higher concurrency. If throttling errors appear, lower it to 1. If requests are refused outright, the IP may be temporarily blocked — wait before retrying; lowering --workers helps only before a block.
Zarr with data variable precip (mm/day) and dims (time, latitude, longitude) on the global CHIRPS grid. Stamped with weather_skills_source=chirps.
There is no --bbox flag: the full 0.05° global grid (~7200×3600 cells, ~104 MB/day as float32) is always fetched. The output is streamed to Zarr one day at a time, so peak resident memory is bounded to ~one day regardless of how long the window is.
--workers is the network-concurrency speed lever and is memory-neutral: each worker transiently holds only the compressed TIF body (a few MB), not a decompressed global array — decompression happens sequentially after the download pool drains. The default is held low because CHC's data server can throttle and temporarily block IPs; raising --workers is an operator choice for infrastructure where that risk is acceptable (see the --workers argument above).
All per-day TIFs are staged to a temp directory before writing, so a very long window is bounded by temp disk, not RAM. For tight-memory hosts, keep the window short and run the clip-region skill immediately after to shrink to your area of interest.
Historical days come from the validated final product and carry no publication lag; the lag and partial-tail behavior described here apply only to the recent tail, which is served by the preliminary product. The CHIRPS v3.0 daily preliminary product is published on a pentad-based schedule: per-day files appear in batches 2 days after each pentad closes (pentads end on the 5th, 10th, 15th, 20th, 25th, and last day of each month). Best-case lag is 2 days (the last day of a pentad, published 2 days later); worst case is ~7 days (the day right after a pentad ends, which waits for the next pentad to close before its batch is published). Average lag is 4-5 days. See https://www.chc.ucsb.edu/data/chirps3 for the official schedule. When the requested --end falls inside the lag window, the script writes a partial zarr covering only the days that were available on the server, logs the missing days and effective end date to stderr, and exits 0. If days are missing from the middle of the range (not the tail), the script exits 2 — that's a server-side data gap, not a lag issue. The weather_skills_history entry's args.end reflects the EFFECTIVE end actually written (not the requested --end), so a re-run against the same --end cache-misses on the partial output and re-attempts the missing tail.
The output stamps a JSON-encoded weather_skills_history attr: an append-only array of
per-step entries {skill, version, args, input}. For this fetcher it is a
length-1 array with skill="chirps-fetch" and input=null; downstream
zarr-writing skills append their own entry. args records the run's flag
values under underscored names (e.g. a flag --time-dim is recorded as
time_dim); version is the value printed by --help. Inspect a written
output's provenance with the provenance skill.
# Absolute window (inclusive both ends)
uv run --script ${CLAUDE_SKILL_DIR}/scripts/fetch.py --start 2026-01-01 --end 2026-02-15 --output /tmp/chirps.zarr
# Last 3 weeks ending at the newest available CHIRPS day (duration idiom: 21 days incl. latest)
uv run --script ${CLAUDE_SKILL_DIR}/scripts/fetch.py --start latest-3w --end latest --output /tmp/chirps.zarr
npx claudepluginhub rhiza-research/forecasting-skills --plugin rhiza-forecastingSearches MemPalace before answering questions about past work, people, projects, or prior decisions. Returns verbatim stored content instead of guessing from model memory.
Guides Payload CMS config (payload.config.ts), collections, fields, hooks, access control, APIs. Debugs validation errors, security, relationships, queries, transactions, hook behavior.
Implements vector databases with Pinecone, Weaviate, Qdrant, Milvus, pgvector for semantic search, RAG, recommendations, and similarity systems. Optimizes embeddings, indexing, and hybrid search.