version: "1.0.0" name: drought-anatomy description: "Assemble a drought view as a set of receipted panels, each labelled with its product, its run or version and its own refusal: precipitation, soil moisture, streamflow and storage as the quantitative panels, snow as a qualitative one. Carries no number of its own and never reduces the panels to a single index."

drought-anatomy

Build the picture of a drought from several products at once, without letting any one of them borrow another's validity. Works by slash command or conversationally ("show me the drought anatomy for this basin, water year 2021 against 2023").

Behavior, in order

1. Parse and show back: the basin, the two periods being compared,

and which panels the request implies.

1. Check the basin against the storage footprint floor first. The

storage panel comes from mascons whose native scale is about one mascon, and a basin below that floor has no storage signal of its own. Say so before drawing anything: a four-panel view with a stated refusal is the honest answer, and a four-panel view with an unexplained gap is not.

1. Consult the bundle for each panel. Consult installed knowledge

concepts first, as the core consult-knowledge skill sets out. Each panel is governed by its own concepts and they do not transfer: the precipitation run rule, the evapotranspiration fill classes, the soil moisture retrieval's coverage and its short record, the regulated-gauge caution, the mascon footprint and lag, and the snow product's model-output status. Restate what applies per panel and cite each by path.

1. Assemble panels, not an index. Build one period's receipt

through uv run ${CLAUDE_PLUGIN_ROOT}/skills/drought-anatomy/scripts/load_drought_panels.py (the script ships with this plugin beside this file) with --label, --basin, --precipitation-receipt (a receipt from the load-precipitation skill's script, so the run is already declared there), --soil-moisture, --streamflow, --storage, --snow, --regulated where the outlet is, and --out; one period per call, so the tool never sees two periods at once and cannot reduce them. Every panel carries its product, its run or version, its period, the count behind each number, and the one sentence that says what it cannot show. Panels are shown side by side and are never averaged, ranked together, or reduced to a single drought number: they measure different quantities with different validity and a composite index would hide exactly the disagreements that make the view worth having.

1. Say what each panel cannot show, in the panel. A regulated

outlet is an operations record rather than a runoff signal. A soil moisture mean over a winter day rests on the cells that were retrieved, which in a snowy basin can be a small fraction of the basin. A storage panel stops where the record stops. A snow panel is a model output and a comparison, never a term.

1. Summary as provenance: basin with its area and its standing

against the floor, the periods, each panel with its product and version, the fixtures or receipts behind it with their hashes, the concepts consulted, and every refusal that fired.

Must NOT

• Never reduce the panels to a single index, score or ranking.
• Never draw a storage panel for a basin below the footprint floor,

and never extend one past the record's last epoch.

• Never put a snow number into a water budget, or offer one as the

explanation of a residual without labelling it a model output outside the identity.

• Never compare a precipitation panel's present against a climatology

built from a different run.

• Never report a soil moisture mean without the count of cells behind

it, and never read a fall in that mean across a freeze as drying.

• Never read a flat streamflow panel at a regulated gauge as evidence

that a drought is absent or over.

• Never restate a product rule this skill could consult: every rule

above lives in a concept and is read from it per run, which is what lets a corrected concept change this skill's behavior without an edit here.

The shape of a good answer

The panels disagree about how large the drought is, because they measure different things over different depths and times. That disagreement is the finding, not a defect to be reconciled. A view whose panels all agree is usually a view whose panels were drawn from one product.