From trove
Answer questions about U.S. nonprofit hospital financials, beds, staffing, charity care, and community benefit by querying CMS Medicare Cost Reports (HCRIS) and IRS Form 990 Schedule H filings. Use when the user asks about a specific hospital's cost-report numbers, peer context across hospitals, the meaning of a worksheet/line/column code, the difference between related CMS and IRS charity-care reporting fields, or any analytical question over the trove project's published HCRIS+990 data bundles.
How this skill is triggered — by the user, by Claude, or both
Slash command
/trove:hcris-analystThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
A skill for analyzing U.S. nonprofit hospital financials. Powered by the [trove project](https://github.com/cbetz/trove)'s parsed HCRIS Hospital 2552-10 data and IRS 990 Schedule H filings, published as Parquet bundles at troveproject.com.
A skill for analyzing U.S. nonprofit hospital financials. Powered by the trove project's parsed HCRIS Hospital 2552-10 data and IRS 990 Schedule H filings, published as Parquet bundles at troveproject.com.
Invoke when the user asks about:
The data lives as Parquet bundles served over HTTPS from troveproject.com. DuckDB can query them directly. Two execution paths:
duckdb -c "SELECT ... FROM read_parquet('https://troveproject.com/data/...')"import duckdb; duckdb.sql("...").df()Both work in any environment with bash / Python and DuckDB available. If the user has the trove repo cloned and uv sync --all-packages has been run, the Python packages are also importable directly (from hcris import pivot_wide, etc.) — that's faster for repeated queries but not required.
Reference docs (read these before answering):
references/parquet_layout.md — bundle URLs, schemas, common query patterns.references/dictionary.md — every HCRIS variable the bundles expose, with worksheet/line/column source.references/schedule_h.md — IRS 990 Schedule H field map.references/cohorts.md — peer cohort dimensions and useful canned cohorts.references/examples.md — runnable example queries that pattern-match the most common question types.charity_care_cost is HCRIS Worksheet S-10 line 23 column 3, not the gross charges line 22 column 1. Always look up the right column.|hcris_fy_end_dt - sched_h_tax_period_end| ≤ 1 month. ~28% of EIN-matched systems pass this filter; the rest are misaligned and should not be treated as direct like-period views.charity_gap has many legitimate explanations (definitional differences in what each form requires; cost-to-charge ratio handling; charity vs. bad debt classification; non-Medicare patient mix). Explain that the two fields are related but non-identical reporting measures, and that a nonzero difference does not by itself say whether either filing is wrong.hcris_2023_wide.parquet").When you produce output that quotes data values, cite the source bundle.
Note on ADI: the public bundles do not include Area Deprivation Index (ADI) columns. UW's Neighborhood Atlas terms are non-sublicensable. Users who download the trove repo and the raw ADI data themselves (with a UW account) get ADI in their local matched hospital parquet via packages/sdoh/, but the public skill working over troveproject.com bundles only has access to SVI.
v1.1, May 2026 (SVI added). Bug reports and questions: github.com/cbetz/trove/issues.
Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub cbetz/trove --plugin trove