The FDA publishes an approval package for each newly approved drug — medical review, statistical review, pharmacology review, chemistry review, label, and other documents that describe the basis for the agency's decision. The documents are public on accessdata.fda.gov, but the path from "drug name" to "the medical review that explains why this drug got approved" is several clicks of guesswork through FDA's site.

fda-analyst is a Claude Code skill that knows where those documents live, how to identify the right one for a given question, and how to read and cite the relevant section. It runs in your own Claude Code or Claude Desktop session and fetches FDA PDFs at query time — no preloaded summaries, no training-data approximations.

Install

/plugin marketplace add cbetz/trove
/plugin install trove@trove

Or copy the skill directory directly:

git clone https://github.com/cbetz/trove
cp -r trove/skills/fda-analyst ~/.claude/skills/

Example prompts

"What was the basis for Lenmeldy's approval — what trials, what endpoints?"
"What adverse events did the FDA flag in the Elahere medical review?"
"Was Krazati approved via accelerated approval? What was the pathway?"
"Compare the regulatory pathways used for Aduhelm and Leqembi."
"Which 2023 approvals used a single-arm trial as the pivotal evidence?"
"What was the primary endpoint in the Vyvgart approval?"

What this skill does

Finds the application. The skill queries trove's FDA approvals index (a Parquet bundle with 218 novel drug approvals from 2021–2024 — CDER NMEs plus CBER cell & gene therapies) to locate the application number, sponsor, and approval-package URL for the drug in question.
Picks the right document. Different questions need different documents — the medical review for "basis for approval" and "adverse events", the statistical review for endpoint analyses, the pharmacology review for ADME, the chemistry review for manufacturing. The skill knows the mapping.
Reads it at query time. No pre-extracted summaries. The skill fetches the PDF from accessdata.fda.gov, reads the relevant section, and quotes back with a citation.
Cites the source. Answers reference the specific document and page so users can verify.

What this skill doesn't do

Clinical advice or dosing. It's a reference tool over public approval documents, not a substitute for the prescribing label or clinical judgment.
Pre-2021 drugs. The current index covers 2021–2024. Earlier drugs aren't ingested yet.
Devices, biosimilars, or generic (ANDA) approvals. The trove index covers FDA's curated Novel Drug Approvals lists (NMEs and novel BLAs) only.
Real-time updates. The bundle is rebuilt on a release cycle, not continuously. Approvals within the last few weeks may not be in the current bundle.

Coverage

218 FDA novel drug approvals from 2021–2024 — 192 CDER (small molecules + most antibody biologics) plus 26 CBER (cell & gene therapies). Browse the index at /drugs/ — each drug has its own permalink page with the approval-package URL, e.g. /drugs/elahere/, /drugs/leqembi/, /drugs/lenmeldy/, /drugs/casgevy/.

Source and license

MIT-licensed. Source code, parser, raw artifacts: github.com/cbetz/trove. The data is U.S. government work, public domain — FDA Novel Drug Approvals and Drugs@FDA.