agentleFS
MichaelRDionneMichaelRDionne/clinical-agent-skills/skills/evidence-check/SKILL.md

evidence-check

Skill0 starsChanged 4 months agoMITView on GitHub

Verify a clinical, drug, or evidence claim against primary sources — FDA labels (openFDA / DailyMed) and PubMed — when a local reference library cannot confirm it, or to cross-check a medical-GPT answer before it ships. Free, no API key, no PHI in queries.

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use the operator's actual licensed title in any prose this check feeds.

    line 29
  2. Distinctive:

    If the check feeds teaching-case content, give the board-style answer first and put label / currency nuance in parentheses.

    line 81

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

evidence-check — verify against primary sources

The second tier of a "gold sources before web" protocol. Order of checks:

  1. Local reference library first — the textbooks and manuals already on disk (psychopharmacology handbooks, DSM, board-prep manuals, specialty diagnostic texts). Cite those when they cleanly cover the fact.
  2. This skill — when the library cannot confirm it: a label / currency question (is X approved for Y? boxed-warning status?), a specialty fact outside the local books, or recent literature. Hit FDA + PubMed directly.
  3. General web search — only if neither covers it.

These are free public APIs. No key, no MCP required.

Firewall

  • No PHI in any query. Send only drug names, clinical concepts, or generic questions — never a patient identifier, name, MRN, or case detail tied to a real person. Fictional teaching cases are fine; real patients never.
  • Use the operator's actual licensed title in any prose this check feeds.
  • The public endpoints carry no PHI risk. The constraint is what you send.

The endpoints (verified fetchable)

FDA drug label — openFDA (cleanest for "what does the label say"):

https://api.fda.gov/drug/label.json?search=openfda.brand_name:<BRAND>&limit=1
https://api.fda.gov/drug/label.json?search=openfda.generic_name:<GENERIC>&limit=1

Fetch it with a prompt like: "From this FDA label JSON, quote the indications_and_usage / boxed_warning / warnings section verbatim for <drug>; is it approved as monotherapy or adjunct for <indication>?" Returns the authoritative current label.

DailyMed (alternate label source, SPLs):

https://dailymed.nlm.nih.gov/dailymed/services/v2/spls.json?drug_name=<NAME>

PubMed literature — NCBI E-utilities (two steps):

# 1) search → get PMIDs + count
https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&term=<QUERY+terms>&retmax=5&retmode=json
# 2) summaries for those PMIDs (titles/journal/year)
https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esummary.fcgi?db=pubmed&id=<PMID,PMID>&retmode=json
# (abstracts if needed)
https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pubmed&id=<PMID>&rettype=abstract&retmode=text

URL-encode spaces as +. Prefer reviews and guidelines; note the year (currency matters).

FAERS adverse-event signal (optional):

https://api.fda.gov/drug/event.json?search=patient.drug.medicinalproduct:<NAME>&count=patient.reaction.reactionmeddrapt.exact

Procedure

  1. State the claim to verify (one line).
  2. Pick the source: drug-label fact → openFDA; literature / evidence → PubMed; both if useful.
  3. Fetch, extracting the specific section (quote the label; list PMID titles and years).
  4. Return a verdict: CONFIRMED / REFUTED / NUANCED, with the primary-source quote + URL + date.
  5. If the check feeds teaching-case content, give the board-style answer first and put label / currency nuance in parentheses.

Absence traps — never conclude "no such FDA document exists" from a bare fetch

  • accessdata.fda.gov returns a ~420-byte stub with HTTP 200 to a default curl User-Agent. The 200 makes it look like a successful fetch of an empty result, and it has already produced a false "no FDA letter exists." Set a real browser User-Agent before concluding an FDA document is absent, and sanity-check the response size — a few hundred bytes is a stub, not a document. Same failure class as WAF bot-blocking on ordinary sites.
  • Verify negative findings generally. "I searched and it isn't there" is a claim that needs the same evidence as a positive one. State the query you ran and the surface you searched, or don't assert absence.

When not to use

  • A fact the local library already cleanly covers → cite the library, skip the web.
  • Anything requiring a real patient's data → stop (PHI).
  • Legal / jurisdictional questions (state duty-to-warn statutes, for example) → that is legal research, not FDA / PubMed.

Proven examples

  • Esketamine monotherapy (a 2024 handbook too old to confirm) → openFDA label: "…indicated for Treatment-resistant depression (TRD) in adults, as monotherapy or in conjunction with an oral antidepressant." CONFIRMED.
  • Levetiracetam behavioral effects → PubMed esearch returned a large, reviewable set for a teaching-case claim.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.