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
- Bhenderlong/vallorix-mcp-plugintheir own version, first seen Nov 8, 2026
- ashe-li/agent-skillstheir own version, first seen Oct 29, 2026
- jerseycheese/agent-skillstheir own version, first seen Oct 28, 2026
- Kaagemusha/agent-skillstheir own version, first seen Sep 27, 2026
- chakmarebel/federal-proposal-copilottheir own version, first seen Aug 20, 2026
- reutavidan/zetesis-plugintheir own version, first seen Aug 6, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- chakmarebel/federal-proposal-copilot★ 4same name
- jerseycheese/agent-skills★ 4same name
- ashe-li/agent-skills★ 2same name
- Bhenderlong/vallorix-mcp-plugin★ 0same name
- Kaagemusha/agent-skills★ 0same name
- reutavidan/zetesis-plugin★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository8 other files it gives its agents.
- adhd-modeSkill
- agent-operating-doctrineSkill
- clinical-change-controlSkill
- debugging-playbookSkill
- multi-agent-protocolSkill
- outbound-send-gateSkill
- portfolio-safety-checkSkill
- writing-without-ai-tellsSkill
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
- Distinctive: line 29
Use the operator's actual licensed title in any prose this check feeds.
- Distinctive: line 81
If the check feeds teaching-case content, give the board-style answer first and put label / currency nuance in parentheses.
“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.
Files
evidence-check — verify against primary sources
The second tier of a "gold sources before web" protocol. Order of checks:
- 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.
- 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.
- 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=1Fetch 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=textURL-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.exactProcedure
- State the claim to verify (one line).
- Pick the source: drug-label fact → openFDA; literature / evidence → PubMed; both if useful.
- Fetch, extracting the specific section (quote the label; list PMID titles and years).
- Return a verdict: CONFIRMED / REFUTED / NUANCED, with the primary-source quote + URL + date.
- 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.govreturns a ~420-byte stub with HTTP 200 to a defaultcurlUser-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.
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
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.

