Prisma search

Skill Hash-7777/Hash-Medical-Reasearch-Agent-Skills/skills/prisma-search

Drop-in agent skills that make an AI appraise medical literature like a reviewer — grade every citation against its source, run reproducible PRISMA searches, pool studies safely, and defend RAG against prompt injection. Framework-agnostic, MIT.

Install

npx -y skills add Hash-7777/Hash-Medical-Reasearch-Agent-Skills --skill prisma-search

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

23 days oldThe repository was created 23 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Run a literature search that another person could reproduce — explicit query, named databases, retrieval date, inclusion/exclusion criteria, and a PRISMA-style count of what was found, screened, and kept. Use whenever an agent gathers studies to answer a medical or scientific question.

SKILL.md

3.2 KB, as published. Nobody here has run it

PRISMA Search

"I found some studies" is not a search — it is a claim with no way to check it. A reproducible search states exactly what was asked, where, when, and what was kept or dropped and why. Anyone repeating it should land on the same set. This skill makes an agent's evidence-gathering auditable the way a systematic review is.

When to use

Whenever the agent retrieves a body of literature to ground an answer — a drug question, an association question, a "what does the evidence say" question. Even a lightweight search benefits from stating its terms.

The rule

A search is only as good as its ability to be repeated. Record the five things that make it repeatable.

The query — the actual search string(s), with the boolean/field structure, not a paraphrase of the topic.
The databases — which sources were searched (e.g. PubMed, Europe PMC, a trials registry, GBIF-style records), by name.
The retrieval date — the real date the search ran. Evidence changes; a search without a date cannot be reproduced or trusted later. Stamp the true date, never a placeholder.
Inclusion / exclusion criteria — what qualified a record (study type, population, outcome, language, date range) and what disqualified it.
The count — how many records were found, screened, excluded (with reasons), and included. This is the PRISMA flow.

The PRISMA flow to report

Identified      (n = …)   records from the named databases
   ↓ remove duplicates
Screened        (n = …)   on title/abstract
   ↓ exclude with reasons
Assessed        (n = …)   full text / full record
   ↓ exclude with reasons (list them)
Included        (n = …)   in the answer

How to do it well

Do not let the harvester grab the wrong tokens. A gene name, a drug class, or a framing word is not a study. Only real records enter the count. (A search that "found 40 trials" where 12 were actually the query terms echoed back is a broken search.)
Separate the query from the question. The user's phrasing is a starting point, not the search string. Build explicit terms — synonyms, brand/ingredient families, MeSH-style concepts — and record them.
State coverage honestly. If the search is narrow, say so. A reproducible small search beats an impressive-sounding one nobody can rerun.
Never fabricate the count or the date. If you did not run the search, you do not have a PRISMA flow — say that instead of inventing one.

Output

Lead the evidence section with the reproducibility block:

Query:      ("drug A"[tiab] OR "brand A"[tiab]) AND ("outcome X"[tiab]) AND randomized
Databases:  PubMed, Europe PMC
Date:       2026-07-14
Include:    RCTs in adults, outcome X reported
Exclude:    reviews, animal studies, n<20
PRISMA:     identified 214 → screened 190 → assessed 22 → included 9

If a reader cannot rerun your search from this block, it is not finished.