feynman-health: forking an AI research agent for public health and HPSR
feynman-health-skills is now live. Skills for PubMed and HPSR epistemology are ready to install. See the GitHub repo for installation and usage.
This started as a fork experiment. It’s become a working tool. HPSR researchers: try it and report back.
Prerequisites
You need Feynman installed first. See Feynman’s installation guide.
Once Feynman is installed, proceed below.
The problem
Feynman is a good AI research agent — searches, reads, synthesises, drafts. But its entire source layer runs through AlphaXiv (arXiv, basically). So it knows comp sci, physics, quantitative biology. It does not know whether PubMed exists.
For people doing health policy and systems research, epidemiology, qualitative work, this is unusable. The skills here fix that.
What I built
Two skills, tested and ready
pubmed-research.skill — Tells Feynman to query PubMed first. Teaches MeSH term logic, study design appraisal, LMIC relevance weighting.
Test run: ASHA programme effectiveness — pulled 19 sources (all appropriate: PubMed/PMC/grey, zero arXiv), correctly separated pilot evidence from routine programme findings. Provenance record archived.
hpsr-epistemology.skill — Encodes that qualitative studies, mixed-methods designs, implementation science papers are first-class evidence. Grey literature (WHO, government docs, NHSRC) is primary, not supplementary. Context-sensitivity is a feature.
What’s NOT here yet
- Subscription databases (Embase, Web of Science, CINAHL, Ovid MEDLINE) — out of scope. Free APIs only.
- PubMed CLI connector — this is the next big ask. Feynman ships with
alphafor AlphaXiv; we need the equivalent for PubMed. Needs a developer. - New workflows (
/sysrev,/policybrief,/appraise) — code work, separate from skills.
Install and use
Clone the repo and copy skill files into Feynman’s skills directory:
git clone https://github.com/daktre/feynman-health-skills.git
cp feynman-health-skills/*.skill ~/.feynman/agent/skills/Then run Feynman as usual. Skills auto-activate for relevant queries.
feynman lit "your research question"
feynman deepresearch "your topic"Full details: GitHub README.
Roadmap
Short-term (planned skills)
grey-literature.skill— explicitly prioritise WHO IRIS, MoHFW, NHSRC, NFHS, parliamentary reportsreviewer-appraisal.skill— CASP checklists, Cochrane risk-of-bias, GRADE framing
Medium-term (needs developer)
- PubMed CLI connector — wraps NCBI E-utilities. This unblocks everything.
- WHO IRIS connector (OAI-PMH)
- Cochrane REST API wrapper
Longer-term
- Local document ingestion (curated Indian health policy materials)
- New workflows
For HPSR researchers
Won’t replace protocol-driven systematic reviews (you need Embase, Web of Science, proper screening — use Covidence or Rayyan). But useful for:
- Rapid evidence reviews
- Scoping review preliminary searches
- Grant background sections
- Understanding what’s published on a health systems topic
- Separating pilot/special intervention evidence from routine programme findings
Feedback wanted
From HPSR practitioners: Does this help? What’s missing? What’s wrong?
From developers: PubMed CLI — interested?
GitHub: daktre/feynman-health-skills
Email: [email protected]
Built at IPH Bengaluru as part of broader work on AI tooling for health systems research.
See also
Original planning (archived for reference)
The thinking behind this fork, kept here for context and traceability.
The fork approach
Feynman’s agent behaviour runs on skills — Markdown instruction files in ~/.feynman/agent/skills/. This makes it fork-friendly without touching TypeScript.
The harder part is source access. AlphaXiv gives full-text access, citation chains, Q&A on papers. We need to rebuild that pointing at PubMed, WHO IRIS, Cochrane.
Planned skills (original list)
A grey-literature.md skill telling the Researcher to treat WHO reports, MoHFW documents, NHSRC publications, state NHP materials, parliamentary standing committee reports, IIPS data as primary sources.
A reviewer-appraisal.md skill that replaces default peer review criteria with CASP checklists for qualitative work, Cochrane risk-of-bias logic for trials, and GRADE framing for evidence synthesis.
A policy-writer.md skill teaching the Writer agent what a policy brief looks like, how a rapid evidence review is structured, how HPSR synthesis papers differ from standard journal articles.
Planned workflows (original list)
/sysrev for systematic review support — PICO formulation, PRISMA-compatible search string generation for PubMed and Cochrane, deduplication, PRISMA flow outline.
/policybrief to synthesise evidence into a structured brief — problem statement, evidence summary, policy options, implementation considerations.
/appraise applying CASP, GRADE, or ROBINS-I to a supplied document.
/burden to aggregate epidemiological data from GBD, NFHS, DLHS, and state health bulletins.
/greylit explicitly instructing the Researcher to prioritise institutional and government sources.
/watch (already exists) — monitoring new WHO guideline updates, Lancet publications on a topic, MoHFW policy notifications.
Source connectors (original spec)
The key connector is a pubmed CLI tool wrapping NCBI’s E-utilities (esearch + efetch). Needs to: search by keyword, MeSH term, date range; retrieve full metadata; pull abstracts and PMIDs.
After that: WHO IRIS connector (OAI-PMH), Cochrane REST API for abstracts and Plain Language Summaries, curated India-specific URL corpus (NFHS, HMIS, NHSRC, MoHFW).
For India-specific content, the realistic approach is a curated URL list that the web search agent is explicitly instructed to prioritise: NFHS, HMIS, NHSRC reports, NITI Aayog health chapters, SRS data. Plus a local PDF ingestion path for frequently-used institutional documents.