Weekly PubMed research digest to a Slack channel

Every Monday at 8am (in my local timezone), run a weekly PubMed research digest and post it to a Slack channel I configure. The research topic should be a configurable input — start with the example query 'GLP-1 receptor agonists AND obesity' but make it easy to change to anything (for example, 'CAR-T cell therapy AND lymphoma' or 'transformer architecture[tiab] AND medical imaging[tiab])'.

Here is the flow I want the agent to run each Monday:

1. Compute the date range for the last 7 days (yesterday minus 7 to yesterday, in YYYY/MM/DD format), and build a PubMed query like: '<my topic> AND ("YYYY/MM/DD"[edat] : "YYYY/MM/DD"[edat])'. Use the [edat] (Entrez date) tag — it's the date PubMed indexed the paper, which is what 'new this week' actually means.

2. Call PubMed ESearch against the pubmed database with that query, retmax=50, retmode=json. Always pass api_key and email on every PubMed call (the credential provides both). Sort by most recent. Collect the returned PMIDs.

3. Before doing any work on those PMIDs, dedupe against what's already been posted. Pull recent channel history from the target Slack channel (last ~30 days, paginate as needed), extract any PMIDs that appear in previously-posted messages or links (pubmed.ncbi.nlm.nih.gov/<PMID>), and drop those PMIDs from the candidate list. The goal: the same paper should never appear in the channel twice in a given month.

4. Call PubMed ESummary on the remaining PMIDs (retmode=json) to get title, authors, journal (full + ISO abbreviation), publication date, and DOI. Then call PubMed EFetch on those same PMIDs (db=pubmed, rettype=abstract, retmode=xml) to get the structured abstracts.

5. Have the agent read the titles + abstracts and pick the top 5 to 8 most clinically significant or novel papers. Ranking should favor: large RCTs over small observational studies, novel mechanisms or first-in-class results, papers in high-impact journals (NEJM, Lancet, JAMA, Nature, Science, Cell), and papers that change practice or report unexpected findings. If fewer than 5 papers match the query that week, include all of them. If zero match, post a short note saying 'No new papers matched <query> this week' so I know the workflow ran.

6. For each selected paper, write a 2 to 3 sentence plain-English summary that covers: study design (RCT, cohort, meta-analysis, in vitro, etc.), sample size or n, and the key finding. Keep it tight — this is a scanner-friendly digest, not a review article.

7. Post a single Slack message to the configured channel using the Slack Bot 'Send a Message' action. Format it as one message with a header (something like ':microscope: PubMed weekly digest — <query> — week of <date range>') and one block per paper in this shape:

*<Paper title>*\n_<First author> et al. · <Journal abbreviation> · <pub date>_\n<2-3 sentence summary>\n<https://pubmed.ncbi.nlm.nih.gov/PMID|Read on PubMed>

Use Slack mrkdwn formatting (*bold*, _italic_, <url|text> for links), not standard Markdown. Keep the whole thing in one message so the channel stays clean — if it would exceed Slack's character limit, split into a parent message + threaded replies rather than multiple top-level messages.

Configurable inputs the workflow should expose: the PubMed query (string), the Slack channel ID, the day of week + time + timezone, and how many papers to include (default 5–8).

Use the 'pubmed' integration for ESearch/ESummary/EFetch and the 'slackbot' integration for both reading channel history and sending the message.