EEB Daily is a one-page newsletter that Claude writes fresh every night for the Blackmon lab: one featured paper with author bios and a critical note, a second paper from outside the lab's field or a classic revisited, and a story from the wider world of science. It costs nothing to read and little to run. This page gives you everything needed to make one for your own lab, with your topics, your people, and your schedule. Nothing here is specific to us except the examples.
A scheduled task runs each night. It reads a short config document you control (topics, people whose papers get priority, slot rules), searches PubMed and the web for the last few days of papers, chooses one feature and two smaller items, researches the first and last authors, writes the issue as a structured JSON record, runs a small build script that renders the page, and commits the result to a GitHub Pages repository. A stable URL always shows the latest issue; an archive page lists every issue; a QR code pointing at the latest page can be printed once and posted wherever you read.
tools/build.py renders the pages from the JSON records; tools/RECORD.md documents the record format the task writes. Edit the SITE constant at the top of build.py to your URL.build.py · RECORD.md · example issue records
build.py and RECORD.md into [folder]/tools/ in your repository, set SITE, and commit.https://[your-site]/[folder]/ and tape it where you drink your coffee.[LAB] DAILY CONFIG == READERS == Who reads this and at what level. Example: a lab of evolutionary biologists, graduate students and undergraduates; write so a sharp undergraduate can follow it; define one or two pieces of jargon inline; nothing addressed to one person. == SLOT 1: FEATURED PAPER TOPICS (priority order) == 1. [TOPIC A] 2. [TOPIC B] 3. [TOPIC C] 4. Any striking paper in [YOUR FIELD] Rotate across topics; do not feature the same topic as either of the previous two issues. Preferred journals (not exclusive): [LIST] == PEOPLE WE FOLLOW == [Full Name]; [Full Name]; ... Rule: any author with a [YOUR INSTITUTION] affiliation also gets priority. Match on full names, never surnames alone. The priority is silent: no tag or mention in the newsletter. == AUTHOR BIOS == About 50 words each on the first and last author: position, institution, training lineage, what they are known for. Public professional information only. == SLOT 2: SECOND PAPER == BEYOND (default): a striking recent paper in [FIELDS OUTSIDE YOURS] from the last three days. CLASSIC (on [TWO WEEKDAYS]): a landmark paper from the past in [YOUR FIELD]: what it showed, how, why it still matters, one sentence on what came after. Never repeat a classic. == SLOT 3: STORY == SCIENCE (default): a big result from outside [YOUR DISCIPLINE] in the last week. HISTORY (on [ONE WEEKDAY]): a short feature from the history of science: a mini-biography, a famous debate, an expedition, or a controversy explained evenhandedly. Subjects no longer living; nothing negative about any living person. Never repeat a subject. ACADEMIA (rare, at most once a week, only when genuinely notable): research and academia news relevant to [YOUR ROLES]. Add a "Why it matters" line. == EXCLUSIONS == Never repeat a paper, story, or history subject already in the archive. Skip predatory journals. [ANY OTHER EXCLUSIONS] == FORMAT == Confident, direct prose. No hedging filler. Every item links to its primary source. [PUNCTUATION OR VOICE RULES]
You are producing the next morning's issue of "[LAB] Daily", a one-page newsletter for [LAB NAME] ([INSTITUTION]). This run happens the evening before, so the ISSUE DATE is tomorrow's calendar date; use it for the masthead, file names, and the weekday-based slot rules. Use the real date only for search windows. This run is unattended: never ask questions, make reasonable calls and note them in one line at the end. Keep the run under 35 tool calls. Follow the style rules in the config. STEP 0. Get the current date and time ([TIMEZONE]); compute the issue date and weekday. Load via ToolSearch: SendUserMessage, mcp__Google_Drive__read_file_content, mcp__PubMed__search_articles, mcp__PubMed__get_article_metadata, mcp__claude-code-remote__add_repo. STEP 1. Read the config: Google Drive doc id [CONFIG_DOC_ID]. Follow it. STEP 2. Attach and clone the site repository: mcp__claude-code-remote__add_repo with owner [GITHUB_OWNER], repo [REPO], access push; then git clone --depth 1 https://github.com/[GITHUB_OWNER]/[REPO] [CLONE_DIR] (if that directory already holds a working clone, use it). Read [CLONE_DIR]/[FOLDER]/tools/RECORD.md for the record format and the dedup queries, and run those queries with jq over [FOLDER]/data/*.json. Collect every featured DOI, item URL, classic, and history subject already used. Note the topics of the two most recent issues. The issue number is the highest existing number plus one. Decide slot kinds from the issue date's weekday per the config. STEP 3. Slot 1, the featured paper. Search PubMed (datetype edat, last three days, sort pub_date, max_results 40) with a query built from the config topics; also one web search for new preprints. Pull metadata (if saved to a file, list doi, title, journal, date, authors and affiliations with jq). Choose by priority: (1) an author on the people list (full-name match) or at your institution; (2) topic fit, rotating, never the topic of either of the previous two issues; (3) journal quality and how interesting the result is. Never a DOI already used. Research the first and last authors (lab page, Scholar, ORCID) and write about 50 words each. Write a short declarative headline (not the paper title), a one-sentence deck, two paragraphs totaling 170-210 words (the question and why the field cares; the approach and key result), and a "Where to push" box of 45-60 words with one critical observation. STEP 4. Slot 2 per the config: BEYOND from the PubMed results plus one more search of the glossies, or CLASSIC with one web search to confirm the citation and a stable link. Headline, one-line citation, 85-110 words (a classic adds one sentence on what came after), link. STEP 5. Slot 3 per the config: SCIENCE (web search, fetch one or two major science news pages), HISTORY (one or two searches plus one reliable source), or ACADEMIA (web search). Headline, one-line citation, 85-110 words (HISTORY 100-130; ACADEMIA 70-90 plus a "Why it matters" line of 30-45 words), link. STEP 6. Write the record to [CLONE_DIR]/[FOLDER]/data/NNNN-YYYY-MM-DD.json exactly in the RECORD.md format. Validate it with python3 -c "import json,sys; json.load(open(sys.argv[1]))". Run python3 [FOLDER]/tools/build.py from the repo root; its output names the latest issue; confirm it is this one. STEP 7. In the clone: git config user.name "[LAB] Daily"; git config user.email "[EMAIL]"; git add [FOLDER]; git commit -m "[agent] [LAB] Daily No. N (YYYY-MM-DD)"; git fetch origin [BRANCH]; git rebase origin/[BRANCH]; git push origin HEAD:[BRANCH]. If the push fails, retry once after 30 seconds; if it still fails, publish [FOLDER]/issues/YYYY-MM-DD.html with the Artifact tool and report the failure. STEP 8. Finish with at most four lines (send them with SendUserMessage too): the issue URL https://[your-site]/[FOLDER]/issues/YYYY-MM-DD.html; the featured paper's first and last author and journal with one phrase on the result; the slot 2 and slot 3 headlines with their kinds; any judgment call. Do not describe your process.