Weekly Genomics Literature Review — Claude Code Plugin
Automated pipeline that searches bioRxiv and top genomics journals, downloads full PDFs, and has Claude Code read them and produce critical reviewer-quality assessments with fine-grained scoring.
No API key needed — works entirely within Claude Code. The Python script only handles search and PDF download; Claude Code itself reads the PDFs and writes reviews.
Installation
First, install Claude Code in terminal using the following command (for details, refer to https://code.claude.com/docs/en/setup#installation)
curl -fsSL https://claude.ai/install.sh | bash
Next, start an interactive session of Claude Code
claude
For the step Select login method, Select Claude account with subscription, then it will automatically ask for account credentials and jump to the browser.
Finally, install directly from Claude Code's interactive session:
/plugin marketplace add zhongguojie1998/weekly-lit-review
/plugin install weekly-lit-review
Usage
Once installed, trigger the pipeline in any Claude Code session:
Note: It will consume tokens at the speed depending on the model you select. I recommend using /model haiku command first to set it to the lightest model before running the command.
/weekly-lit-review:weekly-lit-review --days 7
/weekly-lit-review:weekly-lit-review --max-papers 10
/weekly-lit-review:weekly-lit-review --days 3 --no-pdf
Results are saved to ~/Desktop/Claude/week-lit-review-results/.
Daily interactive mode + 12pm reminder
The interactive mode is human-in-the-loop: it fetches titles/abstracts only, shows a
numbered list of candidates, you pick which ones matter, and only the selected papers get
full-text download + deep review.
/weekly-lit-review:weekly-lit-review --interactive
Flow: query titles/abstracts → reply with your picks (1-5, 9 / all / bare DOIs) → deep
review of just those → summary.
Daily 12:00pm reminder (macOS launchd). A local agent fires every day at noon, pre-fetches
today's candidate list (metadata only), and posts a notification so the list is ready when you
sit down. Install once:
cp assets/com.zhongguojie.weekly-lit-review.plist ~/Library/LaunchAgents/
launchctl unload ~/Library/LaunchAgents/com.zhongguojie.weekly-lit-review.plist 2>/dev/null
launchctl load ~/Library/LaunchAgents/com.zhongguojie.weekly-lit-review.plist
- Force-run now:
launchctl start com.zhongguojie.weekly-lit-review
- Check status:
launchctl list | grep weekly-lit-review
- Stop/uninstall:
launchctl unload ~/Library/LaunchAgents/com.zhongguojie.weekly-lit-review.plist
- Logs:
~/Desktop/Claude/week-lit-review-results/launchd.{out,err}.log and the dated
prefetch_*.log.
The plist and scripts/daily_prefetch.sh hardcode the repo path
/Users/guojiezhong/Desktop.personal/Claude/week-lit-review. Edit both if you move the repo.
macOS app (menu bar + window)
A native front-end (app/) wraps the interactive flow so you don't touch the terminal: a
control-panel window (Fetch candidates / Select & review / Open summary / engine picker) plus a
resident 📚 menu-bar app that owns the daily 12:00 auto-fetch. Candidates persist in
candidates/ and are reused across launches; selecting papers deep-reviews only the new ones and
builds a combined summary over everything selected.
Build and install the double-clickable app (then it auto-starts at login):
bash app/build_app.sh --install # creates LitReview.app, copies to /Applications, adds a Login Item
open -a LitReview # launch now
See app/README.md for details (engine selection, dev/debug run via app/run_app.sh, limitations).
Non-interactive (from terminal, experimental)
Run directly from the command line without entering an interactive session:
# Defaults: 7 days, 80 papers, sonnet model
bash scripts/run_review.sh
# Customize
bash scripts/run_review.sh --days 3 --max-papers 10
bash scripts/run_review.sh --days 7 --model opus
bash scripts/run_review.sh --days 7 --no-pdf
Pipeline Overview
┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Read Config │────▶│ Search Papers│────▶│Download Full │────▶│Claude Reads │────▶│ Write Report │
│ (YAML) │ │ bioRxiv+RSS │ │ Text (cascade)│ │ & Review │ │ (HTML+JSON) │
└──────────────┘ └──────────────┘ └──────────────┘ └──────────────┘ └──────────────┘
Step 1: Read Configuration
- Loads
assets/config.yaml
- Configures search parameters: lookback days, max papers, categories, keywords
- Sets up journal RSS feed list and bioRxiv categories
