Literature Monitor Bot — AI skill for Claude Code

**English** · [中文](README.zh-CN.md)

📚 Literature Monitor Bot

An **agent-deployed**, hands-off literature radar. It scans **bioRxiv** (twice a day), **a configurable set of journals** via NCBI (weekly), and optional **non-PubMed journals** via OAI-PMH, runs every paper through a 3-stage filter funnel, and pushes a concise, summarized digest card to a **Discord webhook** (Telegram supported too). You can then ask follow-up questions right in a chat channel — expand an entry, deep-read the full PDF, or pull a specific figure.

📡 **We ship a documented scanning strategy for each source** — bioRxiv API, PubMed-by-ISSN, OAI-PMH for non-PubMed journals, and an RSS pattern — with the gotchas and hardening each one needed. See **[docs/feed-sources.md](docs/feed-sources.md)**.

Where it lives: an always-on machine → your phone

This is designed to run on a computer that **never powers off** — an **HPC login node, a lab workstation, or a NAS**. That box quietly runs the scheduled scans around the clock, and the only thing that reaches *you* is a tidy push to **Discord or Telegram**. The point is to turn a machine you already keep running into an **AI-curated feed**: instead of checking bioRxiv and a dozen journal tables of contents yourself, the always-on host does the reading and the filtering, and delivers just the handful of papers worth your attention straight to the app on your phone — where you can also ask it to go deeper.

The reference setup runs on an **HPC via `cron → sbatch`**, so the scans keep working whether or not you have a session open. The core pipeline is plain Python + the `claude` CLI, so on a workstation or NAS you can drive `run_biorxiv.sh` / `run_journals.sh` straight from `cron` (or a `systemd` timer) and skip the SLURM layer entirely.

---

⭐ What makes this different: it's *agentic*

Most literature-alert tools make **you** hand-write a brittle list of keywords and hope for the best. This one is built to b