First release of the arxiv-fetch agent skill: fetch the latest arXiv papers and generate an on-demand digest, with the agent judging relevance itself — no API keys, no dependencies, no repo clone needed.
Highlights
- Stdlib-only fetch script (
skills/arxiv-fetch/scripts/fetch_arxiv.py): concurrent feeds with retry, cross-feed dedup, explicit failure reporting (stats.failed_subjects, nonzero exit when all feeds fail) - Config search chain: request overrides → project (
.arxiv-fetch/config.json) → install (skill dir) → user-global (~/.config/arxiv-fetch/config.json, survives plugin updates) - Guided first-time setup and agent-managed config (show/edit/move/reset)
- Complete arXiv category taxonomy bundled (155 categories) for topic→subject inference
- Repo doubles as a Claude Code plugin marketplace
Install
/plugin marketplace add yang3kc/daily_arxiv_digest
/plugin install arxiv-fetch@daily-arxiv-digest
Other install routes (copy, clone+symlink, skills CLI): see the README.