v0.4.4 — Paper Deduplication & Publishing Workflow

What's New

Paper Recommendation Deduplication

• Cross-day dedup by arXiv ID: Scans previous 30 days of daily recommendation files (YYYY-MM-DD论文推荐.md) and filters out already-recommended papers
• Title-based dedup for Semantic Scholar: Papers from S2 without arXiv IDs are now also caught by title matching against previous recommendations
• Confirmed fix: 03-06 and 03-07 had 100% paper overlap — this will no longer happen

Plugin Publishing Workflow

• Added publish.sh script for automated .claude-plugin/ sync, commit, and push
• Fixed .claude-plugin/plugin.json schema (no skills field — was causing install failures)
• Created claude-plugin-update fish function for reliable custom plugin updates
• Localized paper skill templates to Chinese

Changes since v0.4.0

Commit	Description
7eca361	feat: dedup S2 papers by title in addition to arXiv ID
6c3cc7c	feat: add cross-day paper dedup to paper-recommend
7829341	feat: localize paper skill templates to Chinese
a451e20	chore: automated publish workflow

Full Changelog

v0.4.0...v0.4.4

v0.4.0 — Paper Reading Skills

What's New

Added 4 paper reading skills ported from evil-read-arxiv, bringing the total to 19 skills.

New Skills

Skill	Description
paper-recommend	Daily arXiv paper recommendations with 4D scoring (relevance, recency, popularity, quality)
paper-analyze	Deep single-paper analysis — extract contributions, methods, results into structured notes
paper-images	Extract figures from arXiv papers (source package first, PDF fallback)
paper-search	Search paper notes in vault by title, author, keyword, domain, or tags

Other Changes

• Added config/research_interests.yaml for configurable research domains and keywords
• Added requirements.txt for Python dependencies (PyYAML, requests, PyMuPDF)
• Updated README with paper reading section and prerequisites

Acknowledgments

Thanks to OrbitOS and the evil-read-arxiv project for the original arXiv paper reading workflow that these skills are built upon.

v0.3.0

v0.3.0

Enhanced Skills

ai-newsletter — Rewritten with:

• RSS feed sources (TLDR AI, The Rundown AI) with DuckDuckGo fallback
• Smart deduplication (80%+ title overlap)
• Ranking by AI relevance, productivity, recency, novelty
• Structured digest with 精选推荐 content angles
• Raw feed archiving in 原始数据/ subfolder

ai-product — Rewritten with:

• Multi-source fetching (Product Hunt, Hacker News, GitHub Trending, Techmeme, Reddit)
• Engagement-normalized ranking (PH votes/500, HN points/100, GH stars/1000)
• Content angle logic (教程, 评测, 对比, 深度分析)
• Categorized digest (LLM, 开发者工具, 生产力, 开源)
• Raw data archiving per source

Both skills now save curated digests + raw data, include proper frontmatter metadata, and use Obsidian CLI for file operations.

v0.2.0

Obsidian Claude v0.2.0

Personal knowledge management plugin for Obsidian vaults with 15 skills.

Skills

Vault Organization: organize-para, organize-clippings, process-inbox, archive
Knowledge Building: wiki-builder, vault-query, spaced-review
Research: deep-research, news-research
Content & Planning: daily-plan, concept-to-action, xiaohongshu, create-base
AI Trends: ai-newsletter, ai-product

Installation

claude plugin marketplace add JeanDiable/obsidian-claude
claude plugin install obsidian-claude