Paper Skills

Claude Code skills and tools for building a research second-brain. Collect papers, read them, and generate structured knowledge — all from the CLI.

Pipeline

paper-collector  -->  paper-reader  -->  paper-wiki  -->  topic-literature
  Search & download     Read & digest      Synthesize         Produce surveys,
  papers by topic       PDFs into          notes into a       briefs, reading
                        structured notes   wiki + graph       lists, gap reports
                                                                     |
                                            <----- loop-back --------+
                                           (gaps emit /paper-collector queries)

Shared workspace:

• ./reference/ — PDFs (paper-collector writes, paper-reader reads)
• ./reference/notes/ — markdown notes (paper-reader writes, paper-wiki reads)
• ./reference/index.md — catalog
• ./wiki/ — synthesized knowledge (paper-wiki owns; topic-literature reads + writes to ./wiki/reviews/)

Skills

paper-collector

Search, download, and organize research papers by topic.

/paper-collector "diffusion models for image generation"
/paper-collector "LLM reasoning" — sources: conference
/paper-collector "neural radiance fields" — sources: conference — year: 2023- — max: 20

Options:

• sources: web (default) — all available sources
• sources: conference — top-tier venues only (NeurIPS, ICLR, ICML, ACL, CVPR, etc.)
• max: N — limit number of papers (default: 10)
• year: YYYY- — filter by year

Naming convention: {year}_{short_name}_{first_author}_{venue}.pdf

• short_name is LLM-generated (e.g. bert, transformer, chain_of_thought)
• Examples: 2017_transformer_vaswani_neurips.pdf, 2020_ddpm_ho_neurips.pdf

paper-reader

Read PDFs from ./reference/ and produce structured markdown notes.

/paper-reader all
/paper-reader all — depth: skim
/paper-reader "2024_mamba_gu_icml.pdf" — depth: deep
/paper-reader "transformer" — depth: standard

Depth modes:

Depth	Pages read	Use case
skim	~5 pages	Triage — decide what's worth a deep read
standard	~15 pages	Default — build your knowledge base
deep	Entire paper	Core papers — detailed method walkthrough

Note schema (written to ./reference/notes/{short_name}.md):

• TL;DR, Problem & Motivation (5-point structure), Method, Key Results, Strengths, Limitations, Relevance, Key Takeaways
• Deep mode adds: Method Details, Experiments in Detail, Connections, Open Questions

Tools

Python CLI tools used by the skills. No external dependencies — stdlib only.

Tool	Purpose	Sources
arxiv_fetch.py	Search arXiv, download PDFs, generate filenames	arXiv API
semantic_scholar_fetch.py	Search published venue papers with citation metadata	Semantic Scholar API

Both tools output valid JSON (including on errors) and support rate limiting.

# Search
python3 tools/arxiv_fetch.py search "attention mechanism" --max 10
python3 tools/semantic_scholar_fetch.py search "LLM reasoning" --max 10

# Download
python3 tools/arxiv_fetch.py download 2301.07041 --dir ./reference/

# Generate filename
python3 tools/arxiv_fetch.py make-filename --short-name "bert" --authors "Jacob Devlin" --year 2019 --venue "NAACL"
# -> {"filename": "2019_bert_devlin_naacl.pdf"}

Directory Structure

paper_skills/
  skills/
    paper-collector/
      SKILL.md               # Skill definition
    paper-reader/
      SKILL.md               # Skill definition
    paper-wiki/
      SKILL.md               # Skill definition
    topic-literature/
      SKILL.md               # Skill definition
  tools/
    arxiv_fetch.py            # arXiv search + download + naming
    semantic_scholar_fetch.py # Semantic Scholar API search

Setup

1. Clone this repo
2. Python 3.10+ required
3. (Optional) Set SEMANTIC_SCHOLAR_API_KEY for higher rate limits
4. Copy skills and tools into your Claude Code configuration:

# Global install (available in all projects)
cp -r skills/* ~/.claude/skills/
cp -r tools/* ~/.claude/tools/

# OR project-level install (available only in that project)
cp -r skills/* /path/to/your/project/.claude/skills/
cp -r tools/* /path/to/your/project/.claude/tools/

No pip install needed — all tools use Python stdlib only.

Additional Skills

paper-wiki

Build and maintain an interlinked research wiki from paper notes. Concept pages synthesize across papers; a JSONL graph tracks relationships.

/paper-wiki init          # bootstrap ./wiki/
/paper-wiki build         # first full build from ./reference/notes/
/paper-wiki update        # incremental — integrate newly read papers
/paper-wiki query "..."   # ask the wiki; optionally --save the answer
/paper-wiki lint          # health-check for orphans, thin concepts, broken links
/paper-wiki status        # quick state summary

topic-literature

Produce topic-level deliverables from the wiki.

/topic-literature survey "<topic>"         # comprehensive field survey
/topic-literature brief "<question>"       # focused 1-2 page synthesis
/topic-literature reading-list "<topic>"   # curated reading order
/topic-literature related-work "<focus>"   # academic-format related work section
/topic-literature gaps                     # research gap report + paper-collector queries

The gaps subcommand emits copy-pasteable /paper-collector queries, closing the pipeline loop.

Planned Skills

• graph-creator — Render wiki/graph/ via Graphify / Obsidian graph view for visual exploration