Research Paper Skills

A growing collection of reusable Codex skills for academic research workflows.

The repository focuses on practical, reproducible research assistance: reading papers, producing structured summaries, organizing evidence, and synchronizing useful outputs with research tools. Each skill lives independently under skills/ and can be installed on its own.

Included Skills

paper-reading-management

Manage an end-to-end academic paper workflow across official sources, Zotero, local reading packets, and optional Feishu/Lark publishing.

The skill enforces a reproducible reading process:

1. Resolve and verify the paper from Zotero or official sources.
2. Use one clean, independent subagent per paper.
3. Produce a structured Chinese reading report with figure/table markers.
4. Organize note.md, retrieval.md, and metadata.json under the Zotero collection path.
5. Audit note structure, reading isolation, and inserted figure/table files before updating the library index.
6. Keep exactly two lightweight Zotero tags by default: one paper-kind tag and one venue/source/version tag.

Repository Layout

Research-Paper-Skills/
├── README.md
├── LICENSE
├── install.sh
└── skills/
    └── paper-reading-management/
        ├── SKILL.md
        ├── agents/
        ├── references/
        └── scripts/

Additional research skills can be added under skills/ without changing the repository structure.

Requirements

Basic use:

• Codex or another agent runtime that supports SKILL.md skills.
• Python 3.9 or newer.

Optional integrations:

• Zotero Desktop with its local API enabled for Zotero reads.
• A Zotero MCP write bridge for Zotero metadata and tag writes.
• Zotero Style to display the remark: value as 简记.
• pypdf or pdfplumber for PDF-related helpers.
• lark-cli for optional Feishu/Lark synchronization.

Default Zotero endpoints:

export ZOTERO_LOCAL_API_URL="http://127.0.0.1:23119"
export ZOTERO_MCP_URL="http://127.0.0.1:23120/mcp"

Installation

Clone the repository:

git clone https://github.com/Charles-Donne/Research-Paper-Skills.git
cd Research-Paper-Skills

Install every skill in the repository:

./install.sh

Install only the paper-reading skill:

./install.sh paper-reading-management

Replace an existing installation:

./install.sh --force paper-reading-management

Restart the agent runtime after installation so it discovers the new skills.

Paper Reading Management

Reading Note Structure

Every complete reading note follows this structure:

0. Concise Summary
   Topic · Problem · Method · Innovation · Significance
   One-sentence summary
1. Motivation
2. Innovation
3. Main Content
   3.1 Design Architecture & Methods
   3.2 Key Algorithms & Mathematical Derivations
   3.3 Models & Training & Dataset
   3.4 Experimental Setup & Results
   3.5 Technical Details
4. Significance and Impact
5. Clarifications and Simplifications
6. Additional Notes

The note is designed for two reading speeds:

• Read only Section 0 to quickly decide whether the paper is relevant.
• Read Sections 1–6 for a detailed understanding of the paper and its evidence.

Reading Packet

Each organized paper can produce:

Artifact	Purpose
note.md	Complete structured reading note covering Sections 0–6.
retrieval.md	Compact search card extracted from Section 0, plus bibliographic metadata.
metadata.json	Machine-readable metadata, core overview fields, layered tags, status, and integration identifiers.
figures/	Optional important figure and table snapshots inserted near the relevant explanations.
INDEX.md	Minimal library-level directory for navigating organized papers.

Section 0 supplies the core retrieval fields stored in retrieval.md and metadata.json. These fields also inform domain, method, and project tag planning; tags are kept separately so they remain concise and controllable.

Core Workflow

1. Resolve the paper from Zotero, a local PDF, or an official web source.
2. Verify bibliographic metadata and official project/code links.
3. Read the full paper and generate a structured Chinese note covering Sections 0–6.
4. Extract Section 0 into reusable retrieval and metadata records.
5. Optionally synchronize the one-sentence overview, lightweight tags, folders, and cloud documents.

Example prompts:

Use $paper-reading-management to read this paper and generate a structured Chinese reading report.

Use $paper-reading-management to summarize this Zotero paper and synchronize the result to Zotero.

The bundled reading prompt is written in English for reuse across models, while the generated reading report, including the one-sentence summary, is required to be Chinese.

The default local paper library is ~/paper. Override it with:

export PAPER_LIBRARY_ROOT="$HOME/paper"

Run the diagnostic:

python3 skills/paper-reading-management/scripts/diagnose.py

Zotero Style 简记

Zotero Style stores its visible 简记 value inside Zotero's Extra field:

remark: 一句话中文概览

The sync helper preserves existing non-remark: lines such as arXiv and OpenReview identifiers and does not overwrite Zotero's abstract field:

python3 skills/paper-reading-management/scripts/zotero_sync_note.py \
  --paper-folder "<paper-folder>" \
  --sync-tags \
  --yes

Adding Skills

Place each new skill in its own directory:

skills/<skill-name>/SKILL.md

Keep skill-specific scripts and references inside that directory. The root installer automatically discovers skill directories containing SKILL.md.

Security

Do not commit API keys, Zotero profiles, paper PDFs, generated reading libraries, or private research data.

License

MIT