🔬 Scientific Research Skills

High-level research methodology skills for AI coding agents.

Not another tool list. Each skill encodes a research workflow — the kind of domain expertise that takes months to develop. Skills are written for AI agents to read and follow; this README is written for humans to configure and choose.

中文版本 (Chinese)

Quick Start

For AI Agents

If a user sends you this repo link, follow this onboarding flow. Do not skip the post-install configuration step — many skills are non-functional without API keys, and silently installed skills are worse than uninstalled ones.

Step 1 — Introduce the skills

Read this README, then present each skill’s name and one-line description to the user. Group them by category (🔧 tool-integrated vs 📋 methodology-only) so the user understands which ones will require configuration.

Step 2 — Confirm what to install

Ask: “Which skills would you like to install? (all / select specific ones)”

Step 3 — Install

Copy the selected skill folders to the appropriate location for your platform:

• OpenClaw: ~/.openclaw/skills/
• Claude Code: ~/.claude/skills/ (user-level) or .claude/skills/ (project-level)
• Codex: ~/.agents/skills/ (user-level) or .agents/skills/ (project-level)

Step 4 — ⚠️ Post-install configuration (do not skip)

After installation, explicitly notify the user which installed skills require configuration before they will work. Use this template:

✅ Installed: skill-a, skill-b, skill-c

⚠️ Heads up — the following skills need configuration before use:

• skill-a requires: TAVILY_API_KEY (free tier available)
• skill-c requires: Zotero API key + User ID

Want me to walk you through setting these up now? I can do them one by one.

Then, for each skill the user agrees to configure, walk them through it interactively:

1. Check existing config first — Run echo $TAVILY_API_KEY (or equivalent) to see if it’s already set. If yes, skip to verification.
2. Explain what the key is for — One sentence on what the skill will use it for.
3. Provide the signup link — Give the direct URL (see Dependencies below). Wait for the user to obtain the key.
4. Help them set it persistently — Append the export line to their shell profile (~/.zshrc for zsh, ~/.bashrc for bash). Don’t just export in the current shell — that won’t survive a restart.
5. Verify — Run a quick test (e.g., a 1-result search) to confirm the key works. If it fails, debug before moving on.
6. Move to the next skill — Repeat until all configurable skills are set up.

Per-skill configuration cheat sheet:

Step 5 — Confirm completion

After configuration, summarize: which skills are fully working, which are installed but waiting on config the user wants to defer, and how to invoke each one in their agent.

Configuration etiquette

• Never assume keys are set. Always check.
• Never write secrets to git-tracked files (e.g., project-level .env without .gitignore).
• Don’t ask for keys all at once. Walk through one skill at a time so the user isn’t overwhelmed.
• If the user wants to skip a config, install the skill anyway but mark it as “⚠️ installed but not configured” so they remember later.

For Humans

1. Browse the Skills table below
2. Pick the skills relevant to your research workflow
3. Follow the Installation guide for your agent platform
4. Configure any required API keys and dependencies

---

Skills

Skill categories

• 🔧 Tool-integrated — require external APIs or tools: literature-search, social-media-paper-triage, zotero-management, academic-figure-generation, paper-fulltext-harvest
• 📋 Methodology-only — pure workflow guidance, no dependencies: paper-reading, related-work-survey

---

Installation

Each skill follows the open agent skills standard: a folder with a SKILL.md (YAML frontmatter + markdown instructions) plus optional scripts/, references/, and assets/ directories. This format is natively supported by OpenClaw, Claude Code, and Codex.

Install all skills

# Clone the repo
git clone https://github.com/jxtse/scientific-research-skills.git
cd scientific-research-skills

# OpenClaw
cp -r skills/* ~/.openclaw/skills/

# Claude Code (user-level, available in all projects)
cp -r skills/* ~/.claude/skills/

# Codex (user-level, available in all repos)
cp -r skills/* ~/.agents/skills/

Install specific skills

# Example: install only literature-search and paper-reading

# OpenClaw
cp -r skills/literature-search ~/.openclaw/skills/
cp -r skills/paper-reading ~/.openclaw/skills/

# Claude Code
cp -r skills/literature-search ~/.claude/skills/
cp -r skills/paper-reading ~/.claude/skills/

# Codex
cp -r skills/literature-search ~/.agents/skills/
cp -r skills/paper-reading ~/.agents/skills/

Project-level vs user-level

Skills are auto-discovered by all three platforms. No restart needed for OpenClaw; restart Claude Code or Codex if a new skill doesn’t appear.

Other agents

Each SKILL.md is self-contained markdown. For agents that don’t support the skills standard, feed the SKILL.md content as system prompt or context directly.

---

Dependencies

Search Engines (for literature-search)

You don’t need all of them. Pick based on your needs:

Minimum recommended: Semantic Scholar (free) + arXiv (free) + one of Tavily/Exa.

# Set API keys in your shell profile (~/.zshrc or ~/.bashrc)
export TAVILY_API_KEY="tvly-..."
export EXA_API_KEY="..."
export GEMINI_API_KEY="..."     # Optional: for deep research
export AMINER_API_KEY="..."     # Optional: for Chinese academic search

Social Media Reading (for social-media-paper-triage)

Minimum recommended: Jina Reader (zero config, works for most URLs).

Zotero (for zotero-management)

1. Install Zotero
2. Create API key: zotero.org/settings/keys
3. Find your User ID: shown on the same settings page

export ZOTERO_API_KEY="..."
export ZOTERO_USER_ID="..."

Academic Figure Generation (for academic-figure-generation)

Requires PaperBanana local deployment. See the skill’s SKILL.md for setup details.

Publisher TDM Keys (for paper-fulltext-harvest)

All keys are optional — the skill works in OA-only mode with just UNPAYWALL_EMAIL set. Configure these to unlock paywalled content via institutional TDM (Text and Data Mining) APIs:

For paywalled publishers without TDM (ACS, RSC, T&F, many Chinese journals), the skill includes a browser fallback that drives the user’s logged-in Chrome via OpenClaw’s browser tool with profile="user".

export ELSEVIER_API_KEY="..."
export ELSEVIER_INSTTOKEN="..."
export WILEY_TDM_TOKEN="..."
export UNPAYWALL_EMAIL="[email protected]"

Critical: TDM APIs require the request to come from an institution-allowlisted IP — you must be on the institutional network or VPN.

---

Philosophy

Why this repo exists:

Existing “AI for research” repos (like claude-scientific-skills) catalog hundreds of specific tools. Most are irrelevant to any given researcher, and they describe what a tool does, not when or why to use it.

This repo encodes research methodology — the decision-making process that experienced researchers internalize over years:

• When to skim vs. deep-read a paper
• How to systematically survey a field and find your positioning
• Which search engine to use for which type of query
• How to go from a vague observation to a concrete research contribution

Each skill is a workflow, not a function call.

Design principles:

1. For AI, not for show — Skills are written as agent instructions, not human documentation
2. High-level over low-level — Methodology over tool invocation
3. Modular — Pick only what you need
4. Cross-platform — Same SKILL.md format works with OpenClaw, Claude Code, Codex, and any agent

---

Contributing

This is a living repo. Skills are added as research workflows mature.

To contribute a new skill:

1. Create skills//SKILL.md
2. Use the existing skills as format reference (YAML frontmatter + markdown body)
3. Focus on when and why, not just how
4. Optionally add scripts/, references/, or assets/ directories
5. PR with a one-paragraph description of the workflow it encodes

License

MIT