• 2026-03-05 — 100 GitHub Stars! Thank you to everyone who has supported LangSkills — your encouragement keeps us going!
• 2026-03-04 — v0.1.0 published to PyPI; skill bundles hosted on Hugging Face with China mirror support
• 2026-03-15 — v0.1.1: 119,608 skills across 21 domain bundles — added 32K+ journal skills, cleaned ghost entries
• 2026-02-28 — v0.1.0: 101,330 skills across 21 domain bundles officially released
• 2026-02-27 — Pre-built SQLite bundles with FTS5 full-text search ready for download
• 2026-02-27 — Journal pipeline online: PMC, PLOS, Nature, eLife, arXiv full coverage

• 📚 Massive Pre-Built Skill Library: 119,608 evidence-backed skills covering 95K+ research papers and 24K+ coding/tech sources — all searchable offline via FTS5-powered SQLite bundles.

• 🔧 Fully Automated Skill Pipeline: Give it a topic → it discovers sources → fetches & extracts text → generates skills with an LLM → validates quality → publishes. One command, zero manual work.

• 🔬 Evidence-First, Never Hallucination-Only: Every skill traces back to real web pages, academic papers, or code repositories with full provenance chains — metadata, quality scores, and source links included.

• 🌐 Multi-Source Intelligence: Integrates Tavily, GitHub, Baidu, Zhihu, XHS, StackOverflow, arXiv, PMC, PLOS, Nature, eLife — 10+ data source providers for comprehensive coverage.

• 🧠 LLM-Powered Quality Gates: Each skill is generated, validated, and scored by LLMs with configurable quality thresholds — ensuring high-signal, low-noise output at scale.

• ⚡ Drop-In Reusability: Download domain-specific SQLite bundles, skill-search any keyword, and get structured Markdown ready to feed into any AI agent, RAG pipeline, or knowledge base.

• 🏗️ Extensible Architecture: Modular source providers, LLM backends (OpenAI / Ollama), queue-based batch processing, and configurable domain rules — built to scale.

• 📦 21 Domain Bundles: From Linux sysadmin to PLOS biology, from web development to machine learning — organized, versioned, and individually installable.

pip install langskills-rai

# Auto-detect your project and install only matching bundles (~50-200 MB)
langskills-rai bundle-install --auto

# Search the pre-built skill library (Vibe Research)
langskills-rai skill-search "kubernetes networking" --top 5

# Generate new skills from any topic (Vibe Coding)
cp .env.example .env   # fill OPENAI_API_KEY + OPENAI_BASE_URL
langskills-rai capture "Docker networking@15"

China users: export HF_ENDPOINT=https://hf-mirror.com before bundle-install for faster downloads.

Pre-built bundles are distributed from Hugging Face. The repo itself only keeps the code and local build workflow.

Full setup details → Installation

95,093 research skills distilled from academic papers + 24,515 coding/tech skills from GitHub, StackOverflow, and the web — all searchable offline.

# Install a domain bundle (downloads from Hugging Face)
langskills-rai bundle-install --domain linux

# Or auto-detect your project type and install matching bundles
langskills-rai bundle-install --auto

# Search skills offline (FTS5 full-text search)
langskills-rai skill-search "container orchestration" --top 10

# Filter by domain and minimum quality score
langskills-rai skill-search "CRISPR" --domain research --min-score 4.0

# Get full skill content as Markdown
langskills-rai skill-search "React hooks" --content --format markdown

skills/by-skill/<domain>/<topic>/
├── skill.md          # The skill content (tutorial / how-to / protocol)
├── metadata.yaml     # Provenance, tags, quality score, LLM model used
└── source.json       # Evidence trail back to original web/paper source

langskills-rai search tavily "Linux journalctl" --limit 20
langskills-rai search github "journalctl" --limit 10

# Basic
langskills-rai capture "journalctl@15"

# Target a specific domain
langskills-rai capture "React hooks@20" --domain web

# All domains
langskills-rai capture "Kubernetes" --all --total 30

langskills-rai validate --strict --package
langskills-rai reindex-skills --root skills/by-skill

langskills-rai build-site
langskills-rai build-bundle --split-by-domain

langskills-rai queue-seed                     # seed from config
langskills-rai topics-capture topics/arxiv.txt # or from file
langskills-rai runner                          # start worker
langskills-rai queue-watch                     # monitor

captures/<run-id>/
├── manifest.json          # Run metadata
├── sources/               # Fetched evidence per source
├── skills/                # Generated skill packages
│   └── <domain>/<topic>/
│       └── skill.md
└── quality_report.md      # Validation summary

LangSkills supports Linux, macOS, and Windows. Python 3.10+ required.

pip install langskills-rai

# Download skill bundles (auto-detect your project type)
langskills-rai bundle-install --auto

# Or install a specific domain
langskills-rai bundle-install --domain linux

# Verify
langskills-rai self-check --skip-remote

bundle-install defaults to auto-detection when you omit both --auto and --domain.

git clone https://github.com/LabRAI/LangSkills.git && cd LangSkills
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
playwright install chromium          # optional: Baidu/Zhihu/XHS sources
cp .env.example .env                 # fill OPENAI_API_KEY + OPENAI_BASE_URL
langskills-rai self-check --skip-remote

git clone https://github.com/LabRAI/LangSkills.git && cd LangSkills
python -m venv .venv && .venv\Scripts\activate
pip install -e ".[dev]"
copy .env.example .env               # fill OPENAI_API_KEY + OPENAI_BASE_URL
langskills-rai self-check --skip-remote

Copy the prompt below and paste it into Claude Code / Codex / Cursor / Windsurf — the AI agent will automatically clone, install, configure, and verify LangSkills for you.

Do the following steps in order. Do NOT skip any step.

1. Install langskills-rai from PyPI:
   pip install langskills-rai

2. Auto-detect my project and install matching skill bundles:
   langskills-rai bundle-install --auto

3. Run the self-check to verify everything is working:
   langskills-rai self-check --skip-remote

4. If self-check passes, run a quick smoke test — search the built-in library:
   langskills-rai skill-search "machine learning" --top 3

5. If I want to generate NEW skills (not just search), ask me for my
   OPENAI_API_KEY and OPENAI_BASE_URL, then set them as environment variables.

Done. Report the results of steps 3 and 4.

LangSkills is available as an OpenClaw skill — giving any OpenClaw-powered agent access to 119K+ evidence-backed skills.

Install from Claw Hub (coming soon):

clawhub install langskills-search

Manual install — save the block below as ~/.openclaw/skills/langskills-search/SKILL.md:

---
name: langskills-search
version: 0.1.0
description: Search 119K evidence-backed skills from 95K+ papers & 24K+ tech sources
author: LabRAI
tags: [research, skills, knowledge-base, search, evidence]
requires:
  bins: ["python3"]
metadata: {"source": "https://github.com/LabRAI/LangSkills", "license": "MIT", "min_python": "3.10"}
---

# LangSkills Search

Search 119,608 evidence-backed skills covering 62K+ research papers and 23K+ coding/tech sources — all offline via FTS5 SQLite.

## When to Use

- User asks for best practices, how-tos, or techniques on a technical topic
- You need evidence-backed knowledge (not LLM-generated guesses)
- Research tasks that benefit from academic or real-world source citations

## First-Time Setup

```bash
pip install langskills-rai
# Install matching bundles for the current project or pick a domain:
langskills-rai bundle-install --auto
```

## Search Command

```bash
langskills-rai skill-search "<query>" [options]
```

### Parameters

| Flag | Description | Default |
|:---|:---|:---|
| `--top N` | Number of results | 5 |
| `--domain <d>` | Filter by domain | all |
| `--min-score N` | Minimum quality score (0-5) | 0 |
| `--content` | Include full skill body | off |
| `--format markdown` | Output as Markdown | text |

### Example

```bash
langskills-rai skill-search "CRISPR gene editing" --domain research --top 3 --content --format markdown
```

## Reading Results

Each result includes: **title**, **domain**, **quality score** (0-5), **source URL**, and optionally the full skill body. Higher scores indicate stronger evidence chains.

## Available Domains

`linux` · `web` · `programming` · `devtools` · `security` · `cloud` · `data` · `ml` · `llm` · `observability` · `research-arxiv` · `research-plos-*` · `research-elife` · `research-other`

## Tips

- Use `--content --format markdown` to get copy-paste-ready skill text
- Combine `--domain` with `--min-score 4.0` for high-quality results
- Run `bundle-install --auto` in a project directory to install only relevant domains

All commands: langskills-rai <command> (or python3 langskills_cli.py <command> from source)

Master config: config/langskills.json — domains, URL rules, quality gates, license policy.

Contributions are welcome! Please follow these steps:

1. Open an issue to discuss the proposed change
2. Fork the repository and create your feature branch
3. Submit a pull request with a clear description

This project is licensed under the MIT License.

Copyright (c) 2026 Responsible AI (RAI) Lab @ Florida State University

• Authors: Tianming Sha (Stony Brook University), Dr. Yue Zhao (University of Southern California), Dr. Lichao Sun (Lehigh University), Dr. Yushun Dong (Florida State University)
• Design: Modular pipeline architecture with multi-source intelligence, built for extensibility and offline-first search
• Skills: 119,608 evidence-backed skills generated from 62K+ papers and 23K+ tech sources via LLM-powered quality gates
• Sources: Every skill traces to real web pages, academic papers, or code repositories (arXiv, PMC, PLOS, Nature, eLife, GitHub, etc.)