A collection of production-tested skills for AI coding agents. Built on the Agent Skills open standard.
Compatible with Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot, Kiro, Kimi Code, and any agent supporting SKILL.md.
# Clone the repo
git clone https://github.com/leon2k2k2k/agent-skills.git
# Install a single skill (Claude Code)
cp -r agent-skills/learn-from-paper ~/.claude/skills/
# Use it
# /learn-from-paper https://arxiv.org/abs/1706.03762| Skill | Category | Description | Env Vars |
|---|---|---|---|
| math-background | Research | Articulate vague mathematical intuitions into precise problem statements | -- |
| learn-from-paper | Research | Interactive paper reading sessions with Q&A, quizzes, and testing | -- |
| external-llm | AI Integration | Multi-turn conversations with any OpenAI-compatible LLM (DeepSeek, OpenAI, Groq, etc.) | LLM_API_KEY |
| vr-loop | Math | Proof verification-and-refinement loop (5 consecutive passes required) | LLM_API_KEY |
| external-review-loop | Math | Section-by-section paper review via external reasoning model | LLM_API_KEY |
| latex-math-writeup | Writing | Generate publication-quality LaTeX papers with theorems and proofs | -- |
| file-organizer | Dev Tools | Intelligent file organization, duplicate detection, and cleanup | -- |
| numerical-investigation | Math | Guided numerical conjecture verification with structured workflow | -- |
Copy a skill directory into your agent's skills folder:
# Claude Code (personal — available in all projects)
cp -r learn-from-paper ~/.claude/skills/
# Claude Code (project-specific)
cp -r learn-from-paper .claude/skills/| Agent | Skills directory |
|---|---|
| Claude Code | ~/.claude/skills/ or .claude/skills/ |
| Gemini CLI | ~/.gemini/skills/ |
| Cursor | .cursor/skills/ |
| OpenCode | ~/.opencode/skills/ |
| Kimi Code | ~/.kimi/skills/ |
| Codex | .codex/skills/ |
Skills that call external APIs need these environment variables:
export LLM_API_KEY="your-key-here"
export LLM_BASE_URL="https://api.deepseek.com" # optional, default
export LLM_MODEL="deepseek-reasoner" # optional, defaultUsed by: external-llm, vr-loop, external-review-loop.
See external-llm for provider-specific configurations (OpenAI, Groq, Together, Ollama, etc.).
Each skill is a single SKILL.md file (plus optional references). Fork and edit to fit your workflow — change the default preamble in latex-math-writeup, adjust verification thresholds in vr-loop, etc.
Apache-2.0