If you can read ~100 lines of Python, you understand Skills.
Overview
"Progressive disclosure: show only what's needed, when it's needed." The simplest way to add specialized skills to your AI agent. A minimal implementation of the Agent Skills standard. ~170 lines of code. - Scan for SKILL.md files - Show name + description (~50 tokens/skill) - Load full instructions (~500-2000 tokens) Skills use progressive disclosure to avoid MCP's upfront schema bloat (10k-25k tokens). See MCP_VS_SKILLS.md for detailed comparison. - Real-world skills - git-expert - Git workflow and conventional commits - code-reviewer - Code quality and best practices - api-designer - REST API design patterns - Simulated API scenarios - travel-agent - Flight/hotel booking simulation ────────────────────────────────────────
README
nanoSkills
中文 | English
“Progressive disclosure: show only what’s needed, when it’s needed.”
The simplest way to add specialized skills to your AI agent.
A minimal implementation of the Agent Skills standard. ~170 lines of code.
install
pip install -r requirements.txt
Set environment variables:
export OPENAI_API_KEY='your-key-here'
export OPENAI_BASE_URL='https://api.openai.com/v1' # optional
export OPENAI_MODEL='gpt-4o-mini' # optional
quick start
With real skills:
python agent.py "review my code for best practices"
python agent.py "help me design a REST API"
With fake travel agent (API simulation):
python agent.py --skills-dir skills-fake "book a flight from SFO to NYC"
how it works
Progressive disclosure in 3 levels:
Level 1: Discovery - Scan for SKILL.md files Level 2: List - Show name + description (~50 tokens/skill) Level 3: Activate - Load full instructions (~500-2000 tokens)
skills = discover_skills("./skills-real")
tools.append(build_activate_tool(skills))
# Agent decides when to activate based on task
Why not MCP? Skills use progressive disclosure to avoid MCP’s upfront schema bloat (10k-25k tokens). See MCP_VS_SKILLS.md for detailed comparison.
skill libraries
skills-real/ - Real-world skills
git-expert- Git workflow and conventional commitscode-reviewer- Code quality and best practicesapi-designer- REST API design patterns
skills-fake/ - Simulated API scenarios
travel-agent- Flight/hotel booking simulation
examples
# Code review
python agent.py "review this Python function for improvements"
# Git help
python agent.py "write a commit message for adding user auth"
# API design
python agent.py "design a REST API for a blog system"
# Travel booking (fake APIs)
python agent.py --skills-dir skills-fake "plan a trip to Paris"
creating skills
mkdir -p skills-real/my-skill
cat > skills-real/my-skill/SKILL.md << 'EOF'
---
name: my-skill
description: Brief description for AI
---
Full instructions here.
EOF
license
MIT
────────────────────────────────────────
⏺ One skill at a time, infinite expertise.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add sanbuphy/nanoskills