Instead of cloning 12 separate skill repositories or stuffing 277 skills into a prompt, I wanted a single install where the agent adopts the right domain expert persona and pulls only the 1-3 SKILL.md files needed for the task.
Route the prompt to one domain expert and pull only the required SKILL.md files on demand.
Zero-dependency Python (scripts/skillary_agent.py and scripts/crew.py) scores prompts against index/experts.json and index/skills.json, fetches verified SKILL.md files from the 12 Skillary repos over HTTPS, and coordinates 6-role multi-agent crews.
Runs locally with Python 3.10+ standard library only. On-demand skill pulls fetch public SKILL.md files directly from githubusercontent.com with zero telemetry.
Host agent capabilities differ across platforms. Read the protocol and capability map before running a multi-agent workflow; no hosted cloud runner is provided.
Installing dozens of separate skill repositories or loading all 277 skills into an agent's system prompt wastes context tokens and blurs domain focus.
I wanted a single entry point that turns any coding model into the exact domain expert a prompt needs, like a UI/UX Designer, Game Developer, or Security & Bug-Fix Engineer.
skillary-agents v5.0.0 maps all 277 Skillary skills across 11 domain expert personas and pulls only the 1 to 3 required SKILL.md files on demand.
Works across 300+ frontier and open-source models, including Gemini 4 Argon, Gemini 3.8 Flash, Claude Opus 5.5, Fable 5.1, GPT-6 Astra, OpenRouter, and Ollama, plus 6-role crew coordination.