dojo.md — University for AI AgentsTrain any model through scenario-based courses. Graduate with a SKILL.md — portable expertise that makes agents reliable in production. No fine-tuning. No weight modification. Just knowledge, distilled and proven.How It WorksRun agents through progressively harder scenarios with mock services. An LLM judge evaluates responses, extracts failure patterns, and generates a SKILL.md document that gets injected into the agent's system prompt. Different models fail differently — each gets its own SKILL.md. The loop repeats until the target score is reached.Features

92 courses across 7 categories: marketing, engineering, writing, support, data, sales, design

4,400+ scenarios with 5 difficulty levels

Any model via OpenRouter (637+ models): Claude, GPT, Gemini, DeepSeek, Grok, Llama, MiniMax, Kimi, Qwen

Auto-training loops with convergence detection

Per-model SKILL.md generation — your blind spots are uniquely yours

Works with Claude Code, OpenClaw, Cursor, Windsurf, and any MCP-compatible agent

LLM-as-judge evaluation: cheap agent does the work, strong judge evaluates quality

Knowledge graduation: even at 100% score, the domain expertise has standalone value

Works With

Claude Code — MCP server integration, train from your IDE

OpenClaw — graduated SKILL.md drops directly into OpenClaw's skill directory. Dojo skills are earned, not written.

Cursor & Windsurf — same MCP config, same SKILL.md output

Any MCP-compatible agent — portable skills, train once, use everywhere

Quick Startnpm install -g dojo.md dojo train stripe-refunds dojo train ad-copy --model deepseek/deepseek-v3.2 --target 90 dojo arena stripe-refundsLinks

GitHub

npm

llms.txt — For AI agents