University for AI agents. 92 courses, 4400+ scenarios, any model via OpenRouter. Auto-training loops generate per-model SKILL.md documents. Works with Claude Code, OpenClaw, Cursor, Windsurf. No fine-tuning required.

dojo.md

Your agent demos well. It fails in production. dojo.md fixes that.

Train any model through scenario-based courses. Graduate with a SKILL.md — portable expertise that makes agents reliable. No fine-tuning. No weight modification. Just knowledge, distilled and proven.

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

"Hey Cursor, train yourself on cold-email-b2b and loop until you hit 90"

  Iteration 1: 31/100 — doesn't know subject line rules, no personalization
  Iteration 2: 58/100 — SKILL.md injected, learns "under 50 chars, no caps"
  Iteration 3: 74/100 — gets structure right, still weak on CTAs
  Iteration 4: 86/100 — nails the pain→solution→ask framework
  Iteration 5: 91/100 — ✓ target reached

  SKILL.md → .cursor/skills/cold-email-b2b/SKILL.md

Now every cold email it writes follows the framework. Permanently.

What Just Happened

You told Claude Code to train itself. It ran through 50 scenarios of progressively harder cold emails — bad subject lines, wrong tone, missing personalization, weak CTAs. An LLM judge scored every attempt. After 5 iterations, it extracted everything it learned into a SKILL.md that lives in your project forever.

Next time you say "write a cold email to this VP of Engineering", it loads the SKILL.md automatically. It knows the rules now.

This works for anything:

# Your agent writes terrible Google Ads? Fix it in 5 minutes
dojo train ad-copy-google-ads --target 85

# Support agent keeps