Contribute
build-3000 is fully open source under MIT. The whole curriculum lives in version-controlled JSON - GitHub is the CMS. If you can edit a file and run one command, you can improve the curriculum.
1. Find what to change
Browse data/modules.json and data/levels.json. Every module is one JSON object; every level is one row of the ladder. Content lives in files, not a CMS.
2. Follow the schema
lib/schemas.ts is the single source of truth. A module needs all fifteen sections: objective, vocabulary, concepts, prerequisites, explanations, examples, mistakes, agent guidance, project brief with agent context, grouped Definition of Done, verification prompt, review questions, capabilities and next-module link.
3. Validate before you commit
Run pnpm validate-curriculum. It checks schemas AND cross-references: levels resolve, prerequisites exist, no cycles, and every live module carries a functional DoD plus a verification prompt. No validation, no merge.
4. Open a conventional PR
One logical change per PR with a Conventional Commit message (feat:, fix:, docs:). CI re-runs lint, typecheck, tests and curriculum validation automatically.
What a module looks like
{
"slug": "my-new-module",
"title": "...",
"level": "level-0-foundations",
"domain": "fundamentals",
"summary": "One line for cards and search.",
"order": 6,
"objective": "What the learner will be able to do...",
"vocabulary": [{ "term": "...", "meaning": "..." }],
"concepts": [{ "name": "...", "explanation": "..." }],
"prerequisites": ["an-existing-module-slug"],
"explanations": [{ "title": "...", "body": "..." }],
"examples": [{ "title": "...", "description": "..." }],
"commonMistakes": [{ "mistake": "...", "fix": "..." }],
"agentGuidance": "How to work with an AI agent here...",
"project": {
"slug": "the-project",
"title": "...",
"brief": "...",
"requirements": ["..."],
"agentContext": "Paste-ready context for the learner's agent."
},
"definitionOfDone": {
"functional": ["..."], "architecture": [], "security": [],
"testing": [], "production": []
},
"verificationPrompt": "The audit prompt for their agent...",
"reviewQuestions": ["..."],
"capabilities": ["..."],
"nextModule": null
}Content integrity rules
- - Technically accurate and verifiable only - no invented statistics or fake sources.
- - Every live module ships a Definition of Done and a verification prompt. Non-negotiable.
- - Plain language over jargon; define a term before depending on it.
- - Projects stay agent-agnostic: transferable skills, tool-specific tips only where unavoidable.
Found something broken? Open an issue or go back to learning.