Where learning apps are made
One person who knows their subject, one AI agent that does the handwork, and programs that hold the rules. Here is how they work together, how we teach the agent, and how it is built.
A workshop, not a factory
What you decide in each phase, what the agent does, and which skill, command, hook or action supports it. Drawn from the architecture, so it shows what is really there.
- ExploreWhat should learners be able to do?
- PlanHow will learners practise?
- CodeBuilt locally, by the agent
- CommitOnline at your yes
Behind the workshop
-
Which assistant to build with
Any coding assistant that can use skills and change files will do. The two setups we use most, and three apps a budget model built.
Assistants and examples -
Prompt engineering
How we make the guidance work for the creator's assistant: skills, plan templates, AGENTS.md, and the evals that show whether the assistant follows the process.
Guidance and evals -
Architecture
The tooling in arc42: its building blocks by phase, the decisions and why we took them, and how it keeps an app from collecting anything in secret.
The arc42 documentation -
GitHub, explained
Where your app lives and is published, and which parts of GitHub the tooling uses, for creators who have never used GitHub.
No GitHub knowledge needed
Reference
- Rules: what every listed app must meet, with id, scope, severity and where each rule is told and checked
- llms.txt: the commands, guidance, skills, rules and schemas, for AI assistants
- Validation report (JSON Schema): what
lernapps checkreports - Review verdict (JSON Schema): what the review agent judged on an app that asks to be listed
- Source on GitHub: the CLI, the shared workflows (build, check, publish) and the presets for updates