A workshop, not a factory
A software factory takes people out of the line and turns specifications into output. lernapps.net is a Manufaktur: one person who knows their subject makes one learning app, for their learners, from their own material.
An AI agent does the handwork and programs hold the rules. The person keeps every decision that needs judgement: what the app teaches, whether it is right, whether it goes online.
What runs through every phase comes first, then the four phases of the plan. Each box is a part of the tooling, names what it is built with and links to its place in the architecture. Hover over a box, or move to it with the keyboard, to read what it does for whom.
- Person decides or is asked
- AI agent works from it
Across all phasesWhat runs through every phase?
- Process guidanceMarkdown
Person Answers the interview and confirms at each checkpoint.
AI agent Gives the four phases, the checkpoints, and which skill comes when; a brief for each sub-agent.
- AGENTS.mdMarkdown file at the app's root
AI agent Reads it on every return: where the current plan is and what holds.
- Plan fileMarkdown with YAML front matter and a versioned JSON Schema
Person Reads where the work stands, in plain words.
AI agent Works the plan step by step and notes where it had to guess.
- Alignment CLITypeScript CLI, Playwright, axe-core
AI agent One command, the same result in the hooks, in CI and at the listing.
ExploreWhat should learners be able to do?
You decide
What the app teaches and for whom: the goal, the learners, your own material, the mistakes your learners make.
Checkpoint research: you confirm what the app teaches.
- lernapps-appMarkdown (agentskills.io)
AI agent Knows what every app must be, and chooses the archetype from what learners do.
- lernapps-content-researchMarkdown (agentskills.io)
Person Brings their curriculum and their own material.
AI agent Researches the curriculum, writes the competencies, opens every source it cites.
- lernapps-exercise-patternsMarkdown (agentskills.io)
AI agent Reads how learners of the subject practise, which points to the archetype; in Plan chooses an exercise per competency and shares data as files.
- lernapps-quiz-designMarkdown (agentskills.io)
AI agent Knows what to ask for a quiz and how its content maps onto rounds and question types.
- lernapps-explainer-designMarkdown (agentskills.io)
AI agent Knows what to ask for an explainer and how its content maps onto chapters, pages and exercises.
- lernapps-interactive-designMarkdown (agentskills.io)
AI agent Knows what to ask for an interactive app and how its content maps onto core, screens and playthroughs.
- Archetype templatesFiles at a pinned commit
Person Hears what the chosen archetype can and cannot do.
AI agent Reads the README of each template before it chooses the archetype.
PlanHow will learners practise?
You decide
In one read: the tasks, the sketches and the content method.
Checkpoint plan: nothing is built before your yes.
- lernapps-content-growthMarkdown (agentskills.io)
AI agent Cuts a later increment into slices, one per competency, for parallel sub-agents.
CodeBuilt locally, by the agent
You decide
Nothing. Time for a coffee: the agent builds on its own and may come back with a question only you can answer.
- lernapps-cliMarkdown (agentskills.io)
AI agent Fixes every failed check from its message and never bypasses one; shows the creator the pages of the app.
- lernapps-quizMarkdown (agentskills.io)
AI agent Writes the question bank of a quiz and adds the images its questions show, the only files it writes there.
- lernapps-explainerMarkdown (agentskills.io)
AI agent Writes the chapters and pages of an explainer in its notation; a component only where no block fits.
- lernapps-interactiveMarkdown (agentskills.io)
AI agent Writes the core, components, data and playthroughs of an interactive app, and its architecture.
- lernapps-content-questionsMarkdown (agentskills.io)
AI agent Writes options and feedback that teach.
- GeneratorTypeScript CLI
AI agent Creates the app from its template, passing every check from the first commit; update brings fixes later.
- Git hooksgit hooks
AI agent Check every commit, and before a push build the app and run the end-to-end tests.
- Lint rulesOxlint JS plugin
AI agent Marks a problem while it writes, such as a request to another server before a click.
- Archetype runtimenpm package, subpath exports per archetype
AI agent Runs the app: engine, accessibility, the page without JavaScript, its own tests.
CommitOnline at your yes
You decide
Whether the app goes online: you look at the preview and say yes.
Checkpoints publish and list: the listing pull request is the review.
- lernapps-explainer-self-assessmentMarkdown (agentskills.io)
AI agent Chooses the exercises of the self-assessment of an explainer and writes it into its content, or notes why there is none.
- lernapps-githubMarkdown (agentskills.io)
Person Makes the one click only the owner of the repo can make, if one is needed.
AI agent Sets up repo and Pages, opens the preview pull request, publishes after your yes.
- lernapps-commitMarkdown (agentskills.io)
AI agent Publishes one squash commit, its message taken from the plan.
- lernapps-listingMarkdown (agentskills.io)
Person Sees the listing pull request and its result.
AI agent Writes the entry and opens the listing pull request from your fork.
- Git hooksgit hooks
AI agent Check every commit, and before a push build the app and run the end-to-end tests.
- App check actionGitHub composite action
AI agent Runs the same checks in CI.
- App publishGitHub composite action, GitHub Pages
AI agent Publishes the checked bundle at your own address.
- App previewGitHub composite action, GitHub Pages
Person Looks at the preview before saying yes.
AI agent Puts the checked bundle of a pull request online as a preview.
- Listing validationGitHub Actions workflow
AI agent Checks the entry against the live app and comments the result.
Then, in the catalogIs it listed?
- Review procedureMarkdown, JSON Schema
Person The platform owner reads the verdict and decides on the listing.
AI agent A review agent judges what no check can, against a rubric.
What you get from it
- As a developerNo toolchain to fight: a template that passes every check from its first commit, messages that say what to fix, and the same checks in the hooks, in CI and at the listing.
- As a managerRules with ids, enforced by programs at fixed points, not by trust. Every app says where it deviates and why. Content quality stays a human judgement, on purpose.
What no machine decides
- What the app teaches, and for whom.
- Whether the content teaches correctly: no check and no review proves that.
- Whether the app goes online, and under whose name.
Next
- Architecture: chapter 5 groups the skills by phase, as here.
- Rules: what every app must meet.
- GitHub, explained: for creators who do not know GitHub.