Coding Agents
Build tutorials with Claude Code, GitHub Copilot or another coding agent, without learning the tool first: the package ships skills that teach your agent how InteractiveCodeScroll works, a validation command it can read, and a Markdown version of every tutorial for your readers’ AI assistants.
Start with an agent
Section titled “Start with an agent”New project: pick your agents in the wizard, or pass them as a flag.
npm create interactive-code-scroll@latest my-talk -- --agents claudeExisting project: copy the skills into your agent’s folder.
npx interactive-code-scroll skills install --agent claude # or copilot, agents, allYour project then has:
- The skills in your agent’s folder (
.claude/skills/,.github/skills/or.agents/skills/), copied from the installed version of the package. AGENTS.md(and aCLAUDE.mdthat imports it, for Claude Code): where the tutorial is, which skills to use and the validation loop.npm run check: the build’s validation in about a second, which the agent runs after each change.npm run skills: installs the skills again, for example after upgrading the package.
Open the project in your agent and ask for what you want.
Skills included
Section titled “Skills included”| Skill | What your agent learns |
|---|---|
ics-authoring |
How a tutorial is made: folders, <Step> and the other components, frontmatter, regions and @var markers, the Preview and Result pane, code variants, series sites and the CLI. |
ics-verify |
The validation loop: check --json, fix each problem with its known fix, repeat, then build. |
ics-tutorial-design |
Planning before writing: audience, goal, the finished app and one idea per step, kept in TUTORIAL-PLAN.md as its memory between sessions. |
ics-tutorial-review |
Reviewing a draft as a first-time reader and as an editor, with findings per step (blocker, major, minor) and concrete fixes. |
ics-talk |
Preparing a tutorial for a talk: timing per step, slides as steps, results shown big, offline fallbacks, the presentation keys and a rehearsal checklist. |
ics-migrate |
Converting a blog post, codelab, README walkthrough or sample repository into a tutorial, and upgrading to a new version of the package. |
Skills use the open Agent Skills format: the agent reads each skill’s short description up front and loads the rest only when a task needs it, so they cost little context until used.
What to ask
Section titled “What to ask”| You want to | Ask, for example |
|---|---|
| Start from an idea | “Plan a 10-minute demo that shows how to search places with this API, for developers new to it.” |
| Bring existing content | “Turn blog/geocoding.md into a tutorial with cURL and Python variants.” |
| Write the next step | “Write the next planned step: the sign-in call, with a field for the client ID.” |
| Check your work | “Review the tutorial before my talk and list what to fix.” |
| Get ready to present | “Prepare this tutorial for a 15-minute talk: timing, big results and a fallback for every request.” |
| Upgrade | “Upgrade the tutorial to the latest interactive-code-scroll and fix what changed.” |
The skills tell the agent to agree on a plan with you first and to write one step at a time, checking each one, so you stay in charge of the story.
The validation loop
Section titled “The validation loop”Agents learn fastest from errors. interactive-code-scroll check --json runs every validation the build runs, in about a second, and prints each problem with its file, line and column:
{ "severity": "error", "file": "tutorial/tutorial.mdx", "line": 12, "column": 1, "message": "<Step> region \"config\" not found in code/main.js" }The ics-verify skill maps every message to its fix and finishes with a full build. What check cannot see (how the page looks, whether the Preview app works) needs a look in the dev server.
Markdown for your readers’ AI
Section titled “Markdown for your readers’ AI”Readers use AI assistants too. Every tutorial publishes a plain Markdown version (tutorial.md, one per code variant) with the prose, each step’s code and the final project, and the site lists them in llms.txt. The Ask AI action links it, so the assistant can read the whole tutorial, not only the selected lines.
Supported agents
Section titled “Supported agents”--agent |
Folder | Agents |
|---|---|---|
claude |
.claude/skills/ |
Claude Code |
copilot |
.github/skills/ |
GitHub Copilot |
agents |
.agents/skills/ |
Agents that read this generic folder, such as VS Code’s agent mode |
For other agents, check where they read project skills and copy the folders from node_modules/interactive-code-scroll/dist/skills/. Commit the skill folders so everyone working on the tutorial gets them, and run npm run skills (or skills install) after upgrading.