Using AI skill packages

apprenti.dev publishes a small instruction package for five AI coding assistants — Claude Code, Cursor, OpenAI Codex, OpenCode, and GitHub Copilot — plus a template for a personal ChatGPT Custom GPT. Each one teaches the assistant the curriculum's JSON+Markdown shapes, the locale-overlay rules, the personal-data boundaries, pedagogical guidance for writing quality content, and a validation checklist to run before calling any change finished. Download them from the AI Skills page.

None of these replace apprenti creator (the visual authoring extension) — they're for whichever coding assistant you already work in, including inside apprenti creator's own VS Code window if you use one there too. Sideload the extension itself with Installing apprenti creator.

Claude Code

Unzip the Claude Code package into the root of your curriculum repository. Claude Code discovers project skills automatically — no restart or registration step is needed. The skill activates on its own when you're working with curriculum files, and loads its two reference files (JSON shapes, validation checklist) only when it needs the detail, so you don't need to open them yourself.

Cursor

Unzip the Cursor package into the root of your curriculum repository. Cursor picks up project rules from that folder automatically. This one is scoped to curriculum files — tasks, terms, the curriculum manifest, resources, competencies, and instructions files — so it stays out of the way everywhere else in the repository.

OpenAI Codex

Unzip the Codex package into the root of your curriculum repository. It creates a project instructions file that Codex reads automatically at the start of every session in that repository.

If your repository already has its own project instructions file, don't overwrite it — merge this package's content into yours instead, or nest it under the curriculum folder if you'd rather scope it to that subtree (Codex loads these files from nested directories too).

OpenCode

Unzip the OpenCode package into the root of your curriculum repository. It creates the same project instructions convention OpenCode reads automatically. If you'd rather not touch your main instructions file, register it separately through OpenCode's own configuration instead — see its docs for the exact setting. You can also regenerate project guidance any time with OpenCode's init command and merge this content in afterward.

GitHub Copilot

Unzip the Copilot package into the root of your curriculum repository. It creates two files: a repository-wide instructions file Copilot Chat reads automatically, and a path-scoped instructions file that only applies to curriculum files specifically (task, term, and curriculum JSON, resources, competencies, instructions files) — currently read by Copilot's cloud agent and code review. Each is self-contained, so either one on its own is complete.

If your repository already has its own repo-wide Copilot instructions file, don't overwrite it — merge this package's content into yours instead.

ChatGPT (Custom GPT)

This one is a different shape from the five above, because plain ChatGPT has no mechanism to discover a file in your repository on its own. Unzip the package locally instead of into a repository — it contains an instructions file to paste into a personal Custom GPT's Instructions field, and three reference files to upload as that GPT's Knowledge files. Full setup steps are in the package's own README, since the flow (create a GPT, paste, upload) doesn't map onto "unzip into your repo" the way the others do.

The real difference in daily use: this Custom GPT never sees your repository unless you paste or upload a file into the chat yourself, and you're the one who copies its answer back into your repository and commits it — nothing here reaches your files on its own.

What to expect

Ask your assistant to draft a new task, translate a term into another locale, or add a competency, and it should produce something matching the shapes documented in Curriculum structure basics — stable ids independent of file paths, a complete AI policy block, evidence types a mentor can actually open, and calm, non-gamified task copy. It should also refuse to touch any apprentice or mentor personal-data folders, and should never commit or push on its own — those stay your decision.

Keeping a package up to date

These packages are regenerated whenever the curriculum format changes. If you're mid-way through authoring and a new version ships, re-download and unzip over the existing files — nothing else in your repository is affected.