carlvellotti/free-ai-courses: teaching AI tools by pasting one line into your assistant

The distribution model is a sentence you paste into a chat box

Most course repositories ask you to clone, install dependencies, and work out where to start. This one asks for a single line. Open your AI app, start a new project, and paste:

Fetch fullstackpm.com/cli and help me get started

The assistant fetches a CLI manifest, installs the Full Stack PM app, lists what is available and begins teaching. The same paste works in Claude Desktop, where it goes in the code tab, and in Cursor or Codex, where it goes into the chat. The reasoning behind the design is stated in the repository description itself: every lesson runs inside your AI app, with the AI as the guide, so there are no videos to sit through and no slides to skim.

That single decision reorganises everything else about the project. A lesson is not a document you read, it is a set of instructions an assistant walks you through using the tool being taught. Teaching Claude Code inside Claude Code means the exercises have somewhere to happen, and the repository is explicit that this is not where the teaching happens. Its own framing is that it holds the source content, and that you do not need to clone it to take a course because the paste prompt handles everything.

For someone evaluating the project, the consequence is that the repository is the wrong place to judge the teaching. What you can judge there is the catalogue, the packaging, and the terms. A reader who wants to know whether a lesson is any good has to install it and try it, which is either the intended experience or an unfortunate gap, depending on your point of view.

One npm package, one binary, and a table of course identifiers

The delivery mechanism is a single npm package named `fullstackpm` that installs a command called `fspm`. A shell installer is offered for machines without Node, one script for macOS and Linux and a PowerShell equivalent for Windows.

npm i -g fullstackpm

The install step is separate from the onboarding step, and the onboarding step is separate from taking a course. After installing, you run `fspm onboard` once, then install individual courses by identifier:

fspm onboard
fspm get claude-code-for-pms
fspm get gsd-framework

The identifier table is the spine of the project, and it is worth reading closely because it shows how the catalogue is organised. Four long-form courses cover Claude Code for product managers, Claude Code for a general audience with no technical background assumed, Cursor for product managers, and Codex for product managers. Each has its own domain: a `ccforpms` domain, a `ccforeveryone` domain, a `cursorforpms` domain and a `codexforpms` domain. A fifth entry, for Google's Antigravity, is listed in the same table but carries a footnote saying it starts from its own site for now, which is the one place where the uniform delivery story has a hole.

Beneath those sits a second tier called the Short Stacks, four mini-courses described as finishable in a sitting: a planning framework for building real projects with AI, advanced techniques for daily Claude use, two named practitioners' workflows gathered from the field, and a combination of Claude with Obsidian for notes and knowledge. The identifiers follow the same pattern, from `gsd-framework` through `vin-obsidian-workflows`, and the paste prompt works for these too by swapping the course name. Two reference documents round out the system, an agent-facing page and a human-facing one at the same address, which is a small piece of design worth noticing: the project is written to be read by a model and by a person from the same source.

What the repository actually contains

The directory listing is a plain map of the catalogue. Each course has a top-level directory named after it, so `course-materials` holds the Claude Code for product managers content, and there are matching directories for the general Claude Code course, the Cursor course, the Codex course and the Antigravity course. The Short Stacks live in two directories together, `gsd-framework` and `mini-lessons`, which matches the pattern of a course with a full structure sitting next to a pile of standalone lessons.

Three other entries are worth pausing on because they tell you what the project is for. There is a directory for a complete guide to Claude Cowork, which is not one of the listed courses. There are two directories labelled as bonus material, one attached to the Claude Code for product managers course and one attached to the Codex course. And there is a `website` directory that is the source for one of the public course domains rather than for all of them, which implies the other domains are built elsewhere.

The supporting files follow the same practical bent. There is a file of test commands, which suggests the CLI surface is exercised somewhere in the repository even though no package manifest is published as the project entry point. There is a `CLAUDE.md` at the root, the instruction file an assistant reads when working in a directory, which for a project whose content is consumed by assistants is not decoration but part of how the content is shaped. There is an `assets` directory holding a welcome banner, a `docs` directory, a `scripts` directory, an `.obsidian` directory indicating the material is also edited as an Obsidian vault, and a `.gitignore`.

The primary language of the repository is recorded as HTML, which tells you the website directory is the largest thing by line count even though the teaching content is what the project is for. That inversion is a fair summary of the whole approach: the public face is a small site, and the product is a set of instructions.

Why this content is licensed against modification

The terms are Creative Commons Attribution-NonCommercial-NoDerivatives 4.0, attributed to Carl Vellotti for 2025 and 2026, and the repository restates them in one sentence: you may view and share the course content with attribution, but commercial use and modifications are not permitted.

That is an unusual choice for a course, and the usual reason for it is that the material is both the product and the marketing. A permissive licence on a free course invites someone to fork it into a competing paid course, which is precisely what no-derivatives exists to stop. The attribution clause keeps the name attached, the non-commercial clause blocks resale, and the no-derivatives clause blocks remixing, so the only permitted redistribution is a link or a share of the original.

For an individual learner the restrictions are close to irrelevant, since reading and running a course on your own machine is neither commercial nor derivative. For a team, they are decisive in a way that is easy to miss. You cannot adapt a lesson to your own product vocabulary, you cannot cut it down to a twenty-minute onboarding module, and you cannot translate it into another language for colleagues who need one. If your organisation needs any of that, this repository is a reference to read rather than a curriculum to adopt.

The licence also shapes how the deliverable should be judged. Since the content is not designed to be modified, the quality of the writing is the whole product, and there is no mechanism in the repository for a learner to contribute an improvement back. Whether that is a feature or a limitation depends entirely on whether you wanted a course or a community.

Tags stop in May while commits continue into August

The maintenance picture is a little uneven and worth reading carefully before pinning anything. Three releases are listed, all of them in the first half of 2026: a version that removed references to a partner, a version that added a note about a mastery programme, and a version that updated a call to action to point at a payment page, dated 2026-05-12. The repository is not archived, and the last push is dated 2026-08-06.

So the tags stop roughly three months before the last commit. That gap can mean several things, and the repository does not say which applies. It can mean the recent work touched content that the CLI does not version, such as the website or the bonus material. It can mean releases are cut on a separate cadence from commits. Or it can mean the project moved to a workflow that does not tag at all.

The practical consequence is narrow but real. If you install by package manager you get whatever the published package is, which is closer to the May state than the August state. If you want the current content, the repository is the fresher artefact, which is the reverse of the usual relationship. That is also why the project tells you not to clone: the paste prompt resolves to the live catalogue rather than to a tag, so the delivery path and the release history point at different things.

The hosted homepage in the repository metadata is a single Vercel deployment, while the individual courses are described as living on their own domains and, in the case of Antigravity, on its own site. Splitting five courses across six hosts plus a home page is a lot of surface for a project that positions itself as simple, and it is worth checking which of those hosts you are actually talking to before you assume they share an owner or a content pipeline.

Editorial conclusion

Read this repository if you are a product manager who would rather practise a tool inside the tool than watch a video about it, and if you can accept that the lessons are delivered by an assistant you install yourself rather than by a course platform with progress tracking. It is a poor fit for corporate training programmes, because the CC BY-NC-ND 4.0 terms permit viewing and sharing with attribution but not modification, so you cannot fork a lesson for your own team's onboarding. Before committing, read the Antigravity entry, which is the one course that does not run through the CLI yet, and check the gap between the last tag in May and the last commit in August before assuming the pinned install matches the current content.

Frequently asked questions

Official sources

1. carlvellotti/free-ai-courses on GitHub
2. Issues
3. Project website
4. README
5. Releases