GitHub Trending's top 3 sum to 1.07M stars — and zero are libraries

What happened

The top three repositories on GitHub Trending today have nothing in common except their position on the page. freeCodeCamp/freeCodeCamp sits at 437.9k stars, kamranahmedse/developer-roadmap at 350.5k, and openclaw/openclaw — a desktop AI agent that bills itself as 'your own personal AI assistant, any OS, any platform, the lobster way' — at 283.1k.

Combined, they account for roughly 1.07 million stars. That's more than Linux, more than the entire React + Vue + Svelte trio, more than TensorFlow. And yet none of them is a library. You cannot `npm install` any of them. You cannot `pip install` any of them. You cannot link against them or import them as a dependency. Two are pedagogical: freeCodeCamp is a coding bootcamp curriculum, developer-roadmap is an interactive set of learning paths rendered largely from markdown and JSON. The third is a standalone end-user application — a chat client wrapping a model, the kind of repo whose value is in its binary, not its source.

This is the entire top of the chart on a major platform that the industry treats as a tech radar. Three slots, three artifacts that no production system depends on.

Why it matters

GitHub stars were originally a proxy for 'I might use this in a project later.' That meaning held when the median repo on Trending was a build tool, a database driver, or a framework. Stars no longer track adoption — they track aspiration. A developer bookmarks freeCodeCamp because they intend to learn JavaScript. They star developer-roadmap because they hope to one day be a senior backend engineer. They star openclaw because they want to vibe-code their own Jarvis on a Saturday. None of those acts predicts that the repo will end up in anyone's `dependencies` field.

The asymmetry is structural. Career-aspiration repos have an audience the size of every developer who has ever considered switching paths. A new ORM has an audience the size of teams currently evaluating ORMs. The first is monotonically growing; the second is a tiny, churning slice. Stars never decay, but appetite for self-improvement compounds across every cohort of new developers entering the industry. A curriculum repo accumulates stars from every person who has ever Googled 'how to become a programmer in 2024,' then 2025, then 2026. A real library is starred mostly by people who happened to be shopping in the week it launched.

The LLM era makes this worse, not better. Projects like openclaw — and the broader cohort of 'personal AI agent' repos that the source describes only as 'the lobster way' — are extraordinarily starrable. Every developer who reads the README thinks 'I should run my own.' Almost none of them actually wire up the model keys, install the desktop runtime, configure the system tray daemon, and live with it for a month. The star-to-actual-user ratio for end-user agent apps is roughly the same as for the kitchen gadgets you bought on Black Friday: bought with enthusiasm, used twice.

This isn't an abstraction. We watched it happen in real time over the last 48 hours on this very feed — three suspicious repos (chokepoint-atlas, reg-factory, ConiferKit/sage) climbed Trending on coordinated star fingerprints and zero code. The platform's signal has been gamed at the low end and captured at the high end. The middle, where actual libraries used to live, is the only band where stars still mean what they used to mean — and almost nothing trends from there anymore.

Community reactions on HN threads about Trending have settled into resignation. The recurring observation: 'I check Trending less than I used to.' The replacement signals senior developers actually use — npm weekly downloads, PyPI installs, GitHub clone counts, Sourcegraph search frequency, package.json prevalence in the public corpus — all paint a wildly different picture of what's hot than the Trending page does. React's 243k stars get outranked by markdown files because React's stars stopped flowing in 2017 and markdown stars are still flowing in 2026.

What this means for your stack

First, stop treating GitHub Trending as a procurement input. If you're picking dependencies off the Trending page, you're letting career-anxiety traffic decide your architecture. The page is optimized to surface what new and aspiring developers are bookmarking, not what production systems are running. Use it for vibes; use npm/PyPI/crates.io download counts for decisions. The npm registry publishes weekly download counts; PyPI exposes them via BigQuery; crates.io has a CSV dump. None of those numbers can be juiced with a throwaway account farm.

Second, if you maintain a library, calibrate your own expectations. Your repo will never out-star freeCodeCamp, and that's fine — it shouldn't. A useful database driver might cap at 5k stars and ship in 80% of Node services. A roadmap repo can pull 350k stars and ship in zero systems. Optimize for adoption signals (downloads, GitHub Insights traffic referrers, issues from real users) and stop benchmarking against the curriculum tier. The two metrics are measuring different markets.

Third, when you encounter a 'personal AI assistant' or 'desktop agent' repo with absurd star counts, ask the boring question: how many active monthly users does it have? The answer is almost never findable in the README. That's the tell. Production libraries publish download graphs because the number is the marketing. End-user apps post screenshots because the actual usage number is embarrassing.

Looking ahead

The gap between starred and used will widen, not narrow. Every new entrant to the industry stars freeCodeCamp and developer-roadmap as a rite of passage; every Saturday-night hacker stars one of the agent repos; the libraries that actually run the internet collect a few hundred new stars a week, if they're lucky. At some point GitHub will either redesign the Trending signal — weighting by recent traffic, fork-to-star ratio, or dependent-repo count — or quietly accept that the page is now a discovery surface for content, not code. Until then, treat the top of the chart the way you'd treat a bestseller list at an airport bookstore: useful as a cultural barometer, useless as a software bill of materials.