Positronick's home page just got a lot more honest.
We rebuilt the landing page around one idea: trust you can check beats trust you're told. Here's what shipped.
We rebuilt the landing page around one idea: trust you can check beats trust you're told. Here's what shipped.
Receipts, not adjectives. Every listing links to its real source and a verified author — traced to a real GitHub identity, not a self-declared badge. Anything you install through Positronick (souls and skills) carries a real install count from our public API and a content hash, so what you install is what was published.
Installs, not stars. Star counts get gamed. A CMU study accepted at ICSE 2026 found roughly 6 million suspected fake GitHub stars spread across more than 18,600 repositories — and AI/LLM repos were the single largest category receiving them (source: https://cmustrudel.github.io/papers/icse2026fakestars.pdf). That's why we lead with install counts and content hashes instead of a star badge: a number you can fake shouldn't be the number you trust.
Neutral by design. No model provider or framework is ranked above another on Positronick, and that isn't a policy we might revisit later — it's the whole point of a shared registry.
Machine-readable, not just human-readable. The full catalog is queryable at /.well-known/ai-catalog.json and llms.txt. Point your agent at it directly — you don't have to take our word for anything in this post either.
Curated, not open-submission. Today the catalog is vetted before anything lists. We're not a security scanner, but every listing traces to a real source and a verified author — not an anonymous upload. If you see your project listed, claiming your profile is free: verify your GitHub handle and it's yours.
Try it: curl https://positronick.com/api/souls/<slug>.md > ~/.hermes/SOUL.md