AI analysis grounded in the code graph — computed facts, not vibes · 2026-09-08T03:03:19Z
This is a collection of 163 "skills" — self-contained folders of scripts, documentation and prompts — that plug into AI coding agents (Cursor, Claude Code, Codex, Antigravity, Pi) via the open Agent Skills standard. Each skill wraps a specific scientific tool or database (e.g. scanpy, stable-baselines3, pymc, PubMed search, imaging-data-commons) so an agent can execute domain-specific workflows — model diagnostics, literature search, schematic generation — rather than improvising from scratch. It's aimed at scientists and engineers who want an AI agent to reliably perform genomics, drug-discovery, clinical or ML research tasks.
The star growth (498 in a single day, 35,390 total) lines up with active README-driven promotion rather than a single technical breakthrough: recent commits are dominated by README rewrites (webinar announcements, blog-entry additions, a Star History chart, Reddit links) and version-badge bumps tracking the growing skill count. The project also rebranded from "Claude Scientific Skills" to "Scientific Agent Skills" to broaden compatibility beyond Claude, which is an explicit hook for wider adoption across agent platforms. The README's direct appeal ("please star this repository") and cross-posting to X/LinkedIn/YouTube suggests growth is substantially marketing-driven, on top of genuinely expanding content (147→163 skills across releases v2.60.0–v2.64.0).
What changed recently, how it's actually built (from the code graph), and whether you should care. Free account — no card, no spam.