Scientific Agent Skills: 165 Skills and a Telling Rebrand
K-Dense's scientific-agent-skills added 1,600 stars today, its strongest day in two weeks of camping on GitHub trending, now at 37.9k. The repo packs 165 validated skills that turn any coding agent into a working research assistant: 100+ scientific databases like PubChem, ChEMBL and UniProt, 70+ Python packages like RDKit and Scanpy, plus platform hooks into Benchling and DNAnexus. MIT licensed, and K-Dense claims 190,000 scientists use it.
The reason to cover it now isn't just momentum, it's the rebrand. This repo used to be called Claude Scientific Skills. It's now Scientific Agent Skills, explicitly targeting any agent that supports the open Agent Skills standard. That one rename says a lot about where the ecosystem is going: skills started as a Claude Code feature, and they're quietly becoming a vendor-neutral packaging format for expertise β write the skill once, run it under whichever agent wins.
The bet underneath is worth spelling out. Model capabilities converge; what doesn't converge is domain knowledge encoded into runnable form. A skill that knows the quirks of querying ChEMBL correctly is useful under Claude, Codex, or whatever ships next year. The people building skill libraries are betting the moat is above the model, and a rename from a specific vendor to a standard is that bet said out loud.
Repo: https://github.com/K-Dense-AI/scientific-agent-skills
← Back to all articles
The reason to cover it now isn't just momentum, it's the rebrand. This repo used to be called Claude Scientific Skills. It's now Scientific Agent Skills, explicitly targeting any agent that supports the open Agent Skills standard. That one rename says a lot about where the ecosystem is going: skills started as a Claude Code feature, and they're quietly becoming a vendor-neutral packaging format for expertise β write the skill once, run it under whichever agent wins.
The bet underneath is worth spelling out. Model capabilities converge; what doesn't converge is domain knowledge encoded into runnable form. A skill that knows the quirks of querying ChEMBL correctly is useful under Claude, Codex, or whatever ships next year. The people building skill libraries are betting the moat is above the model, and a rename from a specific vendor to a standard is that bet said out loud.
Repo: https://github.com/K-Dense-AI/scientific-agent-skills
Comments