Stanford and Princeton open LabClaw with 211 skills for biomedical agent workflows

• Stanford and Princeton researchers have open-sourced LabClaw, which the announcement describes as the “Skill Operating Layer for LabOS” for biomedical agent workflows launch repost.
• A practitioner walkthrough says the repo packages 211 skills so an agent can handle multi-step tasks like database lookup, fold analysis, and literature summarization instead of relying on generic tool wiring repo walkthrough.
• The screenshot in the same thread breaks those skills into biology, pharmacy, medicine, literature, vision, LabOS, and general categories, with the repository released under an MIT license repo screenshot.
• For engineers building AI scientist systems, the practical change is a reusable skill library sitting between a reasoning model and an execution layer, rather than having to assemble the “last mile” from scratch architecture take.

LabClaw is now available as an open-source repository GitHub repo and is positioned by its launch post as the “Skill Operating Layer for LabOS” launch repost. The project is aimed at “dry-lab reasoning, protocol composition, and agentic workflows,” according to the

, which frames it as the layer that connects model reasoning to concrete biomedical actions.

The same

says LabClaw includes 211 “production-ready SKILL.md files” spanning biology, lab automation, vision/XR, drug discovery, medicine, data science, and literature research. The category counts shown there include 66 biology skills, 36 pharmacy, 20 medicine, 29 literature, 5 vision, 7 LabOS, and 48 general skills, with the repo marked MIT-licensed repo screenshot.

The clearest implementation detail in the evidence is the workflow example from the thread: an agent can be told to “find this gene sequence,” “run a fold analysis,” and “write a summary” of related clinical trials, with the skills telling the system “which buttons to push and which APIs to call” repo walkthrough. That makes LabClaw less like a model release and more like an operational tool layer for orchestrating domain-specific actions.

The same thread argues the hard part in AI-for-science is the “last mile” and describes the emerging stack as a reasoning core, a skill library like LabClaw, and an execution layer like LabOS architecture take. That architecture claim is still a practitioner interpretation, not a benchmark, but it gives engineers a concrete starting point for building biomedical agents around a prebuilt skill inventory instead of a generic function-calling scaffold.