What it does

LabClaw is a collection of 240 SKILL.md files—structured prompts that tell an OpenClaw-compatible agent when to use a tool, how to call it, and what output to produce. The skills cover biology, drug discovery, clinical medicine, statistics, literature search, and visualization. You install the whole repo or copy individual folders into your agent’s workspace.

The interesting bit

The project treats “skills” as the unit of reuse, not code. Each file is essentially a contract between a human expert and an LLM: here’s the tool, here’s the context, here’s the expected behavior. It’s closer to a technical cookbook than a software library, which makes it weirdly practical for labs that already have Python stacks but need their AI agents to stop hallucinating Bioinformatics 101.

Key highlights

• 86 biology skills (single-cell, spatial omics, CRISPR screens, proteomics)
• 36 pharmacy/drug discovery skills (RDKit, DiffDock, cheminformatics pipelines)
• 22 clinical skills (trials, precision oncology, medical imaging)
• 54 general data-science skills (scikit-learn, statistics, reproducibility)
• 33 literature skills (PubMed, patents, grants, citation management)
• Designed for OpenClaw runtime; one-line install via install https://github.com/wu-yc/LabClaw

Caveats

• Requires the separate OpenClaw runtime to actually function; the repo itself is just markdown
• Some skill counts in the collapsible catalog don’t match the summary tables (66 vs. 86 in biology), suggesting the README may be slightly out of sync
• Crypto donation addresses prominently displayed with a disclaimer that “all memecoins are scams”—a choice

Verdict

Worth bookmarking if you’re building or operating biomedical AI agents and need structured, domain-specific tool descriptions without adopting another heavy framework. Skip it if you’re looking for executable code or a standalone system.