SciAgent-Skills is a specialized plugin that boosts Claude Code (Opus 4.6) performance on the BixBench-Verified-50 bioinformatics benchmark from 65.3% to 92.0%.
The system utilizes 196 structured Markdown files covering genomics, drug discovery, and biostatistics. Unlike traditional RAG systems, it eschews complex vector databases and embedding pipelines in favor of simple on-disk files. The agent intelligently scans skill descriptions during planning and loads full technical details—including runnable code, parameters, and troubleshooting tables—only when needed.
With 72 toolkits, 53 database connectors, and various analysis pipelines, it can be easily installed by cloning the GitHub repository and using the --plugin-dir flag.
SciAgent-Skills is a specialized plugin that boosts Claude Code (Opus 4.6) performance on the BixBench-Verified-50 bioinformatics benchmark from 65.3% to 92.0%.
The system utilizes 196 structured Markdown files covering genomics, drug discovery, and biostatistics. Unlike traditional RAG systems, it eschews complex vector databases and embedding pipelines in favor of simple on-disk files. The agent intelligently scans skill descriptions during planning and loads full technical details—including runnable code, parameters, and troubleshooting tables—only when needed.
With 72 toolkits, 53 database connectors, and various analysis pipelines, it can be easily installed by cloning the GitHub repository and using the --plugin-dir flag.