Alistair Turcan

Modern scientific ecosystems are rich in procedural knowledge across repositories, APIs, scripts, notebooks, documentation, databases, and pa-pers, yet much of this knowledge remains fragmented across heterogeneous artifacts that agents cannot readily operationalize. This gap between abundant scientific know-how and usable agent capabilities is a key bottleneck for building effective scientific agents. We present SKILLFOUNDRY, a self-evolving framework that converts such resources into validated agent skills, reusable packages that encode task scope, inputs and outputs, execution steps, environment assumptions, provenance, and tests. SKILLFOUNDRY organizes a target domain as a domain knowledge tree, mines resources from high-value branches, extracts operational contracts, compiles them into executable skill packages, and then iteratively expands, repairs, merges, or prunes the resulting library through a closed-loop validation process. The current library snapshot contains 286 skills across 27 domains and 254 subdomains mined from 394 resources. Under the current search based comparison against SkillHub and SkillsMP, 71.1% of the skills have no retrieved match that completes the same task. The library improves the same coding agent on five of the six MoSciBench datasets. On broader evaluations, SKILLFOUNDRY raises SciCode sampled test accuracy from 65.43% to 71.60% and ScienceAgentBench success rate from 30.95% to 50.00%, outperforming EvoSkill on the shared held out test set without task specific training. We further show that SKILLFOUNDRY can design task specific skills that substantially improve cell type annotation and the scDRS workflow. Together, these results show that automatically mined skills improve agent performance on benchmarks and domain specific tasks, expand coverage beyond hand crafted skill libraries, and provide a practical foundation for more capable scientific agents. Code and an interactive website dashboard for exploring the skill library, domain knowledge tree, and mined resources are available at https://github.com/ma-compbio-lab/SkillFoundry and https://ma-compbio-lab.github.io/SkillFoundry/.