Researchers publish 163 procedures for scientific agents

A team led by Timothy Kassis described Scientific Agent Skills, an open library of 163 research procedures spanning 16 areas of practice. The collection covers fields including genomics, cheminformatics, medical imaging, study design and scientific communication.

Each skill centres on a versioned instruction file, sometimes accompanied by reference material and executable scripts. An agent loads the relevant material when a task requires it. The procedures address choices that working code alone does not settle, such as which statistical test to use, which identifier system is authoritative and what caveats belong with a result.

The authors explicitly report no task-level evaluation or measurement of how often agents select the appropriate skill. The library therefore provides reusable procedural knowledge without establishing how reliably agents apply it.

That distinction matters for research automation: a script can execute successfully while the analysis it implements remains methodologically wrong.