dsh-science

dsh-science brings a Claude Science-style research workbench to DeepSeek Harness, turning scientific work from ad-hoc chat into a persistent, traceable process. It organizes questions, hypotheses, experiments, observations, analyses, and conclusions into a research loop stored in a project manifest, making it easier to continue work across sessions and justify each step.

The plugin adds versioned artifacts with provenance, so results can be saved with file hashes, commands, inputs, and environment details, then compared, verified, deprecated, or reproduced. Together with project structure, scientific review skills, and literature connections, it helps users maintain an evidence trail while running analyses, writing manuscripts, and validating claims.

For compute-heavy science, dsh-science includes an SSH-based remote compute engine and host management interface for workstations and HPC clusters. It can submit long-running jobs, monitor their status, fetch outputs, and enforce project-level access approvals. Combined with a companion model tier router that separates light, default, and strong tasks, it is aimed at genomics, pathogen, human health, and bioinformatics teams that need reproducible research workflows and remote compute control.