Skip to main content
Back to timeline
K-Dense-AI/scientific-agent-skillsSource publication:

Scientific agent skills break free of Claude: 177 skills load in any compatible host and can run locally

Synopsis

K-Dense renamed its Claude-oriented scientific skill collection to Scientific Agent Skills and moved it onto the open Agent Skills standard, so any AI agent that supports that standard can load it. The collection holds 177 skills spanning biology, chemistry, medicine, physics, engineering, Earth science and data analysis, each giving the conventions and validation checks for a specific package, data source or workflow. The same organization released K-Dense BYOK, a locally running co-scientist where users bring their own API keys, choose from more than 40 models, keep data on their own machine and can scale heavy workloads to cloud compute via Modal.

Source-provided article image: K-Dense-AI / scientific-agent-skills
Figure 1 · GitHub Trending

Interpretation

The scientific skill collection now reaches any AI agent that supports the open Agent Skills standard, rather than a single host, and its name changed from Claude Scientific Skills to Scientific Agent Skills. Previously the skills were aimed at Claude, so other agents could not directly reuse the same set. The collection follows the open Agent Skills standard and lists Cursor, Claude Code, Codex and Google Antigravity among compatible hosts; it is also a portable Agent Plugins package, so clients supporting that standard can load the whole collection as one plugin.

Each skill writes down how to use a particular scientific package, data source or workflow, including the field's conventions and the validation checks the work needs. An agent can already call any Python package or API on its own, but lacks curated documentation and examples for specific tools. The 177 skills cover bioinformatics and genomics, cheminformatics and drug discovery, proteomics and mass spectrometry, clinical research and evidence workflows, and healthcare AI and biosignal research, with tasks such as single-cell RNA-seq, CRISPR screens, molecular docking, LC-MS/MS processing and clinical trial workflows.

Perspective

The skills target researchers and engineers running research workflows through AI agents, covering biology, chemistry, medicine, physics, engineering, Earth science, data analysis and scientific communication, with each skill addressing one specific package, data source or workflow. K-Dense BYOK is described as running locally on a desktop, with users bringing their own API keys, choosing from more than 40 models, and getting web search, file handling and access to over 100 scientific databases; heavy workloads can scale to cloud compute via Modal.

The source is a project description and gives no success rates on specific tasks or comparisons against other approaches. The healthcare AI and biosignal direction is limited to EHR and model research, physiological signal analysis and retrospective validation, not patient-specific diagnosis, treatment, alarms or deployment decisions. Actual skill performance depends on the host agent and the chosen model.