Research and Scientific Analysis Workflows

Problem

General prompting is often too weak for research-heavy work: an agent may misuse RDKit, Scanpy, or scikit-learn, or lack domain constraints across platforms such as Benchling, DNAnexus, and OMERO. This skill packages curated scientific workflows, database access, and domain guidance into Agent Skills for compatible harnesses such as Claude Code, Cursor, and Codex.

How It Works

Rather than exposing only tools, it gives the agent specialized context:
- 70+ Python package skills: RDKit, Scanpy, PyTorch Lightning, BioPython, pyzotero, PennyLane, Qiskit, OpenMM, MDAnalysis, scVelo, TimesFM, and other scientific and data-science libraries.
- 9 scientific integration skills: Benchling, DNAnexus, LatchBio, OMERO, Protocols.io, Open Notebook, and other scientific platforms or APIs.
- Workflow constraints: from literature/data reading and model analysis to result cleanup, reducing unpredictable outputs caused by open-ended prompting.

Boundaries

The skill assumes a working Python environment, required scientific libraries, database permissions, and local dependencies. Because Agent Skills may run code, install packages, access networks, or modify files, use it in an auditable and isolated workspace, and review upstream security checks and Cisco AI Defense Skill Scanner scan results before relying on outputs.