AtomisticSkills

What AtomisticSkills does

AtomisticSkills builds research tasks from three levels. Tools are strictly typed Python functions exposed through MCP servers, covering structure relaxation, molecular dynamics in NVT, NPT and NVE ensembles, Monte Carlo cluster expansion, machine learning interatomic potential simulation, and DFT input preparation and output parsing. Skills are flexible tutorials that combine several tool calls, each with a SKILL.md, helper scripts and examples.

Skill examples in the README include benchmarking an interatomic potential against a labeled dataset, computing diffusion coefficients and activation energies, calculating thermodynamic stability and E_hull, docking ligands into a protein with AutoDock Vina, and calculating gas adsorption isotherms with grand canonical Monte Carlo. Workflows sit on top as high-level research goals, such as searching for sorption materials in a given chemical space. The project links a documentation site, a benchmark leaderboard and a preprint.

Key features

• Three-level hierarchy of workflows, skills and tools
• MCP tools for relaxation, molecular dynamics, Monte Carlo and DFT input and output handling
• Skills such as MLIP benchmarking, diffusion analysis and AutoDock Vina docking
• Each skill ships a SKILL.md, scripts and examples
• Separate benchmark repository and leaderboard

When to use it

• Running a multi-stage materials discovery campaign driven by a coding agent
• Benchmarking a machine learning interatomic potential and generating parity plots
• Computing diffusion coefficients or adsorption isotherms through an agent

Who it is for: Computational materials, chemistry and drug discovery researchers who want AI coding agents to run simulation workflows.

How it fits with Hermes Agent

The repository description lists Hermes Agent among the agentic IDEs and agents it integrates with, alongside Cursor, Claude Code, Codex and Google Antigravity.

FAQ