Lammps Skill — AI skill for Claude Code

lammps-skill

An AI-agent skill, verified Python toolkit, clean-room teaching engine (`mdlite`), thirteen executed chapter notebooks and a thirteen-lecture undergraduate course for molecular dynamics with LAMMPS (latest stable release primary; `develop` watched). See `CHANGELOG.md` for version history.

What it does

• lammpskill detects and drives every way LAMMPS can be installed — WSL apt, a CMake source build, conda-forge, the PyPI wheel, the Windows installer (Docker and a remote cluster are designed and pinned) — through one Installation contract, and measures what each route gives (references/platforms.md). It builds input scripts from a Python Spec, checks them for the mistakes the manual warns about (14 rules, each citing its manual page), runs LAMMPS through a hidden subprocess or in-process through the official python module with the same Result, and reads and writes data, dump, log, restart-header and EAM potential files with its own code — validated against files LAMMPS wrote. Analysis (block averages, RDF, MSD/diffusion, VACF, S(k), energy drift, elastic constants) and benchmark comparison are independent of the binary; ASE, pymatgen, MDAnalysis and OVITO are optional bridges, never required. lammpskill.viz renders headless atom snapshots and animated GIFs (ASE + matplotlib, needs the [ase] extra).
• mdlite is a numpy molecular-dynamics engine written from the textbooks: cell and Verlet lists, Lennard-Jones, harmonic bonds, EAM from setfl/funcfl tables, velocity Verlet with Berendsen, Langevin and Nosé–Hoover-chain thermostats, steepest descent and FIRE. It is cross-checked against LAMMPS at measured precision and against NIST reference data (references/benchmarks.md).
• SKILL.md holds the agent workflows; references/ what the agent must know; docs/USER_MANUAL.md the manual; AGENTS.md every flag. chapters/ holds thirteen executed notebooks (generated from `build/