GPUMD-skill

Route NEP requests to task-specific subskills. NEP (Neuroevolution Potential) is the native machine-learning potential family of the GPUMD ecosystem — analogous to DeePMD-kit for LAMMPS. Use when the user asks for `nep.in`, `train.xyz`, `test.xyz`, NEP training, NEP89 reuse, prediction mode, fine-tuning, dipole / polarizability auxiliary models, or automation via NepTrain / NepTrainKit.

Train a first NEP potential from labeled extxyz data. Use when the user needs `nep.in`, `train.xyz`, `test.xyz`, parameter guidance, loss.out interpretation, or deployment of the resulting `nep.txt` back into GPUMD. NEP is the native machine-learning potential for GPUMD and plays the role that DeePMD plays for LAMMPS.

Route VASP DFT requests to task-specific subskills based on user intent. Use when the user asks for VASP workflows and you must decide between static SCF, relaxation, DOS, or band-structure task preparation. This orchestration skill does not own detailed input generation logic; it dispatches to the correct VASP subskill and enforces consistent handoff to submission skills.

Prepare VASP static SCF input tasks from a user-provided structure and essential DFT settings. Use when the user needs single-point electronic structure/total-energy calculations with INCAR generation, KSPACING-based k-point policy (or explicit KPOINTS on request), and POTCAR mapping instructions.

Route ABACUS requests to task-specific subskills based on user intent. Use when the user asks for any ABACUS DFT calculation and you need to determine whether the task is SCF, relaxation, MD, or electronic analysis.

Prepare ABACUS single-point (static SCF) task inputs from a user-provided structure and essential DFT settings. Use when the user needs total-energy/electronic SCF evaluation with explicit ABACUS INPUT/STRU/KPT generation, pseudopotential + orbital mapping, and basis-type selection (PW or LCAO).

Generate Quantum ESPRESSO DFT input tasks from a user-provided structure plus user-specified DFT settings. Use when the user wants to prepare QE calculations such as SCF, NSCF, relax, vc-relax, MD, bands, DOS, or phonons starting from a structure file or coordinates together with pseudopotentials, functional choice, cutoffs, k-point settings, smearing, spin/charge, and convergence parameters. This skill prepares the QE task only; use a separate submission skill such as dpdisp-submit to submit the generated task.

Route CP2K requests to task-specific subskills based on user intent. Use when the user asks for CP2K workflows and you must decide between static, relaxation, molecular dynamics, or electronic-analysis preparation. This orchestration skill dispatches to the correct CP2K subskill and enforces consistent handoff to submission skills.

Prepare CP2K single-point (static) task inputs from a user-provided structure and essential DFT settings. Use when the user needs total-energy/electronic SCF evaluation with explicit CP2K basis/potential and SCF controls.

Prepare VASP band-structure workflow inputs from existing SCF context and user-specified band-path settings. Use when the user requests electronic band-structure calculations and needs explicit prerequisite checks, line-mode KPOINTS path setup, and stage-specific INCAR preparation.

Prepare VASP DOS workflow inputs from existing SCF artifacts and user-specified DOS settings. Use when the user requests total/projected DOS setup and needs INCAR/KPOINTS preparation with explicit prerequisite checks against prior SCF runs.

Prepare VASP geometry-relaxation input tasks from a user-provided structure and essential DFT settings. Use when the user needs ionic or cell-coupled relaxation and requires explicit ISIF-driven relaxation intent mapping, INCAR generation, and POTCAR mapping instructions.

Prepare and run equilibrium GPUMD molecular dynamics. Use when the user needs a `model.xyz`, a `run.in`, an ensemble choice (NVE/NVT/NPT), timestep and dump-cadence guidance, or interpretation of `thermo.out` and `movie.xyz`. This is the general-purpose MD subskill; specialized observables (phonons, transport, diffusion, elastic, mechanics) have their own subskills.

Route GPUMD requests to task-specific subskills. GPUMD is a GPU-accelerated molecular dynamics code that pairs naturally with NEP machine-learning potentials. Use when the user asks for GPUMD MD, `model.xyz`, `run.in`, harmonic phonons via `compute_phonon`, thermal transport via EMD/HNEMD/NEMD, diffusion and ionic conductivity, elastic constants, melting, friction, or deposition workflows.

Tooling layer for GPUMD helper repositories, format converters, dataset curation, and local example discovery. Use when the user needs GPUMDkit, upstream `GPUMD/tools`, GPUMD-Tutorials lookup, DFT-to-extxyz conversion, frame selection, dataset splitting, or bootstrapping the local GPUMD tool-source ecosystem.

Run ReacNetGenerator on reactive MD trajectories to generate reaction networks and reports. Use when the user wants to analyze LAMMPS dump/xyz/bond trajectories with ReacNetGenerator. Handles LAMMPS dump quirks like x/y/z vs xs/ys/zs by converting to x/y/z (orthorhombic + triclinic supported via reacnet-md-tools). Can infer atomname order from a LAMMPS data file. Runs via local reacnetgenerator if available or via `uvx --from reacnetgenerator ...`. Writes outputs into `out/<input_basename>/` with logs and a summary.

A command-line utility for converting and manipulating over 50 atomic simulation data formats, including outputs from DFT and MD software (VASP, LAMMPS, Gaussian, QE, CP2K, ABACUS, etc.). USE WHEN you need to convert structural or trajectory files between different computational chemistry formats, or when parsing raw simulation outputs into structured training datasets (e.g., deepmd/raw, deepmd/npy, deepmd/hdf5) for DeePMD-kit.

A versatile CLI tool for converting molecular file formats, generating 3D atomic coordinates from SMILES, rendering 2D chemical structure images, and preparing or extracting structures for computational workflows. USE WHEN you need to convert between chemical file formats (e.g., xyz, pdb, mol, smi, gjf), generate 3D structures from SMILES using `--gen3d`, render molecule images (PNG/SVG), or extract geometries from simulation logs to build new inputs.

A tool for generating initial packed molecular configurations (XYZ format) from single-molecule structures by calculating box dimensions, writing input scripts, and executing Packmol. USE WHEN you need to randomly pack a specific number of molecules into a simulation box (defined by target density or fixed lengths) to create starting geometries for molecular dynamics or related computational chemistry workflows.

Structure manipulation and crystal analysis workflows based on pymatgen. USE WHEN you need to read/write common atomistic formats (CIF, POSCAR, XYZ), build supercells, perform site substitution/doping, inspect symmetry (space group), or compute local structure descriptors for materials tasks.

A standardized CLI wrapper for RDKit 3D/2D conformer generation that samples multiple conformers per molecule (ETKDGv3, default 10), optimizes each with a force field (MMFF94s/UFF), keeps the lowest-energy conformer, automatically falls back to 2D layout on total embedding failure with a printed warning, and writes results to SDF or XYZ format. USE WHEN you need to generate 3D (or 2D fallback) molecular geometries from SMILES datasets (.csv/.smi) for downstream tasks such as docking, visualization, or 3D-descriptor computation.

Run Shell commands as computational jobs, on local machines or HPC clusters, through Shell, Slurm, PBS, LSF, Bohrium, etc. USE WHEN the user needs to submit batch jobs to a cluster, run commands on a remote server, execute tasks via job schedulers (Slurm, PBS, LSF), or safely run long-term/background shell commands that require state tracking and auto-recovery.

USE WHEN requesting core chemical structural data (SMILES, formula, mass, 2D images) via IUPAC, common, or multilingual names. You MUST actively retrieve the data using this skill; DO NOT hallucinate or generate structures yourself. DO NOT USE WHEN asking for physical properties (melting point, solubility), safety/toxicity data (MSDS), or synthesis pathways.

It is a specification for semantic workflows used by agents to plan, generate, formalize, summarize, and execute complex tasks, projects, experiments,and research efforts for agents, requiring explicit structure, lazy loading,scoped context, evidence-grounded routing, and human review at critical checkpoints. USE WHEN the user asks for a complex task, project, experiment, or research effort that needs to be carefully planned before execution USE WHEN the user provides a text-based plan and wants it to be made more detailed and formalized according to this specification. USE WHEN the user asks to summarize ongoing or completed work into a reusable workflow manifest. USE WHEN the user specifies the location of an existing agent workflow and wants it loaded and executed according to the specification.

General phonon-workflow skill built around phonopy, independent of force backend. USE WHEN you need to prepare finite-displacement phonon calculations, build force constants, and analyze phonon properties (band structure, DOS, thermal quantities) while obtaining forces from different engines such as VASP, Quantum ESPRESSO, or ML force fields.

Prepare CP2K electronic-analysis task inputs from prior converged context. Use when the user requests post-ground-state electronic analyses and needs prerequisite-aware setup.

Prepare CP2K molecular-dynamics task inputs from a user-provided structure and MD controls. Use when the user needs finite-temperature trajectories with explicit ensemble, timestep, and thermostat/barostat settings.

Prepare CP2K geometry-relaxation task inputs from a user-provided structure and optimization settings. Use when the user needs ion-only or cell-coupled optimization with explicit optimizer and convergence controls.

Prepare NEP prediction and fine-tuning workflows from an existing model or foundation model such as NEP89. Use when the user wants out-of-the-box evaluation, targeted MD sampling, `prediction 1`, or `fine_tune` from an existing `nep.txt` + `nep.restart`.

Prepare GPUMD workflows for self-diffusion, ionic conductivity, and viscosity. Use when the user needs `compute_msd`, `compute_sdc`, `compute_viscosity`, Nernst-Einstein ionic conductivity, Arrhenius fitting, or species-selective diffusion through group indices.

Prepare GPUMD elastic-constant calculations using the strain-fluctuation method. Use when the user needs `compute_elastic`, the full `C_ij` tensor, or elastic moduli (bulk, shear, Young's) from an anisotropic NPT trajectory.

Prepare GPUMD mechanics-type workflows: friction, deposition, impact, and other group-based interface simulations. Use when the user needs `add_spring`, ghost-atom setups, layered 2D material shearing, deposition event sampling, or impact/collision dynamics.

Prepare GPUMD harmonic phonon-dispersion calculations using `compute_phonon`, `kpoints.in`, and supercell replication. Use when the user needs harmonic phonons, `omega2.out`, `D.out`, phonon DOS, or a GPUMD-based starting point for lattice-dynamics analysis.

Prepare GPUMD thermal-transport workflows for EMD / HNEMD / NEMD / SHC / HNEMDEC. Use when the user needs `compute_hac`, `compute_hnemd`, `compute_hnemdec`, `compute_shc`, thermal-conductivity extraction, or transport-specific sampling and convergence guidance.

Prepare dataset-curation and active-learning workflows around NepTrain and NepTrainKit. Use when the user needs perturbation-based sampling, representative-structure selection (FPS / max-min), automated NEP project scaffolding, interactive outlier inspection, or iterative retrain loops rather than a single manual NEP fit.