HPC-Skills: Full-Stack AI Agent Skill Set for Scientific Computing, Covering CFD to Molecular Dynamics

HPC-Skills provides a portable AI Agent skill set that supports workflow automation for mainstream HPC software like OpenFOAM, SU2, LAMMPS, and GROMACS, covering full scenarios of high-performance computing including MPI parallelism, GPU acceleration, and cluster orchestration. It aims to simplify complex HPC workflows, lower the entry barrier for researchers, and facilitate research and innovation in the scientific computing field.

High-performance computing (HPC) is the core infrastructure for modern scientific research and engineering design. However, traditional HPC workflows involve extensive professional software configuration, compilation, operation, and post-processing. Each software has unique syntax, and the threshold for parallel technology is high. The rise of AI Agent technology provides new possibilities for simplifying workflows. Based on this vision, the HPC-Skills project offers a comprehensive AI Agent skill toolset for the scientific computing field.

Skill Design Philosophy: Follow principles of portability (unified calling interface), composability (skill building block combination), reproducibility (containerization + package management to ensure consistent environments), and intelligence (contextual decision optimization).

Parallel Computing and Acceleration: Support MPI parallelism (automatically handle process mapping and load balancing), GPU acceleration (CUDA/OpenCL support and migration guidance), and heterogeneous computing (intelligent allocation of CPU+GPU resources).

Cluster Orchestration Management: Compatible with scheduling systems like Slurm/PBS; automatically query cluster status and submit jobs; intelligent resource request strategies to avoid waste; support cross-platform migration in multi-cloud environments.

Software Ecosystem Coverage: CFD field (OpenFOAM/SU2/FEniCS), structural mechanics (CalculiX/ElmerFEM), molecular dynamics (LAMMPS/GROMACS), quantum chemistry (Quantum ESPRESSO/VASP/Gaussian), as well as a complete ecosystem including numerical libraries, pre/post-processing tools, and package managers.

Workflow Automation: Built-in mesh generation (automatic selection of tools like Gmsh), solver configuration (recommendation of empirical parameter libraries), result analysis (visualization + key indicator extraction); support Git/DVC version control and collaboration.

Applicable in fields such as aerospace (aerodynamic simulation/shape optimization), automotive industry (collision/aerodynamic analysis), materials science (multi-scale simulation), biomedicine (protein folding/drug development), energy (wind energy/nuclear energy optimization), climate science (regional climate models), etc. Educational institutions can lower the threshold for HPC teaching and provide example tutorial resources.

Community Ecosystem: Open-source project with active contributors improving the skill set and strict code review; provide detailed documentation, tutorials, case studies, and regular seminars; collaborate with supercomputing centers, vendors, and academic institutions to ensure synchronization.

Future Directions: Introduce AI capabilities like reinforcement learning/graph neural networks; expand support for quantum computing/edge computing software; strengthen cloud-native technology adaptation; develop visual workflow editors and enhanced natural language interfaces; promote HPC democratization.

HPC-Skills makes complex HPC workflows accessible through AI Agent technology. Its extensive software support, intelligent parallel optimization, and cluster orchestration capabilities make it a powerful assistant for researchers and engineers. As AI and HPC deeply integrate, the project will continue to evolve and provide stronger momentum for scientific discovery.