wrf-run

Claude Code Skill for submitting WRF simulations using natural language.

What It Does

• Parses natural language into WRF namelist parameters
• Generates namelist.input and namelist.wps
• Runs WPS (geogrid → ungrib → metgrid) and WRF (real.exe → wrf.exe)
• Supports both local execution and remote server via SSH
• Archives output per trial for traceability

Quick Start

1. Clone into your Claude Code skills directory or project
2. In Claude Code, type: /wrf-run
3. On first run, answer the configuration questions:
   • Local or remote server?
   • Server address (if remote)
   • Paths to WRF, WPS, ERA5 data, GEOG data
4. Then submit a simulation:

   "Run WRF for Chengdu area, 2025-01-08 00:00-03:00, 4 domains 27km/9km/3km/1km"

Prerequisites

• WRF + WPS compiled on the target machine (use /wrf-install if needed)
• ERA5 reanalysis data in grib format
• WPS_GEOG static data
• Python 3.8+

Remote server (additional)

• ssh-skill installed: ~/.claude/skills/ssh-skill/
• SSH access configured (key or password)

Configuration

All settings in config/user_config.json:

{
  "mode": "remote",
  "ssh_alias": "wrf-server",
  "paths": {
    "wrf_dir": "/home/user/Build_WRF/WRF",
    "wps_dir": "/home/user/Build_WRF/WPS",
    "era5_dir": "/home/user/Build_WRF/data/ERA5",
    "geog_data_path": "/home/user/Build_WRF/WPS_GEOG/geog"
  }
}

See config/user_config.example.json for all options.

Files

wrf-run/
├── SKILL.md                 # Entry point for Claude Code
├── scripts/
│   ├── executor.py          # Local + Remote execution backends
│   ├── config_generator.py  # Namelist generation
│   └── provider_wrf.py      # WPS + WRF pipeline
├── config/
│   ├── user_config.example.json
│   └── templates/           # Namelist .tmpl files
└── examples/

Version

• V1: WPS + WRF execution only
• Future: validation, parameter optimization