Rmcprofile Skill — AI skill for Claude Code

rmcprofile-skill

An AI-agent skill and verified Python toolkit for [RMCProfile](https://rmcprofile.ornl.gov/), the Reverse Monte Carlo program for total scattering — with executed chapter notebooks and an undergraduate course to follow in later releases.

RMCProfile itself is not included: download it from the RMCProfile site, unpack it, and set `RMCPROFILE_HOME` to its `RMCProfile_package` directory. Everything here that does not need the binary works without it.

What it does

• Drive an RMCProfile 6.7.9 installation: read and write every input format (.dat keyword blocks, .rmc6f configurations, data files, the Bragg family, .dw), check an input set for the mistakes the manual warns about (and two it does not: the silent .poly wait and the exit-code-0 stop), run the binary with the environment its setup script would set, and parse every output (χ² history, fit CSVs, partials).
• Analyse configurations without the binary: partial g(r), Keen's G(r) in barn, F(Q), coordination numbers, bond-angle distributions, the average cell — validated on exact geometry and on the identity G(r→0) = −(Σ c_i b_i)² that the package's own SF6 data reproduces.
• Cross-check itself against the installed package: stage any shipped exercise, run it, and compare our partials and G(r) with RMCProfile's own CSVs to recorded tolerances.
• Teach with rmclite, a clean-room Reverse Monte Carlo engine whose calculated functions equal RMCProfile's by construction: moves, χ² terms, constraints, a bond potential, Metropolis acceptance — small enough to read. The chapters (one per data type the program fits) and the course follow.

Install

scripts\install_rmcprofile_windows.ps1        # Windows: conda env `rmcprofile` + kernel; -DryRun first
bash scripts/install_rmcprofile.sh            # Linux / macOS / WSL; --dry-run first
export RMCPROFILE_HOME=/path/to/RMCProfile_package    # the directory with exe/ and tutorial/
python scripts/ve