What is multiwfn-skills?

MulTiwFn-skills is a collection of Codex skills for computational chemistry, built around Multiwfn, a program used to analyze molecular wavefunctions. It lets an AI coding assistant like Codex automatically find wavefunction files, structure files, and rendering programs on your machine, run Multiwfn to generate cube data from them, and then produce publication quality molecular images using VMD with its Tachyon renderer, or ChimeraX as a fallback. To install it you clone the repository and copy the full skill folders, not just their SKILL.md files since each one also depends on its own scripts, reference data, and agent files, into Codex's skills directory, then restart your Codex session. You separately need Multiwfn itself installed on your machine, and either VMD or ChimeraX for rendering, with VMD used by default for its white background, orthographic, lit images. The repository provides four skills. One turns a SMILES string into a 3D structure, asks for any missing calculation parameters, recommends a method and basis set from a curated set of benchmark papers, and writes Gaussian or ORCA input files. Another prepares aromaticity and delocalization analyses, supporting methods like NICS, HOMA, and FLU among others, and can generate a Gaussian NMR template when one is needed. A third generates cube files for specific molecular orbitals such as HOMO and LUMO and renders front and side view images of each. The fourth generates an electrostatic potential surface colored by charge from formats like fchk, molden, wfn, or wfx. Each pipeline can be called directly from the command line with a range of options, or invoked inside Codex using a short skill command. Every run writes a manifest file recording which files and programs it found, the exact commands it ran, and any warnings, and the README recommends against committing large calculation outputs to the repository since its gitignore already excludes common ORCA outputs and cube files. The README also notes some limitations plainly: the SMILES to 3D step only produces one conformer, charge and spin multiplicity are never guessed automatically, and users of Multiwfn are asked to cite the relevant paper in their own work.