InChI Skill Set

A skillset for teaching AI agents to generate valid InChI identifiers from MDL MOL files using pure algorithms — without external cheminformatics libraries.

What Is This?

This project contains attempts at creating an InChI skill set for AI agents. There are currently 3 attempts:

• mol-to-inchi-ver1/ — First attempt at the skill
• mol-to-inchi-ver2/ — Second attempt (refined approach)
• mol-to-inchi-ver3/ — Third attempt (skill-creator and python code allowed, self-improving skill didnt work)

Project Structure

InChI-skill-set/
├── mol-to-inchi-ver1/      # First attempt
├── mol-to-inchi-ver2/      # Second attempt
├── mol-to-inchi-ver3/      # Third attempt
├── src/formula             # Python code for hill formula
├── data/                   # .mol and .inchi files
└── README.md

Data Folder

The data/ folder contains paired .mol molecule files and their corresponding .inchi reference files for testing and validation.

Docs Folder

The docs/ folder contains documentation about the InChI codebase.

Hill formula

The src/formula/ folder contains AI generate Python code to see whether code can be generate to replicate the InChI C code for the Hill formula.

CRITICAL: No External Tools

This skillset teaches generation without cheminformatics libraries. Do NOT use:

• OpenBabel
• Datamol
• RDKit
• CDK
• Python chemistry libraries
• InChI binaries

Only implement the algorithms described in the skill documentation.

License

MIT