mlff-molecular-screening

name	mlff-molecular-screening
description	Use this skill for MLFF single-point screening or pre-relaxation of molecules, conformers, and clusters before xTB or ORCA validation.
license	project-local
allowed-tools	ls read_file write_file edit_file execute get_avail_remote_task get_remote_task_spec remote_submission remote_submission_batch

mlff-molecular-screening

Overview

Use an enabled molecular-capable MLFF backend for fast triage without replacing quantum-chemistry evidence.

Quick Start

1. Put uniquely named molecular structures directly under a clean input/.
2. Query mlff_sp or mlff_relax for the selected molecular-capable backend in one call, using template_overrides={"backend": "<enabled-backend>"} and detail="full"; do not infer provider fields from another backend's schema.
3. For FairChem UMA, use omol and set charge plus multiplicity-style spin for every structure.
4. Submit a stage or batch of complete stages, then inspect all per-input summaries/errors.
5. Send the shortlist to xTB/ORCA with charge, multiplicity, geometry, and provenance intact.

Allowed tools

• ls
• read_file
• write_file
• edit_file
• execute
• get_avail_remote_task
• get_remote_task_spec
• remote_submission
• remote_submission_batch

Workflow

1. Choose a molecular-capable backend

• FairChem UMA omol is the initially validated molecular route. Do not use periodic UMA domains for isolated molecules.
• Query current availability; do not assume every remote deployment installs the same providers.

2. Preserve physical metadata

• Put shared UMA metadata in backend_config.defaults and per-file exceptions in backend_config.items.
• Use exact paths relative to input/. Preserve charge and spin when handing selected geometries to xTB/ORCA.

3. Group only comparable structures

• One stage reuses model loading. Group conformers or similarly sized candidates, and split heterogeneous or large relaxations into balanced stages.
• Use SP for ranking unchanged geometries and relax only when preoptimization is intended.

4. Escalate to quantum chemistry

• MLFF does not replace ORCA frequencies, TS/IRC, thermochemistry, excited states, NMR, or final spin-state energetics.
• Report the MLFF shortlist criterion and the reference method still required.

Method-critical defaults

• UMA omol requires explicit charge and multiplicity-style spin; common values include spin 1 for singlet and spin 3 for triplet.
• Keep backend/model and relaxation settings fixed across candidates being ranked.
• Treat MLFF energy gaps as screening values until checked at the requested xTB/ORCA level.

Output Contract

Return backend/model, operation, charge/spin mapping, stage/batch path, receipt context, batch summary, shortlist, and the planned xTB/ORCA validation.

References

• Canonical stage layout: skills/execution/remote-stage-layouts/SKILL.md#mlff_sp-and-mlff_relax
• Full UMA domain notes: skills/materials_worker/mlff-screening-and-relaxation/references/fairchem_uma.md