Research workflow and open science tool
CHARMM-GUI System Builder
An AI Agent skill that records CHARMM-GUI input checks, browser jobs, download recovery, and GROMACS file validation as separate steps.
- Current public release
- v2.2.0 Audit Closure and Stage Handoff Edition
- License
- AGPL-3.0
- Status
- Public, unofficial community project
An unofficial community project. It is not affiliated with or endorsed by CHARMM-GUI, the Im Lab, or Lehigh University.
01 / Overview
A finished download is not yet a simulation-ready system.
Preparing a membrane-protein system through CHARMM-GUI involves an input structure, browser forms, backend jobs, downloads, archive extraction, and GROMACS topology checks. A green signal at one step cannot speak for the steps that follow.
This skill records each part separately: which parameters were submitted, where a browser job stopped, whether the download is an archive or an HTML error page, and whether GRO, TOP, ITP, and MDP files agree. It is not a new dynamics engine and does not replace CHARMM-GUI.
02 / Why it exists
Browser jobs, downloaded packages, and topologies fail in different places.
A page can look ready without being submitted; the backend can finish while the browser download fails; a file named `.tar.gz` can contain an HTML error; and readable GROMACS files do not prove that custom ligand parameters reached the final topology.
The skill refuses to cover those differences with one green check. Browser state is supported by browser records, file state by type and integrity checks, and topology state by component inspection. Whether the system should enter production MD remains a separate human decision.
03 / Capabilities
From the original structure to the GROMACS package.
- 01
Audit PDB, SDF, MOL2, and ligand-parameter inputs.
- 02
Capture browser parameters, URLs, screenshots, and job state.
- 03
Check membrane orientation, system dimensions, lipid ratios, and ion settings.
- 04
Distinguish running, stalled, backend complete, and waiting for a human next-step submission.
- 05
Recover interrupted browser downloads while preserving evidence.
- 06
Identify tar, tar.gz, HTML error pages, damaged packages, and incomplete downloads.
- 07
Validate GRO, TOP, ITP, MDP, and system components.
- 08
Check that custom ligand parameters truly enter the final GROMACS topology.
- 09
Use a structural-environment ledger to distinguish bulk membrane material from structure-resolved lipids, ions, cofactors, and pocket components.
- 10
Record pose-preservation evidence across cleaning, orientation, assembly, and final coordinate conversion.
- 11
Verify restraint files, macros, equilibration release schedules, and expected production state while handing unresolved obligations to a later stage.
- 12
Close Stage 2 through separate read-only audit, explicit locking, and independent Receipt 2.0 verification.
- 13
Prevent a technical file pass from being reported as production-ready.
04 / Cross-agent support
The same checks can run through different agent clients.
The public release provides adapters for the following environments while keeping the checklist and takeover conditions the same.
- Codex
- Claude Code
- OpenClaw / Javis
- Hermes Agent
- Other Agent Skills-compatible clients
When a runtime has no browser control or human takeover, the skill stops at the browser step. It does not rewrite 'cannot perform' as 'completed'.
05 / Scientific boundary
Complete files do not authorize production simulation.
Stage 2 checks establish that inputs, records, and output files agree. They do not decide whether protonation, membrane composition, or ligand parameters are scientifically appropriate, and they do not authorize MD simply because a topology can be preprocessed.
- It does not issue Production_Approved automatically.
- It does not run gmx mdrun.
- It does not use gmx grompp -maxwarn to conceal topology problems.
- It does not choose membrane composition, protonation, protein segmentation, or ligand parameters for the researcher.
- It does not describe a prepared structure as binding-site evidence.
production_ready=false · md_execution_allowed=false · no_mdrun=trueBefore production, an expert still reviews molecular identity, protonation, ligand parameters, membrane orientation, topology preprocessing, and the plausibility of the whole system.
06 / Origin and acknowledgement
It grew out of membrane builds that actually became stuck.
The expensive part of a membrane build is often not completing a form once, but knowing how to resume a long job, whether a failed download hides a completed backend task, and whether the package that finally arrived is intact. The project turns those recurring problems into executable checks.
The public release keeps only reusable methods. Specific unpublished molecules, internal job IDs, private structures, local paths, and runtime credentials do not enter the repository or this page.
Developed with research guidance from Nian Yin and computational workflow support from NianLab.
07 / Public links