rfd4-proteina-search
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill documentation and referenced files describe a legitimate scientific workflow for protein design. Analysis of the instructions and examples confirms that all operations are restricted to project-specific CLI tools (
rfd4-generate) and local helper scripts. - [EXTERNAL_DOWNLOADS]: The skill mentions that model checkpoints are downloaded over public HTTPS. This is standard behavior for the RFD4-Proteína model and targets trusted scientific repositories. Per security guidelines, downloads from these well-known sources for their intended purpose are considered safe.
- [COMMAND_EXECUTION]: The skill utilizes the
Bashtool to execute a preflight script (scripts/preflight.sh) and a SLURM launch script (scripts/launch_generate.sh). These are local components of the RFD4-Proteína project structure and do not incorporate untrusted remote payloads. - [DEPENDENCIES]: The skill references standard scientific packages and environments, such as
alphafold-colabfoldand specific CUDA-based images. These dependencies are well-documented and sourced from established organizations like DeepMind and RosettaCommons.
Audit Metadata