The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones the official ProteinMPNN repository from GitHub (https://github.com/dauparas/ProteinMPNN.git) and installs standard dependencies (torch, numpy) using pip.
[REMOTE_CODE_EXECUTION]: The skill executes protein_mpnn_run.py from the cloned repository. This is the official implementation of the ProteinMPNN model as described in published research.
[COMMAND_EXECUTION]: The skill uses shell commands to set up the environment, clone the repository, and run the protein design scripts.
[DATA_EXFILTRATION]: The skill processes local protein backbone files (PDB format) and user-provided JSONL files for fixed positions. The resulting amino-acid sequences are written to a local output folder. No sensitive file access or unauthorized network transmission was detected.
[PROMPT_INJECTION]: The skill contains standard instructional content for using the tool and does not attempt to override agent safety guidelines or behavior.
[DATA_EXFILTRATION]: Evaluated for indirect prompt injection surface.
Ingestion points: Protein structure files (.pdb) and configuration files (.jsonl) loaded via command-line arguments.
Boundary markers: Standard shell argument separation is used; no specialized delimiters for input data are present.
Capability inventory: Includes repository cloning, package installation, and execution of local Python scripts.
Sanitization: Inputs are parsed by the specialized ProteinMPNN engine; no specific sanitization for prompt injection is implemented as PDB/JSONL are not typical natural language injection vectors.

proteinmpnn