The Agent Skills Directory

[COMMAND_EXECUTION]: The file references/patterns.md contains Python patterns that use subprocess.run to execute external bioinformatics tools including colabfold_batch, TMalign, and alphafill. These commands are constructed using lists, which mitigates shell injection risks, but they depend on the presence of these binaries on the host system.
[EXTERNAL_DOWNLOADS]: The documentation in references/patterns.md suggests the installation of the colabfold package using pip and refers to the external alphafill.eu service for structural data enhancement.
[PROMPT_INJECTION]: The skill exhibits an attack surface for Indirect Prompt Injection because it ingests and processes untrusted data from external biological files.
Ingestion points: Protein sequence data (FASTA) in predict_structure_colabfold, structural data (PDB) in calculate_rmsd, run_dssp, and calculate_sasa, and confidence data (JSON) in load_pae_from_json.
Boundary markers: None; external file content is read and processed directly without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill can perform file system operations (reading and writing files) and execute specific external binaries via subprocess.run.
Sanitization: There is no evidence of sanitization or validation for the content of the processed files or the resulting command-line arguments beyond basic string conversion.

protein-structure