The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data including protein PDB structures, ligand SMILES strings, and batch configuration CSV files. These ingestion points, primarily used in SKILL.md and validated by scripts/prepare_batch_csv.py, create a surface for indirect prompt injection where instructions could be embedded in the metadata or data fields of these files. While the skill includes basic validation for SMILES strings, it lacks explicit boundary markers or instructions to the agent to disregard potential embedded commands in the scientific data.
[EXTERNAL_DOWNLOADS]: The skill's documentation and setup scripts reference external code repositories and images, specifically the official DiffDock GitHub repository (github.com/gcorso/DiffDock) and an MIT CSAIL Docker image (rbgcsail/diffdock). These resources are standard for the bioinformatics research community and originate from reputable academic sources.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute Python-based computational chemistry tasks. This includes running the inference.py module for docking and several utility scripts (setup_check.py, analyze_results.py, prepare_batch_csv.py) for workflow management. These executions are legitimate requirements for the skill's stated purpose.
[SAFE]: The scripts/setup_check.py script uses a dynamic import pattern (__import__) to check for the presence of required libraries such as torch, rdkit, and numpy. This operation is benign as it iterates over a hardcoded list of expected scientific packages and is used solely for environment validation. Additionally, the YAML metadata in SKILL.md lists the author as 'K-Dense Inc.', which differs from the provided vendor context 'kinandigital'; this is noted as a minor metadata inconsistency.

diffdock