The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation instructions in SKILL.md and references/workflows_examples.md direct the user to clone the DiffDock repository from github.com/gcorso/DiffDock.git and pull a Docker image from rbgcsail/diffdock. These sources are outside the provided trusted vendor list and represent external code dependencies.
[DYNAMIC_EXECUTION]: The scripts/setup_check.py utility uses the __import__ function to dynamically check for the presence of required Python libraries. Although the library names are sourced from a static internal list, this is a form of dynamic module loading at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted protein structure data (PDB) and ligand descriptions (SMILES) which are interpolated into Bash commands for docking inference. This creates a potential attack surface for command injection if the input data contains shell metacharacters.
Ingestion points: The protein_path, ligand_description, and protein_sequence fields in SKILL.md workflows and the CSV files processed by scripts/prepare_batch_csv.py.
Boundary markers: None identified in the prompt instructions to delineate untrusted data.
Capability inventory: The skill uses the Bash tool to execute python -m inference and several local helper scripts.
Sanitization: scripts/prepare_batch_csv.py performs basic SMILES validation using RDKit and checks for the existence of local protein files, but does not perform shell-safe escaping.
[COMMAND_EXECUTION]: The skill relies on the Bash tool to perform environment checks, run molecular docking models, and execute results analysis scripts.

diffdock