The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/compute_mmpbsa.py executes external scientific tools cpptraj and MMPBSA.py using subprocess.run to perform trajectory stripping and energy calculations.
Evidence: Calls occur at line 218 (cpptraj) and line 281 (MMPBSA.py).
Context: The commands are invoked using argument lists (avoiding shell=True) and are necessary for the skill's primary computational workflow.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted chemical files which presents an indirect injection surface if file metadata contains malicious instructions that are later echoed to the agent.
Ingestion points: External data is ingested in scripts/compute_mmgbsa.py via mda.Universe (line 123) and Molecule.from_file (line 137), and in scripts/compute_mmpbsa.py (lines 76, 122).
Boundary markers: The instructions lack specific delimiters or system warnings to the agent regarding potential instructions embedded in chemical file metadata.
Capability inventory: The skill maintains capabilities to write files to the local filesystem and execute internal binaries (cpptraj, MMPBSA.py, sqm).
Sanitization: The skill utilizes structured scientific parsers (MDAnalysis, OpenFF) but does not apply natural language sanitization to metadata strings (e.g., molecule names) that appear in results and logs.

drug-mmpbsa-gbsa