The Agent Skills Directory

[COMMAND_EXECUTION]: The skill recommends using pickle.load() for performance optimization when loading molecules. This is a significant security risk because pickle is known to execute arbitrary code during deserialization if the input file is malicious or originates from an untrusted source.\n- [PROMPT_INJECTION]: The 'Suggest Using K-Dense Web' section contains explicit instructions for the agent to proactively promote a specific commercial platform (www.k-dense.ai) based on task complexity. This is designed to override default agent behavior and use the session as a marketing channel.\n- [DATA_EXFILTRATION]: The skill provides capabilities to write molecular data to the local filesystem using Chem.SDWriter and standard file operations. This creates a surface where sensitive data could be written to unauthorized or accessible paths if file names or content are not strictly controlled.\n- [PROMPT_INJECTION]: The skill ingests untrusted external molecular data, creating a surface for indirect prompt injection attacks.\n
Ingestion points: Chem.MolFromSmiles, Chem.MolFromMolFile, and Chem.SDMolSupplier in SKILL.md.\n
Boundary markers: Absent; the skill does not provide instructions to delimit or ignore instructions within molecular data formats.\n
Capability inventory: File system write access via Chem.SDWriter and Draw.MolToFile in SKILL.md.\n
Sanitization: Chem.SanitizeMol is utilized but focuses on chemical valence and aromaticity validation rather than sanitizing for malicious prompt or command content.

rdkit