The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation references the installation of the map4 featurizer from a third-party GitHub repository (github.com/reymond-group/map4) that is not part of the standard package registry or the trusted vendor list.
[COMMAND_EXECUTION]: The skill utilizes parallel processing via the n_jobs=-1 parameter in the MoleculeTransformer class, which involves spawning multiple subprocesses to perform molecular feature calculations across CPU cores.
[DYNAMIC_EXECUTION]: The skill provides functions to save and load featurizer states (from_state_yaml_file and from_state_json_file). Loading these states involves deserializing data from local files, which could pose a risk if the configuration files are malicious and processed using unsafe deserialization methods.
[INDIRECT_PROMPT_INJECTION]: The skill processes external SMILES strings and chemical data, which creates a surface for indirect prompt injection if malicious instructions are embedded in the molecular datasets or error logs.
Ingestion points: External SMILES strings and datasets are ingested via the MoleculeTransformer and FPCalculator classes (e.g., transformer(smiles_list) in SKILL.md).
Boundary markers: The skill documentation does not provide specific instructions for using delimiters or warning messages to isolate external molecular data from the agent's control flow.
Capability inventory: The skill has access to the Bash tool and can spawn multiple subprocesses via parallelization (n_jobs). It also includes file-writing capabilities through state persistence methods (to_state_yaml_file).
Sanitization: There is no evidence in the provided documentation of sanitizing molecular data or error messages before they are returned to the agent context.

molfeat