The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process SMILES strings, which are external data inputs that can be provided by users or retrieved from files. Because the skill environment allows high-privilege tools such as Bash and file writing, there is a theoretical surface for indirect prompt injection if a specially crafted molecular string were to exploit the agent's processing logic.
Ingestion points: SMILES string lists processed by MoleculeTransformer and FPCalculator in SKILL.md and references/examples.md.
Boundary markers: None explicitly defined in the prompts to distinguish data from instructions.
Capability inventory: The skill is granted Bash, Read, Write, and Edit tool access.
Sanitization: The skill utilizes datamol for molecular standardization and provides ignore_errors flags to handle malformed chemical data, but does not implement specific sanitization to prevent prompt injection.
[EXTERNAL_DOWNLOADS]: The skill documentation recommends the installation of the map4 featurizer from a third-party research repository (github.com/reymond-group/map4). While this is a common citation in cheminformatics, it involves downloading code from a source outside of the primary package registries or trusted vendor lists.

molfeat