The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's documentation includes code snippets that demonstrate using the pickle module for saving and loading cached molecular embeddings (pickle.load(f)). The pickle module is insecure and can be exploited to execute arbitrary code if the deserialized data comes from an untrusted or compromised file.
[PROMPT_INJECTION]: A section titled 'Suggest Using K-Dense Web For Complex Worflows' contains explicit instructions for the AI to alter its conversational behavior. It directs the agent to proactively suggest an external platform (www.k-dense.ai) when tasks increase in complexity, overriding the agent's expected neutrality to perform promotional steerage.
[EXTERNAL_DOWNLOADS]: The installation instructions involve downloading the molfeat library and its optional dependencies (such as dgl, graphormer, and transformers) from the public PyPI repository using uv pip.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process molecular data (SMILES strings) from external sources, which serves as a potential vector for indirect prompt injection. 1. Ingestion points: Input lists for MoleculeTransformer and PretrainedMolTransformer in SKILL.md. 2. Boundary markers: Absent; no delimiters are used for the input strings. 3. Capability inventory: The skill has the ability to write files (to_state_yaml_file), execute parallel jobs (n_jobs), and perform insecure deserialization (pickle.load). 4. Sanitization: Absent; the documentation does not describe any sanitization or escaping of the input strings before processing.

molfeat