The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill utilizes the unsafe pickle module to cache processed database results in references/data-access.md.
Evidence: The code snippet uses pickle.load(f) to deserialize data from a local file named drugbank_parsed.pkl.
Risk: The pickle module is vulnerable to arbitrary code execution if an attacker manages to replace the local cache file with a maliciously crafted one.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large quantities of drug information from external sources, creating a surface for indirect prompt injection.
Ingestion points: The download_drugbank function in references/data-access.md fetches XML data from go.drugbank.com.
Boundary markers: There are no specific instructions or delimiters provided to the agent to treat the XML content as untrusted data or to ignore embedded instructions.
Capability inventory: The skill possesses capabilities to write files (export_chemical_properties in references/chemical-analysis.md) and perform network operations (requests.get in references/data-access.md).
Sanitization: No sanitization, escaping, or schema validation is implemented for the ingested XML data.
[EXTERNAL_DOWNLOADS]: The skill downloads large data files from a remote repository.
Evidence: The download_drugbank() utility in references/data-access.md initiates downloads of the DrugBank database ZIP file.
Note: The source domain go.drugbank.com is a well-known service associated with the skill's primary function.
[METADATA_POISONING]: The skill's internal metadata identifies the author as "K-Dense Inc.", which differs from the provided author context of "leonchaox".
Evidence: The YAML frontmatter in SKILL.md contains skill-author: K-Dense Inc.

drugbank-database