The Agent Skills Directory

[SAFE]: The skill identifies itself as a clinical decision support tool and limits model invocation to ensure deterministic behavior where possible. It encourages the use of a local, offline pharmacology reference script (scripts/pharmacology_ref.py) to minimize external dependencies for primary queries.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a structural ingestion surface where it fetches data from external research databases (PubMed, DrugBank, DailyMed) to populate its generated markdown reports. This is a legitimate and necessary function for the skill's primary role.
Ingestion points: Untrusted text is retrieved from external APIs via PubMed_search_articles, drugbank_get_drug_basic_info, and FAERS_count_reactions in the analysis scripts.
Boundary markers: The skill uses markdown headers and bold labels within its generated reports to distinguish between drug profiles, mechanisms, and clinical evidence.
Capability inventory: The skill has file-writing capabilities through the python scripts (ddi_pipeline.py) which create and append to local markdown report files.
Sanitization: String truncation is applied to the retrieved descriptions to prevent excessively large prompt contents, though active sanitization of markdown syntax within the fetched strings is absent.
[DATA_EXF_EXPOSURE]: While the skill includes an .env.template requesting several API keys, these are standard placeholders for biological and medical research services. The provided scripts do not contain logic to exfiltrate these credentials or send sensitive local files to unauthorized domains.

tooluniverse-drug-drug-interaction