The Agent Skills Directory

[SAFE]: No malicious patterns or behaviors were identified. The skill is entirely focused on its stated purpose of drug profiling and pharmaceutical research.
[EXTERNAL_DOWNLOADS]: The skill fetches data and images from well-known scientific and governmental services. This includes retrieval from the National Center for Biotechnology Information (NCBI) PubChem database, the EMBL-EBI ChEMBL database, and the FDA's DailyMed service. These sources are considered trusted within the scientific domain.
[PROMPT_INJECTION]: The workflow instructions (e.g., 'Report-First Approach', 'MANDATORY completeness') are legitimate operational constraints designed to ensure the quality and structure of the research report. There are no attempts to override safety filters or hijack the agent's behavior.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from sources like PubMed abstracts and ClinicalTrials.gov records.
Ingestion points: External data enters the context through tool outputs from the literature search (PATH 8) and comparative analysis (PATH 9).
Boundary markers: While explicit delimiters for tool outputs are not defined, the skill uses a highly structured markdown template with predefined section headers to contain the data.
Capability inventory: The skill possesses the capability to create and update local files (e.g., [DRUG]_drug_report.md). It does not have general subprocess or non-whitelisted network capabilities.
Sanitization: No explicit sanitization of external content is mentioned, but the risk is mitigated by the structured extraction process which focuses on data rather than instruction following.

tooluniverse-drug-research