The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill retrieves data from a variety of authoritative scientific sources, including PubChem, ChEMBL, the FDA, and ClinicalTrials.gov. These network operations are strictly limited to data fetching from well-known public services and are fundamental to the skill's purpose.
[COMMAND_EXECUTION]: Research is conducted using a predefined suite of tools. The skill does not contain instructions or scripts for arbitrary command execution, privilege escalation, or the installation of external software.
[PROMPT_INJECTION]: The workflow instructions are focused on research methodology and report structure. There are no attempts to subvert agent behavior or bypass safety protocols. An assessment of the indirect prompt injection surface found a safe risk profile:
Ingestion points: Data is ingested from PubMed, clinical trial registries, and pharmacological databases.
Boundary markers: Ingested content is not explicitly delimited, but the agent is directed to process it into a structured markdown report.
Capability inventory: The skill allows for network retrieval and local file creation (markdown reports); no dangerous system-level capabilities are exposed.
Sanitization: The skill relies on the agent's internal synthesis capabilities without additional manual sanitization steps, which is appropriate given the reputable nature of the data sources.
[DATA_EXFILTRATION]: No patterns of credential harvesting or unauthorized data exfiltration were found. The skill only transmits chemical identifiers (like SMILES strings or compound names) to scientific databases to retrieve relevant research data.

tooluniverse-drug-research