The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructions direct the agent to write and execute Python code for statistical analysis, including calculating Proportional Reporting Ratios (PRR) and Reporting Odds Ratios (ROR). This dynamic execution is a functional requirement for the skill's pharmacovigilance purpose. The instructions provide specific code templates in PHASE_DETAILS.md and rely on established libraries like pandas and scipy, which minimizes the risk of arbitrary code execution.
[COMMAND_EXECUTION]: The skill involves running shell commands via Bash to execute Python analysis scripts and manage data outputs. These commands are localized to the agent's computation environment and are used to process data retrieved from trusted medical databases.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external scientific sources, such as PubMed abstracts and clinical trial records. Ingestion points include PubMed_search_articles and EuropePMC_search_articles. While explicit boundary markers are limited to the structured report template, the skill mitigates risk by requiring the agent to map data into a rigid evidence grading system (T1-T4) rather than executing instructions directly from the fetched text. The capability inventory for file-write and network operations is constrained to the primary purpose of report generation and database queries.

tooluniverse-adverse-event-detection