The Agent Skills Directory

Dynamic Code Generation and Execution: The skill's workflow involves an AI agent generating Python code for a Monte Carlo simulation (simulate.py). This code is executed to produce the forecast numbers. To manage risk, the skill implements a multi-agent review process where an independent 'critic' agent is required to audit the code for statistical correctness and bugs before results are finalized.
Workflow Orchestration via Command Execution: The skill uses shell commands and helper scripts to manage its multi-phase execution. These commands orchestrate sub-agent delegation, process registry records, and handle background simulation jobs. The orchestration logic includes validation steps to ensure that file operations remain within the designated project folder.
Indirect Prompt Injection Surface: The skill processes third-party data from clinical trial registries and research documents (such as PDFs). To mitigate potential indirect prompt injection, prompts for the researcher and fact-checker agents include explicit instructions to treat all external content as data and to ignore any instructions embedded within source materials.
Multi-Agent Task Separation: The skill utilizes delegation to partition tasks among different agents (lead, builder, critic, researcher, checker). This architecture ensures that sensitive tasks, such as generating simulation parameters and verifying headline claims, are performed by separate entities, providing a layer of independent verification throughout the forecasting process.

clinical-trial-forecast