From Prompt to Agentic AI: The PASTAS Checklist for Context Engineering
Leo S. Lo · portal Libraries and the Academy · 2026
abstract: This article introduces PASTAS - Purpose and Audience, Authority, Structure and Style, Tools and Access, Accountability and Safeguards, Signals and Review - as a structured approach to context engineering for agentic artificial intelligence (AI). Moving beyond one-shot prompts, PASTAS offers a repeatable blueprint for designing multi-step, library-aligned AI workflows that integrate trusted sources, enforce ethical safeguards, and produce decision-ready outputs. Using a policy research support agent as an example, the author illustrates how each component of the model guides tool selection, workflow logic, and review cycles. The framework embeds librarian values in agent design, ensuring accuracy, transparency, and human oversight remain central in AI-enabled services.