progressive-disclosure

To reduce LLM context window bloat, use progressive disclosure to divide information into three staged layers: metadata, core instructions, and on-demand references. This structured workflow prevents prompt bloat by loading resources only when needed.

Staged prompt loading optimizes LLM interactions by separating metadata, core content, and optional scripts into distinct layers. This progressive disclosure mechanism ensures focused interactions and prevents context windows from becoming overloaded with irrelevant resources.

Structure prompts by defining Stage 1 metadata, Stage 2 core content, and Stage 3 on-demand references. This 3-stage loading model enforces clear boundaries and frontmatter metadata to optimize reuse and collaboration across modular AI workflows.

Yes, progressive disclosure applies to Skills, Agents, and Prompts. It enforces frontmatter metadata and structured content separation across all three, enabling teams to contribute modular assets that are loaded only when needed for scalable workflows.

The best way to organize on-demand scripts and assets is separating them into a dedicated Stage 3 layer. This progressive disclosure approach keeps them out of the initial LLM context window, loading resources dynamically only when the workflow requires them.