I want every paragraph to have a unique ID and support composing pages from those paragraph blocks.
The workflow should have two stages:
- Prepare and engineer the context by selecting the relevant paragraph IDs for the task.
- Use the engineered context to execute the task with only the relevant information.
The goal is for agents to process the content, identify relevant paragraph IDs, and generate task-specific disposable pages that exclude irrelevant paragraphs in a token-efficient way (by only producing IDs).
Before the task is executed, pages containing paragraph ID references should materialize into Markdown text when injected into the LLM context.
I want every paragraph to have a unique ID and support composing pages from those paragraph blocks.
The workflow should have two stages:
The goal is for agents to process the content, identify relevant paragraph IDs, and generate task-specific disposable pages that exclude irrelevant paragraphs in a token-efficient way (by only producing IDs).
Before the task is executed, pages containing paragraph ID references should materialize into Markdown text when injected into the LLM context.