Log inGet started
Package

context.tool

Newest version

This package holds more than this page lists. See all of its files in its world.

README.md

context

The context a harness that composes its own prompt sends beside each request, in one read: where you stand (engine: the world, the mode, the active layer and the scenes), the world's skills (skills, the open ones marked), the caps in force on generations (generation), the brief of the world's memories (memory: what the world's writers and their agents keep for each other, one line per entry, read and kept with the memory toolbox), and the sections the harness states itself (credits after the skills, agents last, any other between them). A harness that carries one of the engine's sections in its conversation itself names it in kept, and the answer leaves it out: DotBot keeps the memory brief as a block of its conversation, read through memory.brief and written again only when it changed, and asks with kept = { "memory" }. A use_tool call of it that names no client is answered by your own sub-client while that one takes calls, so it describes the copy of the world you are working in, and by the user's client otherwise (you hold none yet, yours is starting, or yours was stopped); it never starts a sub-client or waits for one.

For example, DotBot reads it before every request with its credits and agents lines as owned, and sends the text it answers as the tail of the request.