Read along

yat:read walks a book, paper or article with you one section at a time. You read and explain. The tutor asks, and cites the page for everything it says about the source.

What you say

You say	What happens
read <source> with me	Starts where your reading position says, one section at a time
I'm at page N	Moves your position
where does it say that	A quote of at most three sentences, with its location
where does <source> talk about X	The locations that match, by page

Why it never summarizes

A summary does your reading for you. So the loop is: predict from the heading, read the section yourself, answer questions that name their page, and explain the section in your own words. Anything the tutor says from outside the source is labelled as such.

Before you start

A source has to be absorbed first with yat:media, which builds its page index. Web pages are absorbed too, never fetched live. Audio and video work once a text transcript exists.

Files it touches

Reads	Writes
media/out/ (the extract and its index)	envs/<env>/interactions.jsonl
envs/<env>/env.toml, interactions.jsonl	Your reading position, in the graph
memory/ontology/graph.jsonl	Notebook captures, when an explanation passes

Terminal: forbidden. Needs python3 for the reading index.

Source

harness/skills/read/SKILL.md