Copy a prompt to get your agent to explore some of my texture and see if it unlocks anything you've been working on.
Agent context should answer what it can and cannot know.
That should be explicit.
It can know the public notes.
It can know the public writing.
It can know the visible source trails.
It can know the stated uncertainty.
It cannot know the whole person.
It cannot know the held context.
It cannot turn a public pattern into a private claim.
The best agent context makes the boundary easier to respect.
sourced Jun 3 2026 · medium confidence · argue it, don't flatten it
Every page here carries a machine-readable layer underneath: the sources, dates, confidence levels, and connections behind what you just read. I call it undertext. If you read with an agent, give it this page's undertext:
https://jared.is/undertext.json?target=/thinking/2026-06-03-agent-context-should-answer-what-it-can-and-cannot-know/