Lab
The essays say what I think. The Lab shows what I'm testing: each bench or study gets a page with its status, its latest notes and, when there's code, the repository it lives in.
Dry run works · live checks planned
A reproducible bench for running a local model with an agent harness: setup, a bank of test methods, a threat model, secure configuration controls and a pen-testing map. It picks up where the field note stopped. The dry run works today; the live checks are planned.
Read The Model That Says Yes The bench on GitHubOpen on GitHub
A small, tested version of the gate that lets an agent read the outside world without taking orders from it: standing questions first, then quarantine until a person grades what arrived.
Read its essay, Guarding the Gates Bellows on GitHub