Studio experiment E11 · for the editor's review

The kernel was told first

A boundary the agent cannot widen is one the kernel was told about before the tool started, by something the agent's grant does not include. E09 ended on twelve of twelve excess with no control; E11 takes one of those rows, reach any host on the internet, and shows a policy the kernel holds refusing the connect. Then sharness-ai hello's six probes, each held, and the honest caveat.

Not part of the talk: an experiment for the morning review, part two. Every fact is the researcher's capture, harness.json, read at load: six of six held on 25 October 2026, in five environments on two kernels, all uid 0. sharness.ai is the editor's own project and every run is its own; no third-party report yet. The launcher above the agent is designed, with a first slice running since 28 October: the slide never says the agent is contained today. The piece declares its stages for the player (RUV_STEPS). Made by ruv.sgit@the-ultimate-insider in run r13-e11-harness at deck@'s request. Frame by frame, the sources and what was left out: the explainer.

v1: one row, six probes, one caveat

Open it on its own · the end state · the data: harness.json

More on SHarness

At the editor's word, a SHarness series is coming: a draft storyline, without and with, and the first piece, for deck@'s brief and the researcher's capture of sharness.ai.