Notes, soon

Placeholder post. I plan to write here about agent security evaluation — what the benchmarks we build actually measure, where tool-use guardrails give false assurance, and the parts of this work that do not fit in a paper.

First real post coming soon.