Rubrics for evaluating AI agents
A repeatable, rubric-based way to evaluate how AI agents behave when they act on real systems through MCP tools — consistent scores instead of one-off manual checks.
Projects
Things I have built, tested or written about. More coming soon.
Reproducible dev environments on Aurora OS
Using Distrobox containers on an immutable Linux desktop to keep every toolchain isolated, disposable and reproducible.
narmaku.com
This site: a Jekyll landing page plus a Chirpy blog, built in one pipeline and served from Cloudflare's edge, mirrored on GitHub Pages.