Software engineer. .NET, distributed systems, building stuff from scratch.
I build production software with AI agents and ship open source tools from what I learn. ResearcherSkill and Yggdrasil came from watching agents say "done" when they're not even close.
What I'm working on:
Yggdrasil. Architecture enforcement for AI coding agents. You write rules in plain Markdown, a reviewer verifies every change the agent makes and feeds violations back into the agent's loop before it moves on. Works with Claude Code, Cursor, Copilot, Codex, Cline.
AutoReview. Lighter sibling of Yggdrasil. Per-file Markdown rules with a trigger DSL, verified on every commit. Claude Code plugin, zero npm deps, runs on local Ollama by default.
ResearcherSkill. Autonomous deep research skill for Claude Code.
Bigger picture: infrastructure for software that builds itself. Mechanisms that enforce correctness, because better prompts don't hold and agents won't do it on their own.
