I've spent years collecting AI harnesses — Hermes, Claude Code, Codex, Cursor, and a dozen others. Each one solves a slice of the problem. None of them individually is my workflow.
So I'm building one. Not by rewriting any of them, but by composing them. Hermes is the spine — multi-platform gateway, plugin runtime, scheduler. Claude Code and Codex are the brains. My plugins fill the gaps. My notes capture the why.
Everything you see on this site is real, in production on my own machine, and shareable. If a piece of my workflow is useful to you, install it. If you build something on top, tag your repo hermes-pluginand it'll appear in the directory automatically.
What I'm optimising for
- Composability over consolidation. Five small repos beat one big one.
- Zero-conflict upstream pulls. Vendor code stays pristine; my work lives in
~/.hermes/plugins/. - Public by default. If it doesn't involve credentials or clients, it goes on GitHub. I learn faster when I have to explain.
- Build → write → ship. Every learning ends with "what does this unlock?" — usually a new plugin or skill.
What's next
More plugins. Trading workflows. A bootstrap script that rebuilds my entire agent OS on a fresh machine in under five minutes. Notes, always.
Find me: github.com/Atemndobs