Been exploring loop engineering. Addy Osmani has a good post on it: https://addyosmani.com/blog/loop-engineering/

The hard part of loop engineering IMO is the machinery around it, and omnigent sits above pi, claude, codex, etc. and wraps each in a uniform api. The things it adds are exactly that machinery.

  - Parallel git worktrees so concurrent agents don't step on each other.
  - Approval and cost policies enforced at the harness layer rather than living in a prompt the agent can talk its way around.
  - A maker/checker split where the reviewer can run on a different vendor than the writer, which is a more honest check than a second pass from the same model with the same blind spots.

I can see this being useful if only for the fact I can search all my conversations with my 50 different agents on different providers easier. I spend so much time looking through my cc/codex session or the desktop app or Hermes agent or antigravity session (back when I used that). Hopefully having a layer above will abstract that away.

If I understand it correctly the meta harness will help me scan across all sessions and not have to drop into individual ones.