Show and tell: MartinLoop — governed runtime for AI coding agents

[Keesan12]

AI coding agents get expensive when they fail without a contract.

The narrow problem MartinLoop focuses on is simple: should the next attempt run at all, under what budget, against what verifier, and what record gets left when it stops? That is separate from model choice and separate from outbound tool control. If you are running Claude, Codex, OpenCode, or your own loops, that runtime boundary is usually where blind retries turn into real spend.

Repo: https://github.com/Keesan12/martin-loop
Website: https://martinloop.com