MartinLoop - control plane for autonomous AI agents

[Keesan12]

We've created an open source governed runtime solution to ralph wiggum loops

This has saved us a ton of $$$$ in dead end agentic loops in overnight runs, avg is 55-65% cost reductions

Check the repo and test it yourself, you can see the cli and run test comparisons

It's an open source core oss release

https://github.com/Keesan12/martin-loop/

Never deal with stupid ralph loops again

Who else has been burned by ralph loops? Tue overspend is insane

[Keesan12]

Small follow-up from the MartinLoop side: we just pushed the first concrete docs/examples that make the governance layer easier to evaluate without hand-waving.

• Claude Code walkthrough
• OpenCode-style adapter example
• Ralph-style loop safety guide
• GitHub Actions budget gate example

The through-line is the same: keep the coding agent useful, but put a reviewable contract around it with budget caps, verifier gates, explicit stop reasons, and JSONL run records.

The main thing I’m still trying to learn from other teams is which failure hurts most in practice:

• budget overspend
• verifier churn
• scope drift
• weak auditability
• secret or approval-boundary exposure

If anyone here is already running long unattended coding loops, I’d be especially interested in what your current stop condition and post-run evidence trail look like.