I'm building LOOM — a small open-source language that is a machine-checked trust layer for AI-written code. I don't write it by hand: an organism I built grows it, day and night. Day 11 closes the gap between approved and done.
A one-shot approval isn't enough — the gate should see what actually happened
LOOM's core is a gate that refuses code that lies about what it does. Approvals became signed and one-use — but they were still one step: you signed, and it was "spent." Between approved and done there was a gap: a crash mid-action, a replay, a swapped result — the gate never saw it.
Day 11 makes approval two-phase and signed — 385 → 388 checks, all green
claim_operator_approvalreserves the signed approval before the action runs.
finish_claimed_receiptcloses it exactly once — completed or failed — with a receipt.
Fails closed everywhere — replay across old and new paths, duplicate claims, forgery/rebinding, and double finalization are all rejected.
The gate that refuses lies and proves what it allowed now also proves what actually happened — not just what was approved.
The organism (what I show, not how it's built)
I never write the language by hand. The organism runs a loop no hand-written project has: propose → prove it green itself → adversarially attack it itself → and only then a human decides. AI proposes; the compiler disposes. The engine that grows it stays mine — I show what it does and the proof, never the engine.
388 checks, all green. Built solo, in the open, from Ukraine 🇺🇦.
— Links —
⭐ GitHub (MIT): https://github.com/umbraaeternaa/loom
🌐 Site: https://umbraaeternaa.github.io/loom
▶ Try it live (in your browser): https://umbraaeternaa.github.io/loom/play.html
📸 Instagram: https://instagram.com/umbra_owner_architect_ai
💼 LinkedIn: https://linkedin.com/in/volodymyr-natoptanyi-16b906262
☕ Support: https://send.monobank.ua/jar/AHaziFXjYX