@loopyc/verify

Verify-before-you-run for Monkey D Loopy loops.

A codegen-free interpreter dry-runs a LoopSpec through the real @loopyc/runtime with mocked effects — no network, no shell, no model calls — and proves it is bounded, deterministic (two fresh runs converge), and resume-stable (a process restart between every iteration changes nothing). Then a 0–100 scorecard grades termination safety (including grounding — whether real evidence or the agent's own report decides when the loop stops), caps, observability, resumability, and determinism.

import { verifyLoop, scoreLoop, type VerifyOptions } from "@loopyc/verify";

const options: VerifyOptions = {
  fixtures: { shell: { done: true }, http: { status: "green" } },
};
const report = await verifyLoop(spec, capsInjected, options);
const card = scoreLoop(spec, report); // { total: 100, grade: "A", dimensions: [...] }

Fixtures are cloned data returned by the dry-run mocks. They make realistic exit paths reproducible without enabling network, shell, or model calls. Observer points require executable evidence—an observe.hooks.completed action or an active top-level notification—not inert metadata.

Most users want the CLI instead: npm i -g @loopyc/cli → loopc verify spec.yaml --fixtures fixtures.json · loopc score spec.yaml --fixtures fixtures.json.