Skip to content
An agent guts a failing test to make the suite green; Cadence's settle gate still refuses, then accepts once the bug is actually fixed

CADENCE

The DRAFT → BUILD → SETTLE loop, with quality gates that refuse to settle broken work.

CADENCE scaffolds and runs the DRAFT → BUILD → SETTLE loop on your project — and uses that same loop on itself. Quality gates are configurable per profile and tier, and the always-fire build/test gate will refuse to close a phase whose tests do not pass, even when the AI marked the task done.

“Your CI is green. Cadence still said no.” An agent hits a failing test, guts the assertion, and leaves a plausible excuse. The suite goes green, tasks get marked DONE. Then cadence settle run --auto refuses anyway — naming the specific AC and the specific dodge. Real output, offline, mock verifier, zero npm deps: full 4-beat narrative.

Also see: the $100/3 demo (arithmetic bugs, not agent cheating) — a lost-penny bill-split bug the build-test-must-pass gate blocks even though the AI marked the task DONE.

Understand the loop

Concepts — profiles × tiers and the full gate matrix.

API

Generated API docs for the published packages.