ouroboros

Ouroboros calls itself an Agent OS for replayable, specification-first AI coding workflows: a local-first runtime layer that turns non-deterministic agent work into a replayable, observable, policy-bound execution contract. Its motto is to stop prompting and start specifying. The README argues that most AI coding fails at the input, because vague prompts make the AI guess, a missing spec lets the architecture drift, and manual QA is not verification.

The answer is a loop of interview, crystallise, execute, evaluate and evolve. A Socratic interview exposes hidden assumptions, an immutable seed spec locks intent before code is written, and a three-stage automated evaluation gate checks the result; the CLI shows an ambiguity score against a threshold and asks whether to keep interviewing. It works across many agent runtimes, including Claude Code, Codex CLI, OpenCode, Gemini, Kiro and Copilot. Install with the one-line script, open your agent and type `ooo`, then for example `ooo interview "I want to build a task management CLI"`. The README describes a three-repository stack (a terminal shell, application plugins and the core), and the project is MIT licensed.