Ouroboros
01

1. What is Ouroboros

Ouroboros is an open-source tool that manages the requirements, execution, evaluation, and records of AI coding work.

Ouroboros 0.50.4upstream 456a134verified 2026-07-16 against docs and code

1.1 Definition

Ouroboros does not write code itself. Code is written by an AI coding tool such as Claude Code or Codex. Ouroboros works before and after that tool: it pins down requirements, hands off execution, verifies results, and records the process.

Diagram: an idea passes through Ouroboros, which produces a Seed and records, while the runtime produces code and tests
The input is an idea. Ouroboros manages the work. The runtime writes the code. The output is code plus records.

1.2 What you put in

You describe the feature you want or the problem you want fixed, in plain sentences. You do not need a polished specification. Missing decisions are collected by Ouroboros through questions.

InputExample
A one-line idea"A web app that adds to-dos and marks them done"
Answers to questions"The list should survive a page refresh"
An existing repository (optional)A project folder you already have

1.3 Division of roles

Ouroboros and the AI coding tool do different jobs. Ouroboros is the management layer; the AI coding tool is the execution layer. Ouroboros calls the execution layer the runtime.

ActorResponsibility
YouProvide the idea, answer questions, approve the Seed, check the result
OuroborosAsk questions (Interview), write the work spec (Seed), manage execution, evaluate, record
runtime (Claude Code, etc.)Write files, edit code, run commands, run tests

The Seed is the work-spec document that connects the two layers. Goal, constraints, and completion criteria are saved as a YAML file, and the runtime works against that file. The Seed is covered in chapter 6.

1.4 What remains after a run

More than code remains. The following five things are kept.

ArtifactContentLocation
CodeFiles and tests written by the runtimeYour working folder
SeedThe work spec used for this run~/.ouroboros/seeds/
Test resultsWhether lint, build, and tests passedEvaluation output
Evaluation resultsVerdict on the completion criteriaEvaluation output
Execution recordsEvery execution event~/.ouroboros/ouroboros.db

Because execution records persist, you can find interrupted work again and check later what was run and why.

1.5 Before and after

With an AI coding tool alone, the request, the code, and the verification all flow by inside a chat. With Ouroboros, requirements live in a Seed file, the process lives in records, and completion lives in an evaluation verdict. Why this difference matters is explained in chapter 2.