wtf is finn-loop?

Finn-loop is a small set of tools that lets Claude Code, an AI coding assistant, turn ideas into working software through a structured, human-controlled process. It connects two services: Linear, where you track tasks, and GitHub, where code lives. The core idea is that AI handles the repetitive work of writing specifications, building features, and reviewing code, while a person makes the final call on whether to accept each change. The system uses three commands. The first, finn-spec, interviews you about an idea until the requirements are clear, then creates a Linear issue with specific acceptance criteria and stated non-goals. The second, finn-build, picks up an issue you have marked as ready, writes the code to satisfy exactly that issue, and opens a pull request. The third, finn-review, examines open pull requests against their linked issues and required checks, then posts a verdict grouped into clear categories. You merge only what passes. Installation involves copying three skill files into your repository and running a setup prompt inside Claude Code. The setup checks your software versions, connects Linear and GitHub, creates needed labels like agent-ready and loop-approved, and validates the configuration. The README recommends connecting Linear's GitHub integration so that merged pull requests automatically update the status of related issues. The daily workflow takes about fifteen minutes. You run the spec command when you have an idea, review the resulting issue, and apply the agent-ready label if you approve. Then you start a build loop and optionally a review loop. You merge only pull requests that are marked loop-approved, free of conflicts, and passing all checks. If something is blocked, you answer questions and remove the blocked label to resume work. A strict rule underpins the whole system: humans merge. The AI never merges code or enables auto-merge. The loop-approved label is evidence for your decision, not permission for the AI to act. The README also outlines optional patterns for scaling up, such as self-converging pull requests and Slack notifications, but these are suggestions for future steps, not features included here. The full README is longer than what was shown.