finna/Finn-loop

The Finn-loop: a 3-skill AI software factory for Claude Code — spec, build, review. Humans merge.

What it solves

Finn-loop transforms a standard development workflow using Linear and GitHub into an AI-driven "software factory." It automates the tedious parts of the software lifecycle—specification, implementation, and review—while ensuring that humans remain the final authority on what gets merged into the codebase.

How it works

The system operates as a set of three custom skills for Claude Code that integrate with Linear (for issue tracking) and GitHub (for version control and CI):

  1. Specification (/finn-spec): The AI interviews the user to turn a vague idea into a detailed Linear issue with clear acceptance criteria and non-goals.
  2. Implementation (/finn-build): The AI claims an agent-ready issue from Linear, implements the required changes, verifies them, and opens a GitHub Pull Request.
  3. Review (/finn-review): The AI reviews open PRs against the original Linear contract and GitHub status checks, providing a verdict on whether the PR is ready for merge.

Users can run these processes in loops (using /loop) to continuously process the queue of ready issues and pending reviews.

Who it’s for

Software engineers and product owners who use Claude Code, Linear, and GitHub and want to automate the execution of small, well-defined tasks without giving up control over the final merge decision.

Highlights

  • Human-Gated Workflow: Strictly enforces a rule where only humans can merge code.
  • Contract-Driven Development: Uses Linear issues as the single source of truth; if a requirement isn't in the issue, the agent won't build it.
  • Integrated Tooling: Leverages the GitHub CLI (gh) and Linear connectors for seamless state management.
  • Extensible Architecture: Provides a roadmap for evolving from a simple loop into a full factory with Slack integrations, risk-aware merging, and automated watchdogs.

Related

  • Project
    vercel-labs/eve-software-factory-template

    An AI software factory that automates the development loop by using a pipeline of specialized agents to turn GitHub and Linear issues into reviewed pull requests.

  • Project
    umputun/cc-thingz

    A marketplace of plugins for Claude Code that adds specialized skills for structured planning, autonomous task execution, PR reviews, and release management.

  • Project
    Wirasm/prp

    A framework of prompts and agent skills for Claude Code that transforms product requirements into production-ready code through structured planning, implementation, and autonomous review loops.

  • Project
    feiskyer/claude-code-settings

    A collection of custom skills, sub-agents, and configuration templates to extend Claude Code with capabilities like deep research, image generation, and GitHub automation.