<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> Looper is a design-time tool for Claude Code that enables users to visually design, review, and refine agent loops before execution. It addresses the challenge of creating robust, verifiable, and well-defined automation workflows by providing a structured design process, explicit review gates, and a cross-model "council" for objective evaluation. This empowers engineers and researchers to build more reliable agent loops, reducing errors and improving the quality of automated tasks.

How It Works

Looper functions as a design layer, guiding users through defining a sharp goal, necessary context, checkable verification criteria, and feedback mechanisms. Its core innovation lies in the "council" concept, where a separate model acts as a reviewer or judge, providing an objective critique distinct from the model executing the loop. This process ensures goals are coachable, verification is explicit (programmatic, judge, or human), and termination guards are robust. The output is a portable specification (loop.yaml, loop.resolved.json) and a runnable Python script (run-loop.py), facilitating both immediate in-session execution and external orchestration.

Quick Start & Requirements

Installation is performed as a global personal skill and slash command within Claude Code, via provided PowerShell or bash scripts, or manual git clone and file copying. The primary prerequisite is a functional Claude Code environment. No other non-default dependencies are specified.

Highlighted Details

• Design Coaching: Critiques user-defined goals against best practices, ensuring falsifiability and defined "done" states.
• Review Gates: Implements explicit "plan" and "delivery" gates for programmatic checks, model judges, or human sign-offs.
• Cross-Model Council: Recommends using a different model family for review/judging than the host model for objective critique.
• Termination Guards: Enforces explicit stop conditions including max iterations, revision caps, no-progress signals, and budget limits.
• Artifact Generation: Produces loop.yaml, loop.resolved.json, run-loop.py, plan.md, delivery-N.md, state.json, and run-log.md.

Maintenance & Community

The project is maintained by Kevin Simback (@ksimback on GitHub and X). No specific community channels (e.g., Discord, Slack) or sponsorship details are detailed in the provided README.

Licensing & Compatibility

Looper is released under the MIT license, which is permissive and generally compatible with commercial use and closed-source linking without significant restrictions.

Limitations & Caveats

Looper is a design and specification tool; it does not provide durable orchestration. It does not handle scheduling, persistent retries across restarts, sub-agent lifecycle management, concurrency controls, or production run history storage. Users requiring these guarantees must integrate the generated Looper spec with a separate, dedicated orchestration system.