AI-Builder-Club/skills

Codebase harness + loop engineer

What it solves

It addresses the limitation of task-by-task prompting by enabling the creation of "loop engineers"—autonomous agents that trigger themselves, perform work, verify results, and log learnings into a shared memory to create compounding productivity.

How it works

The project provides a set of plugins for Claude Code that implement two primary frameworks:

  • Codebase Harness: Prepares repositories for agentic development by setting up local or cloud-based isolated environments, end-to-end testing, and verification workflows.
  • Loops: Establishes a shared, file-based knowledge base (using ARCHITECTURE.md, LOG.md, and specialized folders) that allows multiple independent agent loops to read and write to the same memory.

It also includes utilities for generating visual flow diagrams from JSON and delegating tasks to various CLI-based agents within detached tmux sessions.

Who it’s for

Developers and AI engineers looking to move from manual prompting to designing autonomous, compounding agent workflows and managing agentic development environments.

Highlights

  • Codebase harness for making repositories agent-ready with automated testing and verification.
  • Compounding loops that use a shared file-based memory to allow different agents to build upon each other's work.
  • Isolated cloud environments via crabbox-setup to allow multiple agents to ship code in parallel.
  • Agent delegation to coordinate multiple CLI agents (like Claude, Grok, or Aider) using a structured protocol.