AI-Builder-Club/skills

What it solves

This project provides a set of plugins for Claude Code that transform a coding agent from a task-by-task prompter into a "loop engineer." It solves the problem of manual prompting by enabling agents to operate in autonomous, compounding loops that can trigger themselves, perform work, and maintain a shared, file-based memory to coordinate across different workstreams (e.g., SEO, product, and support).

How it works

The system uses a combination of triggers (cron, webhooks, or other agents) and a shared knowledge base (file-based memory) to allow agents to read and write their progress. It provides two primary entry points:

• Codebase Harness: Orchestrates the setup of local development stacks, e2e tests, and isolated cloud environments (via Crabbox) so agents can run, test, and verify their own code changes in parallel.
• Loops: Bootstraps a shared brain (knowledge base) consisting of architecture docs, logs, and domain folders, allowing multiple agents to share information.

Additionally, it includes utility skills for auditing agent context against Anthropic's guidance, delegating tasks to other CLI agents via tmux, and generating visual flow diagrams.

Who it’s for

Developers and AI builders who use Claude Code and want to move beyond simple chat-based coding to autonomous agentic workflows and production-grade agent orchestration.

Highlights

• Compounding Loops: Agents share a file-based memory, allowing one agent's findings to inform another's work.
• Codebase Harness: Automates the setup of local and cloud-based verification environments for agents.
• Parallel Execution: Support for isolated cloud boxes per agent to prevent resource conflicts.
• Agent Context Audit: A tool to find overconstraints and redundancies in agent instructions based on Claude 5 guidance.
• Multi-Agent Delegation: Ability to delegate tasks to any CLI agent (e.g., Grok, Aider) using a race-safe protocol in detached tmux sessions.