ralph-starter and Ralph-Lisa-Loop
These are ecosystem siblings: one provides AI-driven project bootstrapping from external specs, while the other implements a dual-agent code review loop that could validate and refine the bootstrapped code through iterative consensus, addressing different stages of an AI-assisted development workflow.
About ralph-starter
multivmlabs/ralph-starter
Bootstrap projects from Figma, Linear, Notion & GitHub specs. The only AI ralph with source integrations.
Executes autonomous AI coding loops by fetching specifications from external sources (GitHub issues, Linear tickets, Notion docs, Figma designs) and iteratively refining code through error feedback until completion. Includes production-grade safety mechanisms like circuit breakers for stuck loops, rate limiting, cost tracking, and validation backpressure that runs tests after each iteration. Supports multiple coding agents (Claude Code, Cursor, Copilot, Gemini) with git automation for commits, pushes, and PR creation.
About Ralph-Lisa-Loop
YW1975/Ralph-Lisa-Loop
Dual-agent AI code review loop — consensus at every step, not after the fact. Used to evolve an open-source fork into a production AI assistant with zero manual code.
Implements a structured loop where Claude and OpenAI's Codex alternate between code generation and adversarial review, with human arbitration for architectural decisions. Each agent's characteristic failure modes (Claude skipping error handling at scale; Codex over-engineering) are caught by the other's strengths. Requires Node.js 18+, Claude Code API access, and Codex CLI, with optional tmux integration for fully automated workflows.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work