How does Ralph handle large projects?

Ralph uses PRD mode to parse large prompts into structured user stories, creating a clear roadmap in prd.json before executing tasks in priority order.

What is Ralph's 'Iron Law' of verification?

The Iron Law is a mandatory protocol that prevents Ralph from claiming completion until fresh evidence—such as test logs, build outputs, and linting results—proves the task is 100% finished.

Can Ralph run tasks in parallel?

Yes, through Ultrawork mode, Ralph can fire 10+ concurrent background tasks for long operations like package installations and test suites.

What models does Ralph use for execution?

Ralph utilizes smart routing: Haiku for simple lookups, Sonnet for standard implementation, and Opus for complex analysis and final Architect verification.

Ralph: Autonomous Task Orchestration

Name: Ralph: Autonomous Task Orchestration
Author: Yeachan-Heo

byYeachan-Heo

•

2,142

•

生产力与工作流

Executes complex tasks through a self-referential loop, leveraging multi-agent orchestration and mandatory architect verification to ensure 100% completion.

Ralph is a high-performance execution engine for Claude Code designed for end-to-end task autonomy. It combines 'Ultrawork' parallel processing with a rigorous 'Iron Law' of verification, ensuring that every code change is validated through builds, tests, and a final review by a dedicated Architect sub-agent. Whether initializing a project from a PRD or fixing deep-seated bugs, Ralph maintains a self-referential loop that persists until the task is demonstrably complete, utilizing smart model routing to optimize for both speed and intelligence.

主要功能

012,142 GitHub stars

02PRD-to-Implementation: Automatically generates product requirements and user stories from initial prompts.

03Ultrawork Parallelism: Fires multiple independent sub-tasks simultaneously to minimize execution time.

04Smart Model Routing: Dynamically assigns tasks to Haiku, Sonnet, or Opus models based on complexity.

05Iron Law Verification: Enforces mandatory test runs and build checks before allowing task completion.

06Architect-Level Sign-off: Requires a high-tier agent to verify all implementations before final delivery.

使用场景

01Autonomous bug hunting and fixing with guaranteed verification cycles and zero tolerance for regressions.

02Building full-stack features from a single natural language requirement with automated task breakdown.

03Complex codebase refactoring requiring parallel updates and verification across multiple modules.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yeachan-heo/oh-my-claudecode ralph

For use in Claude.ai and ChatGPT

Download Skill