Ec Ralph Loop

What This Skill Does

The Ec Ralph Loop is an autonomous coding agent skill designed to overcome the limitations of LLM context windows during complex development tasks. By leveraging a structured 'user story' approach, the skill breaks large, daunting features into small, atomic units of work. It executes a loop that spawns fresh coding agent instances (utilizing either Codex CLI or Claude Code) for each individual story. As each task is completed, verified via tests, and committed to git, the skill updates the shared prd.json state. This process continues iteratively until the entire project scope defined in the PRD is satisfied. It provides a robust architecture for handling multi-step code generation that would otherwise cause an AI model to lose focus or hallucinate due to token limits.

Installation

To integrate this skill into your OpenClaw environment, use the command line interface to pull the module from the central repository:

clawhub install openclaw/skills/skills/henrino3/ec-ralph-loop

Ensure your workspace has the necessary dependencies for the underlying Codex or Claude Code CLI tools configured, as Ralph acts as an orchestrator for these existing binaries.

Use Cases

This skill is perfectly suited for high-volume or high-complexity technical migrations that require sustained accuracy. Common use cases include:

• Large-scale database migrations (e.g., implementing Row-Level Security across hundreds of tables).
• Systematic refactoring of legacy codebases where manual intervention is tedious.
• Rapid feature prototyping where the workflow involves distinct, repetitive steps like creating a schema, then a service, then a UI component for each database entity.
• Automated test-driven development where individual stories are tied to specific validation suites.

Example Prompts

1. "Ralph, look at the project in ~/apps/crm and implement Row Level Security for all 45 tables listed in the schema, using the existing PRD at tasks/security-audit.md."
2. "Use the Ralph loop to finish the remaining user stories in my prd.json for the OAuth authentication feature, ensuring you use Claude Code for the implementation."
3. "Initialize a new Ralph PRD for the user profile page rewrite, breaking down the requirements into atomic stories for database, API, and frontend tasks."

Tips & Limitations

• Right-Sizing Stories: The success of Ralph depends heavily on how granular your stories are. Avoid ambiguous, large-scale requirements like 'build the login system'; instead, split them into 'create users table', 'implement hash validation', and 'add session persistence'.
• Monitoring: Since this is an autonomous loop, always review the progress.txt file and git logs periodically. The tool is excellent for speed, but deep architectural changes still benefit from human oversight.
• Context Persistence: While the agent resets its memory between stories, the AGENTS.md file serves as a persistent 'lessons learned' document. Use it to keep the agent updated on project-specific coding patterns or global constraints that it should adhere to across all iterations.