An advanced orchestration framework for agents to execute multi-step planning and building loops with full TTY support for interactive coding tools.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install ralph-loop-agent
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install ralph-loop-agent using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Ralph Loop?
Ralph Loop is a specialized technical workflow designed to empower agents within the ecosystem of Openclaw Skills. It implements a structured Ralph playbook flow that separates software development into distinct Planning and Building phases. This ensures that agents do not jump into implementation without a verified architectural strategy.
By utilizing the exec and process tools with PTY support, this skill allows agents to manage interactive CLIs like Claude Code, OpenCode, and Codex without terminal hanging issues. It provides a robust framework for context persistence across iterations using local markdown files, making it a cornerstone for reliable autonomous development when using Openclaw Skills.
Ralph Loop Use Cases
- Orchestrating interactive AI coding agents that require TTY/terminal support to function.
- Implementing a rigorous planning-before-coding architecture to reduce hallucinations in complex projects.
- Automating multi-step software feature development using Openclaw Skills for task management.
- Monitoring long-running background coding tasks with automated completion detection and log polling.
How Ralph Loop Works
- The agent gathers requirements, defines the Job To Be Done (JTBD), and initializes the workdir.
- Focus topics are extracted into individual specification files within a specs/ directory to maintain modularity.
- A Planning Loop is initiated where the agent updates IMPLEMENTATION_PLAN.md to map out tasks without modifying source code.
- A Building Loop follows, where the agent implements specific tasks, runs backpressure test commands, and commits changes to version control.
- The agent utilizes the exec tool with pty: true to launch coding agents in the background while capturing session IDs.
- The process tool is used to poll logs and detect the STATUS: COMPLETE sentinel in the implementation plan to conclude the loop.
Ralph Loop Setup
To utilize Ralph Loop within your environment, ensure your agent has access to the exec and process tools. Configure your coding agents (like OpenCode or Claude Code) and prepare the project directory.
# The agent typically initiates the loop using a command pattern like:
opencode run --model <MODEL_NAME> "$(cat PROMPT.md)"
Ensure PROMPT.md and AGENTS.md are present in the root directory to provide the necessary context for the Openclaw Skills execution loop.
Ralph Loop Data Schema & Taxonomy
The Ralph Loop skill relies on a specific file-based metadata taxonomy to maintain state across agent iterations:
| File | Role | Description |
|---|---|---|
| PROMPT.md | Context | Primary instructions and references for the current iteration. |
| AGENTS.md | Knowledge | Contains test commands, build instructions, and operational learnings. |
| specs/*.md | Requirements | Technical specifications for individual focus topics or features. |
| IMPLEMENTATION_PLAN.md | State | The source of truth for task progress and the completion sentinel. |
Ralph Loop Advanced Features
- Native TTY/PTY support to prevent interactive CLI sessions from hanging during execution.
- Integrated backpressure mechanisms that force agents to run tests and update learnings in AGENTS.md.
- Support for multiple sandboxing environments including Docker and E2B for secure code execution.
- Flexible completion detection using regex to match various status sentinel formats in the implementation plan.
- Session persistence allowing agents to poll, log, and kill background processes across different turns.
SKILL.md
Loading
Related Openclaw Skills
Google Tasks Skill
A lightweight bash-based interface for managing Google Tasks lists, items, and due dates via the REST API.
AUTO-BUILDING AI Information Collector
An AI-driven automated information collection and management system for building custom content platforms.
AgentMail
An API-first email platform that provides AI agents with programmatic inboxes, real-time webhooks, and high-volume sending capabilities.
camoufox-tools
A streamlined CLI toolkit for camoufox and agent-browser designed to simplify anti-detection web automation and high-performance data scraping.
Creative Illustration Factory
A sophisticated AI-driven tool for generating professional illustrations, storyboards, and editorial art through structured command-line workflows.
Slack Controller
Automate Slack interactions including messaging, screen sharing, and status updates directly through browser or desktop app automation.
Featured*
