ralph

Orchestrate multi-phase task pipelines with isolated subagents via the Task tool.

20|5|Updated Nov 18, 2025
One-click install
npx skills add https://github.com/zby/llm-do --skill ralph-zby
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ralph
Source: https://github.com/zby/llm-do/tree/main/arscontexta/skills/arscontexta-ralph
Command: npx skills add https://github.com/zby/llm-do --skill ralph-zby

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill addresses the challenge of context contamination in sequential AI task processing by ensuring each task phase operates with a fresh, isolated context, preventing degradation of attention and ensuring learnings are captured.

Core Features & Use Cases

  • Isolated Subagent Spawning: Mandates the use of the Task tool to spawn a subagent for every single task, guaranteeing context isolation.
  • Phase-Specific Processing: Supports distinct processing phases (extract, create, enrich, reflect, reweave, verify) with tailored prompts and tool usage for each.
  • Serial and Parallel Modes: Offers both serial processing for up to N tasks and parallel processing for concurrent claim handling with up to 5 workers.
  • Dry Run and Filtering: Allows for dry runs to preview actions and filters tasks by batch or type.
  • Use Case: Imagine a content generation pipeline where new claims need to be extracted, notes created, enriched, connected to existing knowledge, and finally verified. Ralph ensures each step is handled independently, maintaining high quality and capturing specific learnings at each stage.

Quick Start

Process up to 5 tasks from the queue using the ralph skill.

Frequently Asked Questions about ralph

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I prevent context contamination in sequential AI task processing pipelines?

Pipeline automation for multi-phase task processing orchestrates tasks through extract, create, enrich, reflect, reweave, and verify stages. It uses isolated subagent spawning to ensure each phase operates with a fresh context, preventing attention degradation while capturing specific learnings.

How do I orchestrate subagent spawning for multi-phase queue processing?

Subagent spawning for multi-phase queue processing is orchestrated by using the Task tool to spawn a subagent for every single task. This guarantees context isolation across phases like extract, create, enrich, reflect, reweave, and verify, while capturing learnings via a handoff block.

Can I use parallel execution modes for concurrent queue processing?

Yes, concurrent queue processing supports parallel execution modes for handling tasks concurrently with up to 5 workers. It also offers serial processing for up to N tasks, allowing you to choose the best execution strategy for your workload.

Does queue processing support dry runs and task filtering by batch or type?

Queue processing supports dry runs to preview actions before execution and includes filtering capabilities to select tasks by batch or type. This allows you to safely test the pipeline and target specific subsets of your task queue.

What is the best way to capture learnings across isolated agent pipeline stages?

The best way to capture learnings across isolated agent pipeline stages is to employ a handoff block for robust context management. This mechanism ensures verifiable outcomes are passed between phases, maintaining quality even when subagent context is isolated.