What is the difference between --fast and --one-shot?

--fast runs a reduced set of essential stages with lower quality thresholds, while --one-shot performs a minimal refinement without any quality gate checks.

What does the 'LLM-as-Judge' feature do?

It uses an AI agent to evaluate the quality of each planning phase against a specific threshold, ensuring the plan meets your standards before moving to the next step.

How do I resume a planning session that was interrupted?

You can use the --continue flag, which optionally takes a stage name or automatically detects where the workflow left off based on the task file content.

Can I use this skill on tasks I have already modified manually?

Yes, using the --refine flag allows the skill to detect your manual edits via git and only update the sections of the plan affected by those changes.

What is the primary purpose of the sdd:plan skill?

It automates the process of turning a rough task draft into a detailed, verified, and parallelized implementation plan ready for coding.

Task Refinement Orchestrator

Name: Task Refinement Orchestrator
Author: NeoLabHQ

byNeoLabHQ

•

542

•

Productivity & Workflow

Transforms draft task specifications into implementation-ready plans through multi-agent analysis and quality-gated verification.

The Task Refinement Orchestrator is a sophisticated workflow manager designed to bridge the gap between initial task drafts and technical execution. It employs a multi-agent approach to perform parallel codebase analysis, research, and business requirement refinement, synthesizing these findings into a cohesive architectural overview. By decomposing tasks into logical steps and optimizing them for parallel execution, it provides a clear roadmap for development. With integrated LLM-as-Judge quality gates and smart 'Refine Mode' that tracks git changes, this skill ensures that your implementation plans are accurate, thorough, and ready for immediate action within Claude Code.

Key Features

01Optimization of implementation steps for maximum parallel execution efficiency.

02Automated architectural synthesis and implementation step decomposition.

03Integrated LLM-as-Judge verification with configurable quality thresholds (1.0 - 5.0).

04542 GitHub stars

05Incremental refinement mode that detects git changes to re-run only affected planning stages.

06Parallel multi-agent analysis covering research, codebase impact, and business logic.

Use Cases

01Converting high-level feature requests into detailed, step-by-step technical execution plans.

02Updating existing implementation specs after codebase changes or architectural shifts using --refine.

03Enforcing high-quality planning standards for critical system updates through automated verification gates.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add neolabhq/context-engineering-kit plan

For use in Claude.ai and ChatGPT

Download Skill